1 2007-09-24 Cluster File Systems, Inc. <info@clusterfs.com>
5 2.6.9-55.0.2.EL (RHEL 4),
6 2.6.16.46-0.14 (SLES 10),
7 2.6.18-8.1.8.el5 (RHEL 5),
8 2.6.18.8 vanilla (kernel.org)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.16 - 2.6.22 vanilla (kernel.org)
12 * Due to recently discovered recovery problems, we do not recommend
13 using patchless RHEL 4 clients with this or any earlier release.
14 * Recommended e2fsprogs version: 1.40.2-cfs1
15 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
19 Description: obd_health_check_timeout too short
20 Details : set obd_health_check_timeout as 1.5x of obd_timeout
24 Description: sanity.sh failed test 103
25 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
30 Description: llapi_file_create() does not allow some changes
31 Details : add llapi_file_open() that allows specifying the mode and
32 open flags, and also returns an open file handle.
36 Description: df doesn't work properly if diskfs blocksize != 4K
37 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
41 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
42 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
46 Description: merge and cleanup kernel patches.
47 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
51 Description: Add journal checksum support.(Kernel part)
52 Details : The journal checksum feature adds two new flags i.e
53 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
54 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
55 indicates that the commit block contains the checksum for
56 the blocks described by the descriptor blocks. Now commit
57 record can be sent to disk without waiting for descriptor
58 blocks to be written to disk. This behavior is controlled
59 using JBD2_FEATURE_ASYNC_COMMIT flag.
63 Description: RedHat Update kernel for RHEL5
64 Details : Kernel config file for RHEL5.
68 Description: OSS needs mutliple precreate threads
69 Details : Add ability to start more than one create thread per OSS.
73 Description: Sanity.sh test_65k failed: setstripe should have succeeded
74 Details : execute lfs setstripe on client
78 Description: mds_obd_create error creating tmp object
79 Details : When the user sets quota on root, llog will be affected and can't
80 create files and write files.
83 Frequency : Always on ia64 patchless client, and possibly others.
85 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
86 Details : This allows the patchless client to be loaded on architectures
91 Description: move common funcs to test-frame.sh
92 Details : cleanup those test-scripts and move some common funcs to
98 Description: disorder of journal start and llog_add cause deadlock.
99 Details : in llog_origin_connect, journal start should happen before
100 llog_add keep the same order as other functions to avoid
105 Description: RedHat Update kernel for RHEL5
106 Details : Modify the kernel config file more closer RHEL5.
110 Description: Build failure against Centos5 (RHEL5)
111 Details : Define PAGE_SIZE when it isn't present.
115 Description: test_7 on sanity-quota.sh can't be used on separate machines
116 Details : add facet to handle it
118 --------------------------------------------------------------------------------
120 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
122 * Support for kernels:
123 2.6.5-7.286 (SLES 9),
124 2.6.9-55.0.2.EL (RHEL 4),
125 2.6.16.46-0.14 (SLES 10),
126 2.6.18-8.1.8.el5 (RHEL 5),
127 2.6.18.8 vanilla (kernel.org)
128 * Client support for unpatched kernels:
129 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
130 2.6.16 - 2.6.22 vanilla (kernel.org)
131 * Due to recently discovered recovery problems, we do not recommend
132 using patchless RHEL 4 clients with this or any earlier release.
133 * Recommended e2fsprogs version: 1.39.cfs8
134 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
138 Description: block reactivating mgc import until all deactivates complete
139 Details : Fix race when failing back MDT/MGS to itself (testing)
141 Severity : enhancement
143 Description: lfs setstripe enhancement
144 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
147 Frequency : mds/oss recovery
149 Description: llog ctxt is refrenced after it has been freed.
150 Details : llog ctxt refcount was added to avoide the race between ctxt free
151 and llog recovery process. Each llog user must hold ctxt refcount
152 before it access the llog. And the llog ctxt can only be freed
153 when its refcount is zero.
155 Severity : enhancement
157 Description: randomly memory allocation failure util
158 Details : Make lustre randomly failed allocating memory for testing purpose.
161 Frequency : only for SLES10
163 Description: Update kernel patch for SLES10 SP1
164 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
166 Severity : enhancement
168 Description: omit set fsid for export NFS
169 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
170 as FS_REQUIRES_DEV for avoid problems with generate fsid.
174 Description: Add sync option to mount_lustre.c
175 Details : Client loses data written to lustre after a network interruption.
179 Description: Error message improvement.
180 Details : Merging of two LCONSOLE_ERROR_MSG into one.
184 Description: don't use GFP_* in generic Lustre code.
185 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
189 Description: obdclass is limited by single OBD_ALLOC(idarray)
190 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
194 Description: updated patchess for new RHEL4 kernel
195 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
199 Description: warnings with build patchless client with vanila 2.6.19 and up
200 Details : change old ctl_table style and replace ctl_table/ctl_table_header
201 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
205 Description: O_DIRECT bypasses client statistics.
206 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
207 read_ahead_stats were updated. Copied stats section from
208 osc_send_oap_rpc() into async_internal().
212 Frequency : only for Cray XT3
213 Description: peer credits not enough on many OST per OSS systems.
214 Details : Use new lnet way to add credits as we need those for pings and ASTs
218 Frequency : only with liblustre
219 Description: Liblustre is not releasing flock locks on file close.
220 Details : Release flock locks on file close.
224 Frequency : only for RHEL4
225 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
226 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
227 Update target file and kernel config.
231 Description: lustre support for RHEL5
232 Details : Add support for RHEL5.
237 Description: ASSERTION(export != NULL) failed in target_handle_connect
238 Details : Assetion hit is result of rare race between disconnect and connet
239 to same nid. target_handle_connect found old connect cockie and
240 tried to reconnect, but can't find export for this cockie.
245 Description: umount blocks forever on error
246 Details : In result of wrong using obd_no_recov and obd_force flags client
247 can hand if cancel or some other requests is lost.
251 Frequency : Only for SLES
252 Description: sanity_quota fail test_1
253 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
254 file, which makes TSTID[2] inunique.
258 Frequency : Only for SLES9
259 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
260 Details : Update target/ChangeLog/which_patch .
264 Frequency : Only for test_10 in sanity-quota.sh
265 Description: a bug in quota test code
266 Details : checking a test flag in a wrong place
270 Frequency : Only for test_13 in sanity-quota.sh
271 Description: a bug in quota test script
272 Details : shouldn't check overfull allocated quota
276 Description: jbd statistics
277 Details : Port older jbd statistics patch for sles10
282 Description: lvbo_init failed for resource with missing objects.
283 Details : Fix returning error if we do stat for file with missing/corrupted
284 objects and i_size set to all sum of size of all avaible objects.
285 if we to truncate/write to missing object - it is recreated.
290 Description: Oops in read and write path when failing to allocate lock.
291 Details : Check if lock allocation failed and return error back.
295 Frequency : When flocks are used.
296 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
297 Details : Flock locks might destroy just granted lock if it could be merged
298 with another existing flock, this is done in completion handler,
299 so teach ldlm_cli_enquque_fini that this is a valid case for
305 Description: reply_lock_interpret crash due to race with it and lock cancel.
306 Details : Do not replay locks that are being cancelled. Do not reference
307 locks by their address during replay, just by their handle.
311 Description: lstripe command fails for valid OST index
312 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
313 instead of lov->desc.ld_active_tgt_count.
317 Description: sanity.sh failed test 103
318 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
323 Description: Build failure against Centos5 (RHEL5)
324 Details : Use getpagesize() instead of PAGE_SIZE.
326 --------------------------------------------------------------------------------
328 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
330 * Support for kernels:
331 2.6.5-7.283 (SLES 9),
332 2.6.9-55.EL (RHEL 4),
333 2.6.16.46-0.14 (SLES 10),
334 2.6.18.8 vanilla (kernel.org)
335 * Client support for unpatched kernels:
336 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
337 2.6.16 - 2.6.22 vanilla (kernel.org)
338 * Due to recently discovered recovery problems, we do not recommend
339 using patchless RHEL 4 clients with this or any earlier release.
340 * Recommended e2fsprogs version: 1.39.cfs8
341 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
342 * Starting with this release, the ldiskfs backing filesystem required
343 by Lustre is now in its own package, lustre-ldiskfs. This package
344 should be installed. It is versioned separately from Lustre and
345 may be released separately in future.
350 Description: server LBUG when shutting down
351 Details : Block umount forever until the mount refcount is zero rather
352 than giving up after an arbitrary timeout.
354 Severity : enhancement
356 Description: add optional extra BUILD_VERSION info
357 Details : add a new environment variable (namely LUSTRE_VERS) which allows
358 to override the lustre version.
361 Frequency : 2.6.18 servers only
363 Description: ll_kern_mount() doesn't release the module reference
364 Details : The ldiskfs module reference count never drops down to 0
365 because ll_kern_mount() doesn't release the module reference.
370 Description: server LBUG when using old ost_num_threads parameter
371 Details : Accept the old ost_num_threads parameter but warn that it
372 is deprecated, and fix an off-by-one error that caused an LBUG.
377 Description: Transient SCSI error results in persistent IO issue
378 Details : iobuf->dr_error is not reinitialized to 0 between two
382 Frequency : sometimes when underlying device returns I/O errors
384 Description: OSTs not going read-only during write failures
385 Details : OSTs are not remounted read-only when the journal commit threads
386 get I/O errors because fsfilt_ext3 calls journal_start/stop()
387 instead of the ext3 wrappers.
391 Description: poor connect scaling with increasing client count
392 Details : Don't run filter_grant_sanity_check for more than 100 exports
393 to improve scaling for large numbers of clients.
396 Frequency : SLES10 only
398 Description: sanity-quota.sh quotacheck failed: rc = -22
399 Details : Quotas cannot be enabled on SLES10.
402 Frequency : liblustre clients only
404 Description: getdirentries does not give error when run on compute nodes
405 Details : getdirentries does not fail when the size specified as an argument
406 is too small to contain at least one entry
408 Severity : enhancement
410 Description: Add LNET router traceability for debug purposes
411 Details : If a checksum failure occurs with a router as part of the
412 IO path, the NID of the last router that forwarded the bulk data
413 is printed so it can be identified.
418 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
419 Details : Due to a race condition, liblustre clients were occasionally
422 Severity : enhancement
424 Description: lfs setstripe use optional parameters instead of postional
427 Severity : enhancement
429 Description: Nanosecond timestamp support for ldiskfs
430 Details : The on-disk ldiskfs filesystem has added support for nanosecond
431 resolution timestamps. There is not yet support for this at
432 the Lustre filesystem level.
435 Frequency : during server recovery
437 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
438 Details : request with rq_no_resend flag not awake l_wait_event if they get a
442 Frequency : nfs export on patchless client
444 Description: connectathon hang when test nfs export over patchless client
445 Details : Disconnected dentry cannot be found with lookup, so we do not need
446 to unhash it or make it invalid
449 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
450 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
451 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
456 Description: open req refcounting wrong on reconnect
457 Details : If reconnect happened between getting open reply from server and
458 call to mdc_set_replay_data in ll_file_open, we will schedule
459 replay for unreferenced request that we are about to free.
460 Subsequent close will crash in variety of ways.
461 Check that request is still eligible for replay in
462 mdc_set_replay_data().
467 Description: disable writes to filesystem when reading health_check file
468 Details : the default for reading the health_check proc file has changed
469 to NOT do a journal transaction and write to disk, because this
470 can cause reads of the /proc file to hang and block HA state
471 checking on a healthy but otherwise heavily loaded system. It
472 is possible to return to the previous behaviour during configure
473 with --enable-health-write.
475 Severity : enhancement
477 Description: 64-bit inode version
478 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
479 to the inode. This will be required for version-based recovery.
484 Description: MDS fails to start if a duplicate client export is detected
485 Details : in some rare cases it was possible for a client to connect to
486 an MDS multiple times. Upon recovery the MDS would detect this
487 and fail during startup. Handle this more gracefully.
489 Severity : enhancement
491 Description: Add -o localflock option to simulate old noflock behaviour.
492 Details : This will achieve local-only flock/fcntl locks coherentness.
497 Description: log_commit_thread vs filter_destroy race leads to crash
498 Details : Take import reference before releasing llog record semaphore
503 Description: Wrong request locking in request set processing
504 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
505 imp_delayed_list, in this place should be used imp_lock.
508 Frequency : when reconnecting
510 Description: Grant leak when OSC reconnect to OST
511 Details : When osc reconnect ost, OST(filter) should check whether it
512 should grant more space to client by comparing fed_grant and
513 cl_avail_grant, and return the granted space to client instead
514 of "new granted" space, because client will call osc_init_grant
515 to update the client grant space info.
518 Frequency : when client reconnects to OST
520 Description: Grant leak when OSC does a resend and replays bulk write
521 Details : When osc reconnect to OST, OST(filter) should clear grant info of
522 bulk write request, because the grant info while be sync between
523 OSC and OST when reconnect, and we should ignore the grant info
524 these of resend/replay write req.
529 Description: Grant space more than avaiable space sometimes.
530 Details : When then OST is about to be full, if two bulk writing from
531 different clients came to OST. Accord the avaliable space of the
532 OST, the first req should be permitted, and the second one
533 should be denied by ENOSPC. But if the seconde arrived before
534 the first one is commited. The OST might wrongly permit second
535 writing, which will cause grant space > avaiable space.
538 Frequency : when client is evicted
540 Description: Grant might be wrongly erased when osc is evicted by OST
541 Details : when the import is evicted by server, it will fork another
542 thread ptlrpc_invalidate_import_thread to invalidate the
543 import, where the grant will be set to 0. While the original
544 thread will update the grant it got when connecting. So if
545 the former happened latter, the grant will be wrongly errased
546 because of this race.
551 Description: Checking Stale with correct fid
552 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
553 to check whether it is stale, because sometimes, we want the
554 enqueue happened anyway, and op_data.fid2 will not be initialized.
556 Severity : enhancement
558 Description: update patchless client
559 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
562 Frequency : only with 2.4 kernel
564 Description: random memory corruption
565 Details : size of struct ll_inode_info is to big for union inode.u and this
566 can be cause of random memory corruption.
571 Description: Memory leak in recovery
572 Details : Lov_mds_md was not free in an error handler in mds_create_object.
573 It should also check obd_fail before fsfilt_start, otherwise if
574 fsfilt_start return -EROFS,(failover mds during mds recovery).
575 then the req will return with repmsg->transno = 0 and rc = EROFS.
576 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
577 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
578 be freed no matter whether fsfilt_commit success or not.
581 Frequency : only with huge numbers of clients
583 Description: Prevents from taking the superblock lock in llap_from_page for
585 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
586 for llap_from_page call in ll_removepage() prevents from taking
587 the superblock lock for a soon died page.
592 Description: Not check open intent error before release open handle
593 Details : in some rare cases, the open intent error is not checked before
594 release open handle, which may cause
595 ASSERTION(open_req->rq_transno != 0), because it tries to release
596 the failed open handle.
601 Description: Set cat log bitmap only after create log success.
602 Details : in some rare cases, the cat log bitmap is set too early. and it
603 should be set only after create log success.
607 Description: Accessing a block bevice can re-enable I/O when Lustre is
608 tearing down a device.
609 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
614 Description: service threads may hog cpus when there are a lot of requests
615 Details : Insert cond_resched to give other threads a chance to use some CPU
620 Description: the cat log was not initialized in recovery
621 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
622 unlink log on mds will not be initialized until mds post
623 recovery. And also in mds post recovery, the unlink log will
624 initialization will be done asynchronausly, so there will be race
625 between add unlink log and unlink log initialization.
629 Description: brw_stats were being printed incorrectly
630 Details : brw_stats were being printed as log2 but all of them were not
631 recorded as log2. Also remove some code duplication arising from
632 filter_tally_{read,write}.
636 Frequency : rare, only in recovery.
637 Description: ASSERTION(req->rq_type != LI_POISON) failed
638 Details : imp_lock should be held while iterating over imp_sending_list for
639 prevent destroy request after get timeout in ptlrpc_queue_wait.
643 Description: replay-single.sh test 52 fails
644 Details : A lock's skiplist need to be cleanup when it being unlinked
645 from its resource list.
650 Description: Short directio read returns full requested size rather than
652 Details : Direct I/O operations should return actual amount of bytes
653 transferred rather than requested size.
655 Severity : enhancement
657 Description: metadata RPC reduction (e.g. for rm performance)
658 Details : decrease the amount of synchronous RPC between clients and servers
659 by canceling conflicing lock before the operation on the client
660 and packing thier handles into the main operation RPC to server.
662 Severity : enhancement
664 Description: add #ifdef HAVE_KERNEL_CONFIG_H
665 Details : kernels from 2.6.19 not need include linux/config.h, but add
666 include linux/autoconf.h in commpiler command line.
668 Severity : enhancement
670 Description: patchless client support for 2.6.22 kernel
671 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
672 removed. In this case we need drop using os depended interface to
673 kmem_cache and use cfs_mem_cache API.
678 Description: fix mal-formatted messages
679 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
683 Frequency : always in liblustre
684 Description: wrong IS_ERR implementation in liblustre.h
685 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
690 Description: Correct condition for output debug message.
691 Details : inode i_nlink equal zero is not enough for output message about
692 disk corruption, i_ctime and i_mode should be also checked.
696 Frequency : always in patchless client
697 Description: add configure check for truncate_complete_page
698 Details : improve checks for exported symbols. This allow run check without
699 sources, but with Module.symvers shipped with kernel distribution.
700 add check for truncate_complete_page used by patchless client.
704 Description: sanity.sh test_77h fails with "test_77h file compare failed"
705 Details : test_77h uses a file which was messed by other test case.
709 Description: Not Check whether lov_tgts is NULL in some lov functions
710 Details : Checking whether lov_tgts is NULL in some functions.
714 Description: replace obdo_alloc() with OBDO_ALLOC macro
715 Details : nothing special is done in obdo_alloc() function, and for
716 debugging purpose, it needs to be replaced with macros.
720 Description: bad return value and errno from fcntl call
721 Details : In liblustre API, errno should be a negative value if error
726 Description: ptlrpc_check_set() LBUG
727 Details : In case of positive reply from server and failed client bulk
728 callback after bulk transfer shouldn't LBUG, but process this
729 request as erroneous.
731 Severity : enhancement
733 Description: VFS operations stats tool.
734 Details : Tool which collects stats by tracking value written in pid,
735 ppid, gid and uses llstat to generate output to plot graph using
737 Updated lustre/utils/Makefile.am
738 Added lustre/utils/ltrack_stats.c
740 Severity : enhancement
742 Description: 2.6.18 server support (lustre 1.6.1)
743 Details : Support for 2.6.18 kernels on the server side.
748 Description: ASSERTION(imp->imp_conn_current) failed
749 Details : an assertion failure is hit if a client node boots and attempts to
750 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
753 Frequency : only for i686
755 Description: 1.4.11 RC1 build fails for RHEL 4, i686
756 Details : Fixed config variable for build.
761 Description: Updated patchess for new RHEL4 kernel
762 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
763 Updated patch jbd-stats-2.6.9.patch
764 Updated patch qsnet-rhel4-2.6.patch
765 Updated patch quota-deadlock-on-pagelock-core.patch
766 Updated patch vfs_intent-2.6-rhel4.patch
767 Updated patch vfs_races-2.6-rhel4.patch
768 Updated series file 2.6-rhel4-titech.series
769 Updated series file 2.6-rhel4.series
770 Updated kernel config files
775 Description: lquota slave complains LBUG when reconnecting with mds
777 Details : quota slave depends qctxt->lqc_import to send its quota request.
778 This pointer will be invalid if mds did failover or broke its
779 connect to osts, which leads to LBUG.
782 Frequency : when qunit size is too small(less than 20M)
784 Description: write is stopped by improper -EDQUOT
785 Details : If the master is busy and qunit size is small enough(let's say 1M),
786 the slave can not get quota from master on time, which will lead
787 slave to trigger a -EQUOTA to client.
792 Description: Deadlock during metadata tests
793 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
794 called with the per-dentry lock held.
797 Frequency : SLES9 only
799 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
800 Details : set CONFIG_NR_CPUS to 128 instead of 8.
802 Severity : enhancement
804 Description: remove fs_prep_san_write operation and related patches
805 Details : remove the ext3-san-jdike patches which are no longer useful.
810 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
811 Details : a disk corruption can cause the mballoc code to assert on a
812 double free or other extent corruptions. Handle these with
813 ext3_error() instead of assertions.
816 Frequency : only with mballoc3 code and deep extent trees
818 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
819 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
820 can cause the filesystem to be remounted read-only.
825 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
826 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
827 few external users of this ioctl number is lfsck's call to
828 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
829 Add the old ioctl number to the handler so both old and new
834 Description: parallel lock callbacks
835 Details : Instead of sending blocking and completion callbacks as separated
836 requests, adding them to a set and sending in parallel.
840 Description: Disable most debugging by default
841 Details : To improve performance, disable most logging (for debug purposes)
842 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
843 default, and HA includes fewer messages.
847 Frequency : only run on patchless client.
848 Description: use do_facet on sanity.sh for test handling recoverables errors
849 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
853 Description: hash tables for lists of nids, connections and uuids
854 Details : Hash tables noticeably help when a lot of clients connect to a
855 server, to faster identify duplicate connections or reconnects,
856 also to faster find export to evict in manual eviction case.
860 Description: Sometimes, when the server evict a client, and the client will
861 not be evicted as soon as possible.
862 Details : In enqueue req, the error was returned by intent, instead of
863 rq_status which make ptlrpc layer not detect this error, and
864 does not evict the client. So enqueue error should be returned
868 Frequency : only at startup
870 Description: Delay client connections to MDT until fist MDT->OST connect
871 Details : If a client tried to create a new file before the MDT had
872 connected to any OSTs, the create would return EIO. Now
873 the client will simply block until the MDT connects to the
874 first OST and the create can succeed.
877 Frequency : only for SLES9
879 Description: Routinely utilize latest Quadrics drivers in CFS releases
880 Details : Update patch qsnet-suse-2.6.patch.
883 Frequency : only for sles10
885 Description: Update patches for SLES 10 SP1 kernel.
886 Details : Update the patch vfs_intent-2.6-sles10.patch.
891 Description: Routinely utilize latest Quadrics drivers in CFS releases
892 Details : Update patch qsnet-rhel4-2.6.patch.
895 Frequency : at statup only
897 Description: mds_lov_synchronize race leads to various problems
898 Details : simultaneous MDT->OST connections at startup can cause the
899 sync to abort, leaving the OSC in a bad state.
903 Description: Using wrong pointer in osc_brw_prep_request
904 Details : Access to array[-1] can produce panic if kernel compiled with
905 CONFIG_PAGE_ALLOC enabled
907 Severity : enhancement
909 Description: Async OSC create to avoid the blocking unnecessarily.
910 Details : If a OST has no remain object, system will block on the creating
911 when need to create a new object on this OST. Now, ways use
912 pre-created objects when available, instead of blocking on an
913 empty osc while others are not empty. If we must block, we block
914 for the shortest possible period of time.
918 Frequency : only in recovery
919 Description: Mark OST as early accessible if his start SYNC.
920 Details : osc_precreate return flag early accessible if oscc marked as
921 OSCC_FLAG_SYNC_IN_PROGRESS.
926 Description: Sometimes precreate code can triger create object on wrong ost
927 Details : Wrong protected or not not restored variables aftre precreate loop
928 can produce creation object on wrong ost.
931 Frequency : oss recovery
933 Description: llog_commit_thread cleanup should sync with llog_commit_thread
935 Details : llog_commit_thread_count should be synced between llog_commit
936 start and cleanup, so new llog_commit thread should not be started
937 when llog_commit threads being stopped to avoid accessing some
940 Severity : enhancement
942 Description: Add printing inode info into message about error in writepage.
944 --------------------------------------------------------------------------------
946 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
951 Frequency : on some architectures
953 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
954 Details : Uninitialized flags sometimes cause configuration commands to
958 Frequency : patchless clients only
960 Description: missing __iget() symbol export
961 Details : The __iget() symbol export is missing. To avoid the need for
962 this on patchless clients the deathrow inode reaper is turned
963 off, and we depend on the VM to clean up old inodes. This
964 dependency was during via the fix for bug 12181.
969 Description: sanity.sh fail: test_52b
970 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
971 incorrect inode's flags to client.
973 --------------------------------------------------------------------------------
975 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
977 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
978 INTEROPERATE with older versions automatically. In many cases a
979 special upgrade step is needed. Please read the
980 user documentation before upgrading any part of a 1.4.x system.
981 * WARNING: Lustre configuration and startup changes are required with
982 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
984 * Support for kernels:
985 2.4.21-47.0.1.EL (RHEL 3),
986 2.6.5-7.283 (SLES 9),
987 2.6.9-42.0.10.EL (RHEL 4),
988 2.6.12.6 vanilla (kernel.org),
989 2.6.16.27-0.9 (SLES10)
990 * Client support for unpatched kernels:
991 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
992 2.6.16 - 2.6.19 vanilla (kernel.org),
993 2.6.9-42.0.8EL (RHEL 4)
994 * Recommended e2fsprogs version: 1.39.cfs6
995 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
998 Severity : enhancement
1000 Description: Async OSC create to avoid the blocking unnecessarily.
1001 Details : If a OST has no remain object, system will block on the creating
1002 when need to create a new object on this OST. Now, ways use
1003 pre-created objects when available, instead of blocking on an
1004 empty osc while others are not empty. If we must block, we block
1005 for the shortest possible period of time.
1007 Severity : enhancement
1009 Description: MountConf
1010 Details : Lustre configuration is now managed via mkfs and mount
1011 commands instead of lmc and lconf. New obd types (MGS, MGC)
1012 are added for dynamic configuration management. See
1013 https://mail.clusterfs.com/wikis/lustre/MountConf for
1016 Severity : enhancement
1018 Description: dynamic OST addition
1019 Details : OSTs can now be added to a live filesystem
1021 Severity : enhancement
1023 Description: startup order invariance
1024 Details : MDTs and OSTs can be started in any order. Clients only
1025 require the MDT to complete startup.
1027 Severity : enhancement
1029 Description: parallel, asynchronous orphan cleanup
1030 Details : orphan cleanup is now performed in separate threads for each
1031 OST, allowing parallel non-blocking operation.
1033 Severity : enhancement
1035 Description: optimized stripe assignment
1036 Details : stripe assignments are now made based on ost space available,
1037 ost previous usage, and OSS previous usage, in order to try
1038 to optimize storage space and networking resources.
1040 Severity : enhancement
1042 Description: Permanently set tunables
1043 Details : All writable /proc/fs/lustre tunables can now be permanently
1044 set on a per-server basis, at mkfs time or on a live
1047 Severity : enhancement
1049 Description: Lustre message v2
1050 Details : Add lustre message format v2.
1052 Severity : enhancement
1054 Description: client OST exclusion list
1055 Details : Clients can be started with a list of OSTs that should be
1056 declared "inactive" for known non-responsive OSTs.
1060 Description: ENOENT returned for valid filehandle during dbench.
1061 Details : Check if a directory has children when invalidating dentries
1062 associated with an inode during lock cancellation. This fixes
1063 an incorrect ENOENT sometimes seen for valid filehandles during
1064 testing with dbench.
1067 Frequency : SFS test only (otherwise harmless)
1069 Description: SPEC SFS validation failure on NFS v2 over lustre.
1070 Details : Changes the blocksize for regular files to be 2x RPC size,
1071 and not depend on stripe size.
1073 Severity : enhancement
1075 Description: fine-grained SMP locking inside DLM
1076 Details : Improve DLM performance on SMP systems by removing the single
1077 per-namespace lock and replace it with per-resource locks.
1079 Severity : enhancement
1081 Description: don't hold multiple extent locks at one time
1082 Details : To avoid client eviction during large writes, locks are not
1083 held on multiple stripes at one time or for very large writes.
1084 Otherwise, clients can block waiting for a lock on a failed OST
1085 while holding locks on other OSTs and be evicted.
1087 Severity : enhancement
1089 Description: Multiple MD RPCs in flight.
1090 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1091 To avoid overly-overloading MDT, introduce a limit on number of
1092 MDT RPCs in flight for a single client and add /proc controls
1093 to adjust this limit.
1095 Severity : enhancement
1097 Description: client read/write statistics
1098 Details : Add client read/write call usage stats for performance
1099 analysis of user processes.
1100 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1101 file access. extents_stats shows chunk size distribution.
1102 extents_stats_per_process show chunk size distribution per
1105 Severity : enhancement
1107 Description: per-client statistics on server
1108 Details : Add ldlm and operations statistics for each client in
1109 /proc/fs/lustre/mds|obdfilter/*/exports/
1111 Severity : enhancement
1113 Description: improved MDT statistics
1114 Details : Add detailed MDT operations statistics in
1115 /proc/fs/lustre/mds/*/stats
1117 Severity : enhancement
1119 Description: VFS operations stats
1120 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1121 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1126 Description: Fix client-side osc byte counters
1127 Details : The osc read/write byte counters in
1128 /proc/fs/lustre/osc/*/stats are now working
1131 Frequency : always as root on SLES
1133 Description: Failure of copying files with lustre special EAs.
1134 Details : Client side always return success for setxattr call for lustre
1135 special xattr (currently only "trusted.lov").
1140 Description: Refcount LNET uuids
1141 Details : The global LNET uuid list grew linearly with every startup;
1142 refcount repeated list entries instead of always adding to
1145 Severity : enhancement
1147 Description: Dynamic service threads
1148 Details : Within a small range, start extra service threads
1149 automatically when the request queue builds up.
1152 Frequency : mixed-endian client/server environments
1154 Description: mixed-endian crashes
1155 Details : The new msg_v2 system had some failures in mixed-endian
1158 Severity : enhancement
1160 Description: Easy OST removal
1161 Details : OSTs can be permanently deactivated with e.g. 'lctl
1162 conf_param lustre-OST0001.osc.active=0'
1164 Severity : enhancement
1166 Description: MGS proc entries
1167 Details : Added basic proc entries for the MGS showing what filesystems
1170 Severity : enhancement
1172 Description: provide MGS failover
1173 Details : Added config lock reacquisition after MGS server failover.
1175 Severity : enhancement
1177 Description: add Linux 2.4 support
1178 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1182 Description: a large application tries to do I/O to the same resource and dies
1183 in the middle of it.
1184 Details : Check the req->rq_arrival time after the call to
1185 ost_brw_lock_get(), but before we do anything about
1186 processing it & sending the BULK transfer request. This
1187 should help move old stale pending locks off the queue as
1188 quickly as obd_timeout.
1191 Frequency : when an incorrect nid is specified during startup
1193 Description: ptlrpc connect to non-existant node causes kernel crash
1194 Details : LNET can't be re-entered from an event callback, which
1195 happened when we expire a message after the export has been
1196 cleaned up. Instead, hand the zombie cleanup off to another
1199 Severity : enhancement
1201 Description: plain/inodebits lock performance improvement
1202 Details : Grouping plain/inodebits in granted list by their request modes
1203 and bits policy, thus improving the performance of search through
1207 Frequency : only if OST filesystem is corrupted
1209 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1210 Details : for a short time RPCs with bulk IO are in the replay list,
1211 but replay of bulk IOs is unimplemented. If the OST filesystem
1212 is corrupted due to disk cache incoherency and then replay is
1213 started it is possible to trip an assertion. Avoid putting
1214 committed RPCs into the replay list at all to avoid this issue.
1217 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1219 Description: System hang on startup
1220 Details : This bug allowed the liblustre (e.g. catamount) client to
1221 return to the app before handling all startup RPCs. This
1222 could leave the node unresponsive to lustre network traffic
1223 and manifested as a server ptllnd timeout.
1225 Severity : enhancement
1227 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1228 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1229 boolean to control whether to print peer debug info when a
1230 client's RPC times out.
1233 Frequency : only for kernels with patches from Lustre below 1.4.3
1235 Description: Remove old rdonly API
1236 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1239 Frequency : only for devices with external journals
1241 Description: Set external device read-only also
1242 Details : During a commanded failover stop, we set the disk device
1243 read-only while the server shuts down. We now also set any
1244 external journal device read-only at the same time.
1247 Frequency : when upgrading from 1.4 while trying to change parameters
1249 Description: The wrong (new) MDC name was used when setting parameters for
1250 upgraded MDT's. Also allows changing of OSC (and MDC)
1251 parameters if --writeconf is specified at tunefs upgrade time.
1254 Frequency : when setting specific ost indicies
1256 Description: QOS code breaks on skipped indicies
1257 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1258 created with --index need not be sequential.
1260 Severity : enhancement
1262 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1263 Details : The uninit_groups feature works in conjunction with the kernel
1264 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1265 up the pass1 processing of e2fsck. This is a read-only feature
1266 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1267 mount filesystems that have had this feature enabled.
1269 Severity : enhancement
1271 Description: Improve multi-block allocation algorithm to avoid fragmentation
1272 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1273 allocation locality and avoid filesystem fragmentation.
1275 ------------------------------------------------------------------------------
1277 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1279 * Support for kernels:
1280 2.4.21-47.0.1.EL (RHEL 3)
1281 2.6.5-7.283 (SLES 9)
1282 2.6.9-42.0.10.EL (RHEL 4)
1283 2.6.12.6 vanilla (kernel.org)
1284 2.6.16.27-0.9 (SLES 10)
1285 * Recommended e2fsprogs version: 1.39.cfs5
1287 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1291 Frequency : occasional, depends on client load and configuration
1292 Bugzilla : 12181, 12203
1293 Description: data loss for recently-modified files
1295 Details : In some cases it is possible that recently written or created
1296 files may not be written to disk in a timely manner (this should
1297 normally be within 30s unless client IO load is very high).
1298 The problem appears as zero-length files or files that are a
1299 multiple of 1MB in size after a client crash or client eviction
1300 that are missing data at the end of the file.
1302 This problem is more likely to be hit on clients where files are
1303 repeatedly created and unlinked in the same directory, clients
1304 have a large amount of RAM, have many CPUs, the filesystem has
1305 many OSTs, the clients are rebooted frequently, and/or the files
1306 are not accessed by other nodes after being written.
1308 The presence of the problem can be detected by looking at
1309 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1310 smaller than the second (nr_unused) then dirty files will not
1311 be flushed automatically to disk. "sync; sleep 10" should be
1312 run several times on the node before unmounting it to update
1313 Lustre (this is also safe to run on nodes without this problem).
1315 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1316 that can cause this same problem, so customers using that kernel
1317 also need to update the kernel in addition to Lustre. In order
1318 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1320 It is normal that files written just before a client crash (less
1321 than 30s) may not yet have been flushed to disk, even for local
1325 Frequency : frequent on thin XT3 nodes
1327 Description: UUID collision on thin XT3 Linux nodes
1328 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1329 randomly, since we relied on an insufficiently-seeded PRNG.
1334 Description: OSS hangs after "All ost request buffers busy"
1335 Details : A deadlock between quota and journal operations caused OSS
1336 hangs after printing "All ost request buffers busy."
1339 Frequency : always on liblustre builds
1341 Description: Cleanup compiler warnings on liblustre
1344 Frequency : always on liblustre builds on XT3
1346 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1350 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1351 > 32000 subdirectories
1352 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1353 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1354 subdirectory count crosses 32000. This will aid e2fsck to
1355 correctly handle more than 32000 subdirectories.
1358 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1360 Description: System hang on startup
1361 Details : This bug allowed the liblustre (e.g. catamount) client to
1362 return to the app before handling all startup RPCs. This
1363 could leave the node unresponsive to lustre network traffic
1364 and manifested as a server ptllnd timeout.
1366 Severity : enhancement
1368 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1369 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1370 boolean to control whether to print peer debug info when a
1371 client's RPC times out.
1376 Description: make O_SYNC working on 2.6 kernels
1377 Details : 2.6 kernels use different method for mark pages for write,
1378 so need add a code to lustre for O_SYNC work.
1383 Description: Failure to close file and release space on NFS
1384 Details : Put inode details into lock acquired in ll_intent_file_open.
1385 Use mdc_intent_lock in ll_intent_open to properly
1386 detect all kind of errors unhandled by mdc_enqueue.
1391 Description: proc file read during shutdown sometimes raced obd removal,
1393 Details : Add lock to prevent obd access after proc file removal.
1396 Frequency : Only for files larger than 4GB on 32-bit clients.
1398 Description: improperly doing page alignment of locks
1399 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1400 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1405 Description: RPCs being resent when they shouldn't be
1406 Details : Some RPCs that should not be resent are being resent. This
1407 can cause inconsistencies in the RPC state machine. Do not
1408 resend such requests.
1411 Frequency : rare, only with NFS export
1413 Description: Crash on NFS re-export node
1414 Details : under very unusual load conditions an assertion is hit in
1415 ll_intent_file_open()
1418 Frequency : only if OST filesystem is corrupted
1420 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1421 Details : for a short time RPCs with bulk IO are in the replay list,
1422 but replay of bulk IOs is unimplemented. If the OST filesystem
1423 is corrupted due to disk cache incoherency and then replay is
1424 started it is possible to trip an assertion. Avoid putting
1425 committed RPCs into the replay list at all to avoid this issue.
1430 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1431 Details : Large single O_DIRECT read and write calls can fail to allocate
1432 a sufficiently large buffer to process the request. In case of
1433 allocation failure the allocation is retried with a smaller
1434 buffer and broken into smaller requests.
1436 Severity : enhancement
1438 Description: Add -o localflock option to simulate old noflock behaviour.
1439 Details : This will achieve local-only flock/fcntl locks coherentness.
1444 Description: versioning check is incomplete
1445 Details : Checking the version difference of client vs. server, report
1446 error if the gap is too big.
1451 Description: add support PG_writeback bit
1452 Details : add support for PG_writeback bit for Lustre, for more carefull
1453 work with page cache in 2.6 kernel. This also fix some deadlocks
1454 and remove hack for work O_SYNC with 2.6 kernel.
1456 Severity : enhancement
1458 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1459 Details : The uninit_groups feature works in conjunction with the kernel
1460 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1461 up the pass1 processing of e2fsck. This is a read-only feature
1462 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1463 mount filesystems that have had this feature enabled.
1465 Severity : enhancement
1467 Description: Improve multi-block allocation algorithm to avoid fragmentation
1468 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1469 allocation locality and avoid filesystem fragmentation.
1471 ------------------------------------------------------------------------------
1473 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1475 * Support for kernels:
1476 2.6.9-42.0.3.EL (RHEL 4)
1477 2.6.5-7.276 (SLES 9)
1478 2.4.21-47.0.1.EL (RHEL 3)
1479 2.6.12.6 vanilla (kernel.org)
1480 2.6.16.21-0.8 (SLES10)
1481 * Recommended e2fsprogs version: 1.39.cfs2-0
1483 * The backwards-compatible /proc/sys/portals symlink has been removed
1484 in this release. Before upgrading, please ensure that you change
1485 any configuration scripts or /etc/sysctl.conf files that access
1486 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1487 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1488 in advance of the upgrade on any system running Lustre 1.4.6 or
1489 newer, since /proc/sys/lnet was added in that version.
1490 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1494 Frequency : only when quota is used
1496 Description: avoid scanning export list for quota master
1497 Details : Change the algorithms to avoid scanning export list in order
1498 to improve the efficiency.
1501 Frequency : MDS failover only, very rarely
1503 Description: "went back in time" messages on mds failover
1504 Details : The greatest transno may be lost when the current operation
1505 finishes with an error (transno==0) and the client's last_rcvd
1506 record is over-written. Save the greatest transno in the
1507 mds_last_transno for this case.
1510 Frequency : always for specific kernels and striping counts
1512 Description: client may get "Matching packet too big" without ACL support
1513 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1514 when trying to access files in certain configurations. The
1515 clients should in fact be denied when mounting because they do
1516 not understand ACLs.
1519 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1521 Description: many clients connecting with IO in progress causes connect timeouts
1522 Details : Avoid synchronous journal commits to avoid delays caused by many
1523 clients connecting/disconnecting when bulk IO is in progress.
1524 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1525 OST_IO_PORTAL to avoid delays behind potentially many pending
1529 Frequency : occasionally with multiple writers to a single file
1531 Description: shared writes to file may result in wrong size reported by stat()
1532 Details : Allow growing of kms when extent lock is cancelled
1535 Frequency : always with random mmap IO to multi-striped file
1537 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1538 Details : Only if the hole is at the end of OST object so that kms is too
1539 small. Fix is to increase kms accordingly in ll_nopage.
1542 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1544 Description: writes to a missing object would leak memory on the OST
1545 Details : If there is an inconsistency between the MDS and OST filesystems,
1546 such that the MDS references an object that doesn't exist, writes
1547 to that object will leak memory due to incorrect cleanup in the
1548 error handling path, eventually running out of memory on the OST.
1553 Description: Creating too long symlink causes lustre errors
1554 Details : Check symlink and name lengths before sending requests to MDS.
1557 Frequency : only if flock is enabled (not on by default)
1559 Description: posix locks not released on fd closure on 2.6.9+
1560 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1561 kernels, this caused such locks not to be released on fd close and
1562 then assertions on fs unmount about still used locks.
1565 Frequency : MDS failover only, very rarely
1567 Description: clients may get ASSERTION(granted_lock != NULL)
1568 Details : When request was taking a long time, and a client was resending
1569 a getattr by name lock request. The were multiple lock requests
1570 with the same client lock handle and
1571 mds_getattr_name->fixup_handle_for_resent_request found one of the
1572 lock handles but later failed with ASSERTION(granted_lock != NULL).
1577 Description: handle->h_buffer_credits > 0, assertion failure
1578 Details : h_buffer_credits is zero after truncate, causing assertion
1579 failure. This patch extends the transaction or creates a new
1583 Frequency : NFS re-export or patchless client
1584 Bugzilla : 11179, 10796
1585 Description: Crash on NFS re-export node (__d_move)
1586 Details : We do not want to hash the dentry if we don't have a lock.
1587 But if this dentry is later used in d_move, we'd hit uninitialised
1588 list head d_hash, so we just do this to init d_hash field but
1589 leave dentry unhashed.
1592 Frequency : NFS re-export or patchless client
1594 Description: NFS exports has problem with symbolic link
1595 Details : lustre client didn't properly install dentry when re-exported
1596 to NFS or running patchless client.
1599 Frequency : NFS re-export or patchless client
1601 Description: Various nfs/patchless fixes.
1602 Details : fixes reuse disconected alias for lookup process - this fixes
1603 warning "find_exported_dentry: npd != pd",
1604 fix permission error with open files at nfs.
1605 fix apply umask when do revalidate.
1608 Frequency : occasional
1610 Description: Crash on NFS re-export node
1611 Details : calling clear_page() on the wrong pointer triggered oops in
1612 generic_mapping_read().
1615 Frequency : rarely, using O_DIRECT IO
1617 Description: unaligned directio crashes client with LASSERT
1618 Details : check for unaligned buffers before trying any requests.
1621 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1623 Description: stale data returned from RAID cache
1624 Details : If only a small amount of IO is done to the RAID device before
1625 reading it again it is possible to get stale data from the RAID
1626 cache instead of reading it from disk.
1629 Frequency : always for sles10 kernel
1631 Description: sles10 support
1632 Details : ll_follow_link: compile fixes and using of nd_set_link
1633 under newer kernels.
1636 Frequency : depends on arch, kernel and compiler version, always on sles10
1639 Description: recursive or deep enough symlinks cause stack overflow
1640 Details : getting rid of large stack-allocated variable in
1644 Frequency : depends on hardware
1646 Description: lustre write performance loss in the SLES10 kernel
1647 Details : the performance loss is caused by using of write barriers in the
1648 ext3 code. The SLES10 kernel turns barrier support on by
1649 default. The fix is to undo that change for ldiskfs.
1651 ------------------------------------------------------------------------------
1653 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1655 * Support for kernels:
1656 2.6.9-42.0.3EL (RHEL 4)
1657 2.6.5-7.276 (SLES 9)
1658 2.4.21-47.0.1.EL (RHEL 3)
1659 2.6.12.6 vanilla (kernel.org)
1663 Frequency : quota enabled and large files being deleted
1665 Description: releasing more than 4GB of quota at once hangs OST
1666 Details : If a user deletes more than 4GB of files on a single OST it
1667 will cause the OST to spin in an infinite loop. Release
1668 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1673 Description: statfs data retrieved from /proc may be stale or zero
1674 Details : When reading per-device statfs data from /proc, in the
1675 {kbytes,files}_{total,free,avail} files, it may appear
1676 as zero or be out of date.
1679 Frequency : systems with MD RAID1 external journal devices
1681 Description: lconf's call to blkid is confused by RAID1 journal devices
1682 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1683 of returning all block devices that match the journal UUID.
1686 Frequency : always, for aggregate stripe size over 4GB
1688 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1689 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1690 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1693 Frequency : always if "lfs find" used on a local file/directory
1695 Description: "lfs find" segfaults if used on a local file/directory
1696 Details : The case where a directory component was not specified wasn't
1697 handled correctly. Handle this properly.
1700 Frequency : always on ppc64
1702 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1703 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1704 perform endianness conversion.
1707 Frequency : rarely (truncate to non-zero file size after write under load)
1708 Bugzilla : 10730, 10687
1709 Description: Files padded with zeros to next 4K multiple
1710 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1711 it is possible that files that are truncated to a non-zero size
1712 immediately after being written are filled with zero bytes beyond
1713 the truncated size. No file data is lost.
1715 Severity : enhancement
1717 Description: Allow recovery/failover for liblustre clients.
1718 Details : liblustre clients were unaware of failover configurations until
1721 Severity : enhancement
1723 Description: user file locks should fail when not mounting with flock option
1724 Details : Set up an error-returning stub in ll_file_operations.lock field
1725 to prevent incorrect behaviour when client is mounted without
1726 flock option. Also, set up properly f_op->flock field for
1730 Frequency : always on ia64
1732 Description: "lfs df" loops on printing out MDS statfs information
1733 Details : The obd_ioctl_data was not initialized and in some systems
1734 this caused a failure during the ioctl that did not return
1735 an error. Initialize the struct and return an error on failure.
1738 Frequency : SLES 9 only
1740 Description: Error of copying files with lustre special EAs as root
1741 Details : Client side always return success for setxattr call for lustre
1742 special xattr (currently only "trusted.lov").
1745 Frequency : rarely on clusters with both ia64+i386 clients
1747 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1748 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1749 clients are doing concurrent writes to the same file it is
1750 possible that the ia64 clients may LASSERT because the OST
1751 extent locks are not PAGE_SIZE aligned. Ensure that grown
1752 locks are always aligned on the request boundary.
1755 Frequency : specific use, occasional
1757 Description: Overwriting in use executable truncates on-disk binary image
1758 Details : If one node attempts to overwrite an executable in use by
1759 another node, we now correctly return ETXTBSY instead of
1760 truncating the file.
1762 Severity : enhancement
1764 Description: Async OSC create to avoid the blocking unnecessarily.
1765 Details : If a OST has no remain object, system will block on the creating
1766 when need to create a new object on this OST. Now, ways use
1767 pre-created objects when available, instead of blocking on an
1768 empty osc while others are not empty. If we must block, we block
1769 for the shortest possible period of time.
1774 Description: chmod on Lustre root is propagated to other clients
1775 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1776 invalid by the follow_mount time.
1779 Frequency : liblustre clients only
1781 Description: Race in 'instant cancel' lock handling could lead to such locks
1782 never to be granted in case of SMP MDS
1783 Details : Do not destroy not yet granted but cbpending locks in
1787 Frequency : replay/resend of open
1789 Description: non null lock assetion failure in mds_intent_policy
1790 Details : Trying to replay/resend lockless open requests resulted in
1791 mds_open() returning 0 with no lock. Now it sets a flag if
1792 a lock is going to be returned.
1794 Severity : enhancement
1796 Description: Checksum enhancements
1797 Details : New checksum enhancements allow for resending RPCs that failed
1800 Severity : enhancement
1802 Description: Tunables on number of dirty pages in cacche
1803 Details : Allow to set limit on number of dirty pages cached.
1808 Description: client crash on unmount - lock still has references
1809 Details : In some error handling cases it was possible to leak a lock
1810 reference on a client while accessing a file. This was not
1811 harmful to the client during operation, but would cause the
1812 client to crash when the filesystem is unmounted.
1815 Frequency : specific case, rare
1817 Description: ETXTBSY on mds though file not in use by client
1818 Details : ETXTBSY is no longer incorrectly returned when attempting to
1819 chmod or chown a directory that the user previously tried to
1820 execute or a currently-executing binary.
1823 Frequency : extremely rare except on liblustre-based clients
1825 Description: Lustre space not freed when files are deleted
1826 Details : Clean up open-unlinked files after client eviction. Previously
1827 the unlink was skipped and the files remained as orphans.
1832 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1833 Details : In some cases it was possible to send a blocking callback to
1834 a client doing a glimpse, even though that client didn't get
1835 a lock granted. When the glimpse lock is cancelled on the OST
1836 the freed lock is left on the waiting list and corrupted the list.
1839 Frequency : all core dumps
1841 Description: Broke core dumps to lustre
1842 Details : Negative dentry may be unhashed if parent does not have UPDATE
1843 lock, but some callers, e.g. do_coredump, expect dentry to be
1844 hashed after successful create, hash it in ll_create_it.
1846 ------------------------------------------------------------------------------
1848 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1850 * Support for kernels:
1851 2.6.9-42.0.2.EL (RHEL 4)
1852 2.6.5-7.276 (SLES 9)
1853 2.4.21-40.EL (RHEL 3)
1854 2.6.12.6 vanilla (kernel.org)
1858 Frequency : always on RHEL 3
1860 Description: Number of open files grows over time
1861 Details : The number of open files grows over time, whether or not
1862 Lustre is started. This was due to a filp leak introduced
1863 by one of our kernel patches.
1865 ------------------------------------------------------------------------------
1867 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
1869 * Support for kernels:
1870 2.6.9-42.EL (RHEL 4)
1871 2.6.5-7.267 (SLES 9)
1872 2.4.21-40.EL (RHEL 3)
1873 2.6.12.6 vanilla (kernel.org)
1878 Bugzilla : 5719, 9635, 9792, 9684
1879 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1880 Details : If a server is under heavy load and cannot reply to new
1881 connection requests before the client resends the (re)connect,
1882 the connection handling code can behave badly if two service
1883 threads are concurrently handing separate (re)connections from
1884 the same client. Add better locking to the connection handling
1885 code, and ensure that only a single connection will be processed
1886 for a given client UUID, even if the lock is dropped.
1888 Severity : enhancement
1890 Description: add TCP zero-copy support to kernel
1891 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1892 sends if the hardware supports scatter-gather and checksumming.
1893 This allows socklnd to do client-write and server-read more
1894 efficiently and reduce CPU utilization from skbuf copying.
1897 Frequency : only if NFS exporting from client
1899 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1900 Details : If setting attributes on a file created under NFS that had
1901 never been opened it would be possible to oops the client
1902 if the file had no objects.
1906 Bugzilla : 9326, 10402, 10897
1907 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1908 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1909 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1910 before ptlrpcd_wake() was called and the request was freed.
1913 Frequency : always for liblustre
1915 Description: liblustre client does MDS+OSTs setattr RPC for each write
1916 Details : When doing a write from a liblustre client, the client
1917 incorrectly issued an RPC to the MDS and each OST the file was
1918 striped over in order to update the timestamps. When writing
1919 with small chunks and many clients this could overwhelm the MDS
1920 with RPCs. In all cases it would slow down the write because
1921 these RPCs are unnecessary.
1923 Severity : enhancement
1925 Description: allow number of MDS service threads to be changed at module load
1926 Details : It is now possible to change the number of MDS service threads
1927 running. Adding "options mds mds_num_threads={N}" to the MDS's
1928 /etc/modprobe.conf will set the number of threads for the next
1929 time Lustre is restarted (assuming the "mds" module is also
1930 reloaded at that time). The default number of threads will
1931 stay the same, 32 for most systems.
1936 Description: OST crash if filesystem is unformatted or corrupt
1937 Details : If an OST is started on a device that has never been formatted
1938 or if the filesystem is corrupt and cannot even mount then the
1939 error handling cleanup routines would dereference a NULL pointer.
1944 Description: NULL pointer deref in llap_from_page.
1945 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1946 incorrect) mapping if the page was truncated/reclaimed while it was
1947 searched for. Check for this condition and skip such pages when
1948 doing readahead. Introduce extra check to llap_from_page() to
1949 verify page->mapping->host is non-NULL (so page is not anonymous).
1952 Frequency : Sometimes when using sys_sendfile
1954 Description: "page not covered by a lock" warnings from ll_readpage
1955 Details : sendfile called ll_readpage without right page locks present.
1956 Now we introduced ll_file_sendfile that does necessary locking
1957 around call to generic_file_sendfile() much like we do in
1961 Frequency : with certain MDS communication failures at client mount time
1963 Description: NULL pointer deref after failed client mount
1964 Details : a client connection request may delayed by the network layer
1965 and not be sent until after the PTLRPC layer has timed out the
1966 request. If the client fails the mount immediately it will try
1967 to clean up before the network times out the request. Add a
1968 reference from the request import to the obd device and delay
1969 the cleanup until the network drops the request.
1972 Frequency : occasionally during client (re)connect
1974 Description: assertion failure during client (re)connect
1975 Details : processing a client connection request may be delayed by the
1976 client or server longer than the client connect timeout. This
1977 causes the client to resend the connection request. If the
1978 original connection request is replied in this interval, the
1979 client may trip an assertion failure in ptlrpc_connect_interpret()
1980 which thought it would be the only running connect process.
1983 Frequency : only with obd_echo servers and clients that are rebooted
1985 Description: kernel BUG accessing uninitialized data structure
1986 Details : When running an obd_echo server it did not start the ping_evictor
1987 thread, and when a client was evicted an uninitialized data
1988 structure was accessed. Start the ping_evictor in the RPC
1989 service startup instead of the OBD startup.
1991 Severity : enhancement
1992 Bugzilla : 10193 (patchless)
1993 Description: Remove dependency on various unexported kernel interfaces.
1994 Details : No longer need reparent_to_init, exit_mm, exit_files,
1995 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1998 Frequency : rare (only users of deprecated and unsupported LDAP config)
2000 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2001 Details : LDAP apparently contains 'lustreName' attributes instead of
2002 'name'. A simple remapping of the name is sufficient.
2005 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2007 Description: waiting_locks_callback trips kernel BUG if client is evicted
2008 Details : Running with the dump_on_timeout debug flag turned on makes
2009 it possible that the waiting_locks_callback() can try to dump
2010 the Lustre kernel debug logs from an interrupt handler. Defer
2011 this log dumping to the expired_lock_main() thread.
2013 Severity : enhancement
2015 Description: Support NFS exporting on 2.6 kernels.
2016 Details : Implement non-rawops metadata methods for NFS server to use without
2017 changing NFS server code.
2020 Frequency : very rare (synthetic metadata workload only)
2022 Description: two racing renames might cause an MDS thread to deadlock
2023 Details : Running the "racer" program may cause one MDS thread to rename
2024 a file from being the source of a rename to being the target of
2025 a rename at exactly the same time that another thread is doing
2026 so, and the second thread has already enqueued these locks after
2027 doing a lookup of the target and is trying to relock them in
2028 order. Ensure that we don't try to re-lock the same resource.
2031 Frequency : only very large systems with liblustre clients
2033 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2034 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2035 all OSTs in parallel. This allows the eviction of stale liblustre
2036 clients to proceed much faster than if they were done in series,
2037 and also offers similar improvements for other set_info RPCs.
2042 Description: excessive CPU usage during initial read phase on client
2043 Details : During the initial read phase on a client, it would agressively
2044 retry readahead on the file, consuming too much CPU and impacting
2045 performance (since 1.4.5.8). Improve the readahead algorithm
2046 to avoid this, and also improve some other common cases (read
2047 of small files in particular, where "small" is files smaller than
2048 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2053 Description: MDS crash when receiving packet with unknown intent.
2054 Details : Do not LBUG in unknown intent case, just return -EFAULT
2056 Severity : enhancement
2057 Bugzilla : 9293, 9385
2058 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2059 Details : Do not serialize getattr (non-intent version) and statfs.
2062 Frequency : occasional, when OST network is overloaded/intermittent
2064 Description: client evicted by OST after bulk IO timeout
2065 Details : If a client sends a bulk IO request (read or write) the OST
2066 may evict the client if it is unresposive to its data GET/PUT
2067 request. This is incorrect if the network is overloaded (takes
2068 too long to transfer the RPC data) or dropped the OST GET/PUT
2069 request. There is no need to evict the client at all, since
2070 the pinger and/or lock callbacks will handle this, and the
2071 client can restart the bulk request.
2074 Frequency : Always when mmapping file with no objects
2076 Description: client crashes when mmapping file with no objects
2077 Details : Check that we actually have objects in a file before doing any
2078 operations on objects in ll_vm_open, ll_vm_close and
2084 Description: Request leak when working with deleted CWD
2085 Details : Introduce advanced request refcount tracking for requests
2086 referenced from lustre intent.
2088 Severity : Enhancement
2090 Description: Cache open file handles on client.
2091 Details : MDS now will return special lock along with openhandle, if
2092 requested and client is allowed to hold openhandle, even if unused,
2093 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2094 closing files for every read/write openration.
2096 Severity : Enhancement
2098 Description: Cache open negative dentries on client when possible.
2099 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2100 negative dentries on lock revocation.
2105 Description: Remounting a client read-only wasn't possible with a zconf mount
2106 Details : It wasn't possible to remount a client read-only with llmount.
2108 Severity : enhancement
2109 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2110 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2111 MPICH is included to add Lustre-specific ADIO interfaces.
2112 This is based closely on the UFS ADIO layer and only differs
2113 in file creation, in order to allow the OST striping to be set.
2114 This is user-contributed code and not supported by CFS.
2119 Description: extended inode attributes (immutable, append-only) work improperly
2120 when 2.4 and 2.6 kernels are used on client/server or vice versa
2121 Details : Introduce kernel-independent values for these flags.
2123 Severity : enhancement
2126 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2127 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2128 granularity. Now, Fractional MB granularity is be supported,
2129 this is very useful for low memory system.
2131 Severity : enhancement
2133 Description: Getattr by fid
2134 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2135 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2136 have locks and avoids getattr rpc requests for it on every stat.
2139 Frequency : Always, for filesystems larger than 2TB
2141 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2142 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2143 filesystems larger than 2TB to be mounted with mballoc enabled.
2146 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2148 Description: filesystem corruption for non-standard kernels and very large OSTs
2149 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2150 filesystems larger than 2TB is mounted then the kernel will
2151 silently corrupt the start of the filesystem. CONFIG_LBD is
2152 enabled for all CFS-supported kernels, but the possibility of
2153 this happening with a modified kernel config exists.
2155 Severity : enhancement
2157 Description: add client O_DIRECT support for 2.6 kernels
2158 Details : It is now possible to do O_DIRECT reads and writes to files
2159 in the Lustre client mountpoint on 2.6 kernel clients.
2161 Severity : enhancement
2163 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2164 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2165 parallel, not waiting for response from every OST before sending
2166 a rpc to the next OST.
2171 Description: setattr vs write race when updating file timestamps
2172 Details : Client processes that update a file timestamp into the past
2173 right after writing to the file (e.g. tar) it is possible that
2174 the updated file modification time can be reset to the current
2175 time due to a race between processing the setattr and write RPC.
2177 Severity : enhancement
2179 Description: Bring 'lfs find' closer in line with regular Linux find.
2180 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2181 -print0 options and obtains all the needed info through the lustre
2184 Severity : enhancement
2186 Description: support up to 1024 configured devices on one node
2187 Details : change obd_dev array from statically allocated to dynamically
2188 allocated structs as they are first used to reduce memory usage
2193 Description: Flush dirty partially truncated pages during truncate
2194 Details : Immediatelly flush partially truncated pages in filter_setattr,
2195 this way we completely avoid having any pages in page cache on OST
2196 and can retire ugly workarounds during writes to flush such pages.
2201 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2202 Details : protect inode from truncation within vfs_unlink() context
2203 just take a reference before calling vfs_unlink() and release it
2204 when parent's i_sem is free.
2207 Frequency : always, if extents are used on OSTs
2209 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2210 Details : OSTs using the extents format would not zero the high 16 bits of
2211 the index physical block number. This is not a problem for any
2212 OST filesystems smaller than 16TB, and no kernels support ext3
2213 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2214 new/modified files) and can be fixed for existing filesystems
2215 with e2fsprogs-1.39-cfs1.
2220 Description: import connection selection may be incorrect if timer wraps
2221 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2222 import connections to be ignored if the 32-bit jiffies counter
2223 wraps. Use a 64-bit jiffies counter.
2226 Frequency : during server recovery
2228 Description: crash after server is denying duplicate export
2229 Details : If clients are resending connect requests to the server, the
2230 server refuses to allow a client to connect multiple times.
2231 Fixed a bug in the handling of this case.
2234 Frequency : very large clusters immediately after boot
2236 Description: LNET request buffers exhausted under heavy short-term load
2237 Details : If a large number of client requests are generated on a service
2238 that has previously never seen so many requests it is possible
2239 that the request buffer growth cannot keep up with the spike in
2240 demand. Instead of dropping incoming requests, they are held in
2241 the LND until the RPC service can accept more requests.
2244 Frequency : Sometimes during replay
2246 Description: Assertion failure in ll_local_open after replay.
2247 Details : If replay happened on an open request reply before we were able
2248 to set replay handler, reply will become not swabbed tripping the
2249 assertion in ll_local_open. Now we set the handler right after
2250 recognising of open request
2253 Frequency : very rare
2255 Description: kernel reports "badness in vsnprintf"
2256 Details : Reading from the "recovery_status" /proc file in small chunks
2257 may cause a negative length in lprocfs_obd_rd_recovery_status()
2258 call to vsnprintf() (which is otherwise harmless). Exit early
2259 if there is no more space in the output buffer.
2261 Severity : enhancement
2263 Description: clear OBD RPC statistics by writing to them
2264 Details : It is now possible to clear the OBD RPC statistics by writing
2265 to the "stats" file.
2270 Description: Client mtime is not the same on different clients after utimes
2271 Details : In some cases, the client was using the utimes() syscall on
2272 a file cached on another node. The clients now validate the
2273 ctime from the MDS + OSTs to determine which one is right.
2278 Description: Inability to activate failout mode
2279 Details : lconf script incorrectly assumed that in python string's numeric
2280 value is used in comparisons.
2283 Frequency : always with multiple stripes per file
2285 Description: Inefficient object allocation for mutli-stripe files
2286 Details : When selecting which OSTs to stripe files over, for files with
2287 a stripe count that divides evenly into the number of OSTs,
2288 the MDS is always picking the same starting OST for each file.
2289 Return the OST selection heuristic to the original design.
2294 Description: mount failures may take full timeout to return an error
2295 Details : Under some heavy load conditions it is possible that a
2296 failed mount can wait for the full obd_timeout interval,
2297 possibly several minutes, before reporting an error.
2298 Instead return an error as soon as the status is known.
2300 ------------------------------------------------------------------------------
2302 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
2304 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2305 INTEROPERATE with older versions automatically. Please read the
2306 user documentation before upgrading any part of a live system.
2307 * WARNING: Lustre networking configuration changes are required with
2308 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2311 * Support for kernels:
2312 2.6.9-22.0.2.EL (RHEL 4)
2313 2.6.5-7.244 (SLES 9)
2314 2.6.12.6 vanilla (kernel.org)
2317 Severity : enhancement
2318 Bugzilla : 7981/8208
2319 Description: Introduced Lustre Networking (LNET)
2320 Details : LNET is new networking infrastructure for Lustre, it includes
2321 a reorganized network configuration mode (see the user
2322 documentation for full details) as well as support for routing
2323 between different network fabrics. Lustre Networking Devices
2324 (LNDs) for the supported network fabrics have also been
2325 created for this new infrastructure.
2327 Severity : enhancement
2328 Description: Introduced Access control lists
2329 Details : clients can set ACLs on files and directories in order to have
2330 more fine-grained permissions than the standard Unix UGO+RWX.
2331 The MDS must be started with the "-o acl" mount option.
2333 Severity : enhancement
2334 Description: Introduced filesystem quotas
2335 Details : Administrators may now establish per-user quotas on the
2338 Severity : enhancement
2340 Description: Configuration change for the XT3
2341 The PTLLND is now used to run Lustre over Portals on the XT3
2342 The configure option(s) --with-cray-portals are no longer used.
2343 Rather --with-portals=<path-to-portals-includes> is used to
2344 enable building on the XT3. In addition to enable XT3 specific
2345 features the option --enable-cray-xt3 must be used.
2350 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2351 Details : A race between allocating a new llcd and re-getting the llcd_lock
2352 allowed another thread to grab newly-allocated llcd.
2354 Severity : enhancement
2356 Description: 2.6 OST async journal commit and locking fix to improve performance
2357 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2358 2.6 kernels are now async as they already were in 2.4 kernels so
2359 that they can commit concurrently with the network bulk transfer.
2360 For block-allocated files the filter allocation semaphore is held
2361 to avoid filesystem fragmentation during allocation. BKL lock
2362 removed for 2.6 xattr operations where it is no longer needed.
2367 Description: lconf incorrectly determined whether two IP networks could talk
2368 Details : In some more complicated routing and multiple-network
2369 configurations, lconf will avoid trying to make a network
2370 connection to a disjoint part of the IP space. It was doing the
2371 math incorrectly for one set of cases.
2376 Description: Fix for potential infinite loop processing records in an llog.
2377 Details : If an llog record is corrupted/zeroed, it is possible to loop
2378 forever in llog_process(). Validate the llog record length
2379 and skip the remainder of the block on error.
2382 Frequency : occasional (liblustre only)
2384 Description: liblustre could not open files whose last component is a symlink
2385 Details : sysio_path_walk() would incorrectly pass the open intent to
2386 intermediate path components.
2389 Frequency : rare (liblustre only with non-standard tuning)
2390 Bugzilla : 7201 (7350)
2391 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2392 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2393 after it has been released, only do this for referenced locks
2395 Severity : enhancement
2397 Description: specifying an (invalid) directory default stripe_size of -1
2398 would reset the directory default striping
2399 Details : stripe_size -1 was used internally to signal directory stripe
2400 removal, now use "all default" to signal dir stripe removal
2401 as a directory striping of "all default" is not useful
2404 Frequency : common for large clusters running liblustre clients
2406 Description: doing an ls when liblustre clients are running is slow
2407 Details : sending a glimpse AST to a liblustre client waits for every AST
2408 to time out, as liblustre clients will not respond. Since they
2409 cannot cache data we refresh the OST lock LVB from disk instead.
2411 Severity : enhancement
2413 Description: doing an ls at the same time as file IO can be slow
2414 Details : enqueue and other "small" requests can be blocked behind many
2415 large IO requests. Create a new OST IO portal for non-IO
2416 requests so they can be processed faster.
2419 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2421 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2422 created file with mode 000
2423 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2424 allow an ftruncate() as a non-root user to a file with mode 000.
2425 Lustre now respects this flag to disable mode checks when
2426 truncating a file owned by the user
2429 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2432 Description: Revoking locks from clients that went dead or catatonic might take
2434 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2435 cancellation of such locks instant on servers without waiting for
2436 any reply from clients. Clients drops these locks when cancel
2437 notification from server is received without replying.
2440 Frequency : liblustre-only, when liblustre client dies or becomes busy
2442 Description: Doing ls on Linux clients can take a long time with active
2444 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2445 granting such locks to it in the first place if possible. Locks
2446 are taken by proxy on the OST during the read or write and
2447 dropped immediately afterward. Add connect flags handling, do
2448 not grant locks to liblustre clients for glimpse ASTs.
2450 Severity : enhancement
2452 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2453 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2454 sequential reads and behaves badly with applications doing random
2455 reads. Improve it by reading ahead at least read region, and
2456 avoiding excessive large RPC for small reads.
2458 Severity : enhancement
2460 Description: Creating more than 1000 files for a single job may cause a load
2461 imbalance on the OSTs if there are also a large number of OSTs.
2462 Details : qos_prep_create() uses an OST index reseed value that is an
2463 even multiple of the number of available OSTs so that if the
2464 reseed happens in the middle of the object allocation it will
2465 still utilize the OSTs as uniformly as possible.
2470 Description: OST or MDS may oops in ping_evictor_main()
2471 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2472 but doesn't restart at beginning of obd_exports_timed list
2475 Severity : enhancement
2477 Description: improve by-nid export eviction on the MDS and OST
2478 Details : allow multiple exports with the same NID to be evicted at one
2479 time without re-searching the exports list.
2482 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2484 Description: MDS may oops in groups_free()
2485 Details : in rare race conditions a newly allocated group_info struct is
2486 freed again, and this can be NULL. The 2.4 compatibility code
2487 for groups_free() checked for a NULL pointer, but 2.6 did not.
2490 Frequency : common for liblustre clients doing little filesystem IO
2491 Bugzilla : 9352, 7313
2492 Description: server may evict liblustre clients accessing contended locks
2493 Details : if a client is granted a lock or receives a completion AST
2494 with a blocking AST already set it would not reply to the AST
2495 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2498 Frequency : lfs setstripe, only systems with more than 160 OSTs
2500 Description: unable to set striping with a starting offset beyond OST 160
2501 Details : llapi_create_file() incorrectly limited the starting stripe
2502 index to the maximum single-file stripe count.
2505 Frequency : LDAP users only
2507 Description: lconf did not handle in-kernel recovery with LDAP properly
2508 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2510 Severity : enhancement
2512 Description: bind OST threads to NUMA nodes to improve performance
2513 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2514 node and do their allocations there to localize memory access
2516 Severity : enhancement
2518 Description: llmount can determine client NID directly from Myrinet (GM)
2519 Details : the client NID code from gmnalnid was moved directly into
2520 llmount, removing the need to use this or specifying the
2521 client NID explicitly when mounting GM clients with zeroconf
2524 Frequency : if client is started with down MDS
2526 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2527 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2528 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2529 ptlrpc_import_delayed_req() because we didn't check if the
2530 request was interrupted, and we also didn't break out of the
2531 event loop if there was no timeout
2536 Description: data loss during non-page-aligned writes to a single file from
2537 both multiple nodes and multiple threads on one node at same time
2538 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2539 inconsistency led to false short-reads, that were cached and later
2540 used by ->prepare_write() to fill in partially written page,
2541 leading to data loss.
2544 Frequency : always, if lconf --abort_recovery used
2546 Description: lconf --abort_recovery fails with 'Operation not supported'
2547 Details : lconf was attempting to abort recovery on the MDT device and not
2550 Severity : enhancement
2552 Description: remove cleanup logs
2553 Details : replace lconf-generated cleanup logs with lustre internal
2554 cleanup routines. Eliminates the need for client-cleanup and
2557 Severity : enhancement
2559 Description: add support for EAs (user and system) on lustre filesystems
2560 Details : it is now possible to store extended attributes in the Lustre
2561 client filesystem, and with the user_xattr mount option it
2562 is possible to allow users to store EAs on their files also
2564 Severity : enhancement
2566 Description: Add possibility (config option) to show minimal available OST free
2568 Details : When compiled with --enable-mindf configure option, statfs(2)
2569 (and so, df) will return least minimal free space available from
2570 all OSTs as amount of free space on FS, instead of summary of
2571 free spaces of all OSTs.
2573 Severity : enhancement
2575 Description: do not expand extent locks acquired on OST-side
2576 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2577 by server: they are not cached anyway.
2580 Frequency : when mmap is used/binaries executed from Lustre
2582 Description: Unmmap pages before throwing them away from read cache.
2583 Details : llap_shrink cache now attempts to unmap pages before discarding
2584 them (if unmapping failed - do not discard). SLES9 kernel has
2585 extra checks that trigger if this unmapping is not done first.
2590 Description: lconf didn't resolve symlinks before checking to see whether a
2591 given mountpoint was already in use
2594 Frequency : when migrating failover services
2595 Bugzilla : 6395, 9514
2596 Description: When migrating a subset of services from a node (e.g. failback
2597 from a failover service node) the remaining services would
2598 time out and evict clients.
2599 Details : lconf --force (implied by --failover) sets the global obd_timeout
2600 to 5 seconds in order to quickly disconnect, but this caused
2601 other RPCs to time out too quickly. Do not change the global
2602 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2604 Severity : enhancement
2605 Frequency : if MDS is started with down OST
2606 Bugzilla : 9439,5706
2607 Description: Allow startup/shutdown of an MDS without depending on the
2608 availability of the OSTs.
2609 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2610 Add appropriate locking and lov-osc refcounts for safe
2611 cleaning. Add osc abort_inflight calls in case the
2612 synchronize never started.
2615 Frequency : occasional (Cray XT3 only)
2617 Description: root not authorized to access files in CRAY_PORTALS environment
2618 Details : The client process capabilities were not honoured on the MDS in
2619 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2620 been accessed by an authorized user then root was able to access
2621 the file on the local client also. The root user capabilities
2622 are now allowed on the MDS, as this environment has secure UID.
2625 Frequency : occasional
2627 Description: ldiskfs "too long searching" message happens too often
2628 Details : A debugging message (otherwise harmless) prints too often on
2629 the OST console. This has been reduced to only happen when
2630 there are fragmentation problems on the filesystem.
2635 Description: Division by zero in statfs when all OSCs are inactive
2636 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2637 lov_statfs divides by value returned by lov_get_stripecnt().
2641 Bugzilla : 9489, 3273
2642 Description: First write from each client to each OST was only 4kB in size,
2643 to initialize client writeback cache, which caused sub-optimal
2644 RPCs and poor layout on disk for the first writen file.
2645 Details : Clients now request an initial cache grant at (re)connect time
2646 and so that they can start streaming writes to the cache right
2647 away and always do full-sized RPCs if there is enough data.
2648 If the OST is rebooted the client also re-establishes its grant
2649 so that client cached writes will be honoured under the grant.
2654 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2655 Details : Now I/O RPCs go to different portal number and (presumably) fast
2656 lock requests (and glimses) and other RPCs get their own service
2657 threads pool that should be able to service those RPCs
2660 Severity : enhancement
2662 Description: Ability to exchange lustre version between client and servers and
2663 issue warnings at client side if client is too old. Also for
2664 liblustre clients there is ability to refuse connection of too old
2666 Details : New 'version' field is added to connect data structure that is
2667 filled with version info. That info is later checked by server and
2671 Frequency : rare, liblustre only.
2672 Bugzilla : 9296, 9581
2673 Description: Two simultaneous writes from liblustre at offset within same page
2674 might proceed at the same time overwriting eachother with stale
2676 Details : I/O lock withing llu_file_prwv was released too early, before data
2677 actually was hitting the wire. Extended lock-holding time until
2678 server acknowledges receiving data.
2681 Frequency : extremely rare. Never observed in practice.
2683 Description: avoid generating lustre_handle cookie of 0.
2684 Details : class_handle_hash() generates handle cookies by incrementing
2685 global counter, and can hit 0 occasionaly (this is unlikely, but
2686 not impossible, because initial value of cookie counter is
2687 selected randonly). Value of 0 is used as a sentinel meaning
2688 "unassigned handle" --- avoid it. Also coalesce two critical
2689 sections in this function into one.
2691 Severity : enhancement
2693 Description: allow liblustre clients to delegate truncate locking to OST
2694 Details : To avoid overhead of locking, liblustre client instructs OST to
2695 take extent lock in ost_punch() on client's behalf. New connection
2696 flag is added to handle backward compatibility.
2698 Severity : enhancement
2699 Bugzilla : 4928, 7341, 9758
2700 Description: allow number of OST service threads to be specified
2701 Details : a module parameter allows the number of OST service threads
2702 to be specified via "options ost ost_num_threads={N}" in the
2703 OSS's /etc/modules.conf or /etc/modprobe.conf.
2707 Bugzilla : 6146, 9635, 9895
2708 Description: servers crash with bad pointer in target_handle_connect()
2709 Details : In rare cases when a client is reconnecting it was possible that
2710 the connection request was the last reference for that export.
2711 We would temporarily drop the export reference and get a new
2712 one, but this may have been the last reference and the export
2713 was just destroyed. Get new reference before dropping old one.
2715 Severity : enhancement
2716 Frequency : if client is started with failover MDS
2718 Description: Allow multiple MDS hostnames in the mount command
2719 Details : Try to read the configuration from all specified MDS
2720 hostnames during a client mount in case the "primary"
2723 Severity : enhancement
2725 Description: Stop sending data to evicted clients as soon as possible.
2726 Details : Check if the client we are about to send or are sending data to
2727 was evicted already. (Check is done every second of waiting,
2728 for which l_wait_event interface was extended to allow checking
2729 of exit condition at specified intervals).
2732 Frequency : rare, normally only when NFS exporting is done from client
2734 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2736 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2739 Frequency : always, if srand() is called before liblustre initialization
2741 Description: Liblustre uses system PRNG disturbing its usage by user application
2742 Details : Introduce internal to lustre fast and high-quality PRNG for
2743 lustre usage and make liblustre and some other places in generic
2744 lustre code to use it.
2746 Severity : enhancement
2747 Bugzilla : 9477, 9557, 9870
2748 Description: Verify that the MDS configuration logs are updated when xml is
2749 Details : Check if the .xml configuration logs are newer than the config
2750 logs stored on the MDS and report an error if this is the case.
2751 Request --write-conf, or allow starting with --old_conf.
2753 Severity : enhancement
2755 Description: Handle symlinks in the path when checking if Lustre is mounted.
2756 Details : Resolve intermediate symlinks when checking if a client has
2757 mounted a filesystem to avoid duplicate client mounts.
2762 Description: lconf can hit an error exception but still return success.
2763 Details : The lconf command catches the Command error exception at the top
2764 level script context and will exit with the associated exit
2765 status, but doesn't ensure that this exit status is non-zero.
2770 Description: failure of ptlrpc thread startup can cause oops
2771 Details : Starting a ptlrpc service thread can fail if there are a large
2772 number of threads or the server memory is very fragmented.
2773 Handle this without oopsing.
2776 Frequency : always, only if liblustre and non-default acceptor port was used
2778 Description: liblustre cannot connect to servers with non-default acceptor port
2779 Details : tcpnal_set_default_params() was not called and was therefore
2780 ignoring the environment varaible TCPNAL_PORT, as well as other
2781 TCPNAL_ environment variables
2786 Description: two objects could be created on the same OST for a single file
2787 Details : If an OST is down, in some cases it was possible to create two
2788 objects on a single OST for a single file. No problems other
2789 than potential performance impact and spurious error messages.
2793 Bugzilla : 5681, 9562
2794 Description: Client may oops in ll_unhash_aliases
2795 Details : Client dcache may become inconsistent in race condition.
2796 In some cases "getcwd" can fail if the current directory is
2802 Description: Inode refcounting problems in NFS export code
2803 Details : link_raw functions used to call d_instantiate without obtaining
2804 extra inode reference first.
2808 Bugzilla : 9942, 9903
2809 Description: Referencing freed requests leading to crash, memleaks with NFS.
2810 Details : We used to require that call to ll_revalidate_it was always
2811 followed by ll_lookup_it. Also with revalidate_special() it is
2812 possible to call ll_revalidate_it() twice for the same dentry
2813 even if first occurence returned success. This fix changes semantic
2814 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2815 reference on a request referred from the intent.
2816 ll_intent_release() then releases such a request.
2819 Frequency : rare, normally benchmark loads only
2821 Description: unlinked inodes were kept in memory on the client
2822 Details : If a client is repeatedly creating and unlinking files it
2823 can accumulate a lot of stale inodes in the inode slab cache.
2824 If there is no other client load running this can cause the
2825 client node to run out of memory. Instead flush old inodes
2826 from client cache that have the same inode number as a new inode.
2829 Frequency : SLES9 2.6.5 kernel and long filenames only
2830 Bugzilla : 9969, 10379
2831 Description: utime reports stale NFS file handle
2832 Details : SLES9 uses out-of-dentry names in some cases, which confused
2833 the lustre dentry revalidation. Change it to always use the
2837 Frequency : rare, unless heavy write-truncate concurrency is continuous
2838 Bugzilla : 4180, 6984, 7171, 9963, 9331
2839 Description: OST becomes very slow and/or deadlocked during object unlink
2840 Details : filter_destroy() was holding onto the parent directory lock
2841 while truncating+unlinking objects. For very large objects this
2842 may block other threads for a long time and slow overall OST
2843 responsiveness. It may also be possible to get a lock ordering
2844 deadlock in this case, or run out of journal credits because of
2845 the combined truncate+unlink. Solution is to do object truncate
2846 first in one transaction without parent lock, and then do the
2847 final unlink in a new transaction with the parent lock. This
2848 reduces the lock hold time dramatically.
2851 Frequency : rare, 2.4 kernels only
2853 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2854 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2855 before unmounting their filesystem in order to avoid deadlock.
2856 The kernel_locked() function in 2.4 kernels only checks whether
2857 the kernel lock is held, not whether it is this process that is
2858 holding it as 2.6 kernels do.
2863 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2864 Details : The client ptlrpc code may be trying to reconnect to a down
2865 server before a previous connection attempt has timed out.
2866 Increase the reconnect interval to be longer than the connection
2867 timeout interval to avoid sending duplicate connections to
2871 Frequency : echo_client brw_test command
2873 Description: fix echo_client to work with OST preallocated code
2874 Details : OST preallocation code (5137) didn't take echo_client IO path
2875 into account: echo_client calls filter methods outside of any
2876 OST thread and, hence, there is no per-thread preallocated
2877 pages and buffers to use. Solution: hijack pga pages for IO. As
2878 a byproduct, this avoids unnecessary data copying.
2882 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2883 Description: Client can oops in mdc_commit_close() after open replay
2884 Details : It was possible for the MDS to return an open request with no
2885 transaction number in mds_finish_transno() if the client was
2886 evicted, but without actually returning an error. Clients
2887 would later try to replay that open and may trip an assertion
2888 Simplify the client close codepath, and always return an error
2889 from the MDS in case the open is not successful.
2892 Frequency : rare, 2.6 OSTs only
2894 Description: OST may deadlock under high load on fragmented files
2895 Details : If there was a heavy load and highly-fragmented OST filesystems
2896 it was possible to have all the OST threads deadlock waiting on
2897 allocation of biovecs, because the biovecs were not released
2898 until the entire RPC IO was completed. Instead, release biovecs
2899 as soon as they are complete to ensure forward IO progress.
2901 Severity : enhancement
2903 Description: Support for specifying external journal device at mount
2904 Details : If an OST or MDS device is formatted with an external journal
2905 device, this device major/minor is stored in the ext3 superblock
2906 and may not be valid for failover. Allow detecting and
2907 specifying the external journal at mount time.
2912 Description: Mounting an MDS with pending unlinked files may cause oops
2913 Details : target_finish_recovery() calls mds_postrecov() which returned
2914 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2915 considers this an error and immediately begins cleaning up the
2916 lov, just after starting the mds_lov process
2918 Severity : enhancement
2920 Description: Implement 'lfs df' to report actual free space on per-OST basis
2921 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2922 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2923 usage of objects; '-h' to report in human readable format.
2925 ------------------------------------------------------------------------------
2927 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
2934 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2935 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2936 at startup during orphan unlinking. Instead, skip buddy update
2937 before setup, as it will be regenerated after recovery is complete.
2942 Description: If an OST is inactive, its locks might reference stale inodes.
2943 Details : lov_change_cbdata() must iterate over all namespaces, even if
2944 they are inactive to clear inode references from the lock.
2946 Severity : enhancement
2947 Frequency : occasional, if non-standard max_dirty_mb used
2949 Description: Client will block write RPCs if not enough grant
2950 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2951 then the client will block writes while waiting for another RPC
2952 to complete instead of consuming its dirty limit. With change
2953 we get improved performance when max_dirty_mb is small.
2955 Severity : enhancement
2956 Bugzilla : 3389, 6253
2957 Description: Add support for supplementary groups on the MDS.
2958 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2959 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2960 lookups for user supplementary groups into a cache.
2964 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2965 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2966 for liblustre O_EXCL is enforced by MDS.
2970 Description: readdir never returns NULL in liblustre.
2971 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2972 dentry in struct dirent.
2976 Frequency : liblustre only, depends on application IO pattern
2977 Description: liblustre clients evicted if not contacting servers
2978 Details : Don't put liblustre clients into the ping_evictor list, so
2979 they will not be evicted by the pinger ever.
2981 Severity : enhancement
2983 Description: Add ability to evict clients by NID from MDS.
2984 Details : By echoing "nid:$NID" string into
2985 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2986 $NID would be instantly evicted from this MDS and from all active
2987 OSTs connected to it.
2991 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2992 Details : lookup_it_finish() used to query file size from OSTs that was not
2997 Description: service threads change working directory to that of init
2998 Details : Starting lustre service threads may pin the working directory
2999 of the parent thread, making that filesystem busy. Threads
3000 now change to the working directory of init to avoid this.
3004 Frequency : during shutdown only
3005 Description: shutdown with a failed MDS or OST can cause unmount to hang
3006 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3011 Frequency : 2.6 only
3012 Description: chmod/chown may include an extra supplementary group
3013 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3014 supplementary group and if none is specified this is used.
3017 Bugzilla : 5479 (6816)
3019 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3020 Details : If lookup is in progress on a file that is unlinked we might try
3021 to revalidate the inode and fail in revalidate after lookup is
3022 complete and ll_file_open() enqueues the open again but
3023 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3026 Frequency : always, if lconf --abort_recovery used
3028 Description: lconf --abort_recovery fails with 'Operation not supported'
3029 Details : lconf was attempting to abort recovery on the MDT device and not
3032 ------------------------------------------------------------------------------
3034 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3039 Frequency : rare (only unsupported configurations with a node running as an
3041 Bugzilla : 6514, 5137
3042 Description: Mounting a Lustre file system on a node running as an OST could
3044 Details : OSTs now preallocates memory needed to write out data at
3045 startup, instead of when needed, to avoid having to
3046 allocate memory in possibly low memory situations.
3047 Specifically, if the file system is mounted on on OST,
3048 memory pressure could force it to try to write out data,
3049 which it needed to allocate memory to do. Due to the low
3050 memory, it would be unable to do so and the node would
3051 become unresponsive.
3053 Severity : enhancement
3055 Description: Addition of lconf --service command line option
3056 Details : lconf now accepts a '--service <arg>' option, which is
3057 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3059 Severity : enhancement
3061 Description: Failover mode is now the default for OSTs.
3062 Details : By default, OSTs will now run in failover mode. To return to
3063 the old behaviour, add '--failout' to the lmc line for OSTs.
3065 Severity : enhancement
3067 Description: Health checks are now provided for MDS and OSTs
3068 Details : Additional detailed health check information on MSD and OSTs
3069 is now provided through the procfs health_check value.
3072 Frequency : occasional, depends on IO load
3074 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3075 after numerous create/delete cycles
3076 Details : The ext3 inode allocation policy would not allocate new inodes
3077 very well on the OSTs because there are no new directories
3078 being created. Instead we look for groups with free space if
3079 the parent directories are nearly full.
3084 Description: Network or server problems during mount may cause partially
3085 mounted clients instead of returning an error.
3086 Details : The config llog parsing code may overwrite the error return
3087 code during mount error handling, returning success instead
3093 Description: MDS can fail to allocate large reply buffers
3094 Details : After long uptimes the MDS can fail to allocate large reply
3095 buffers (e.g. zconf client mount config records) due to memory
3096 fragmentation or consumption by the buffer cache. Preallocate
3097 some large reply buffers so that these replies can be sent even
3098 under memory pressure.
3102 Frequency : rare (liblustre)
3103 Description: fsx running with liblustre complained that using truncate() to
3104 extend the file doesn't work. This patch corrects that issue.
3105 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3106 ATTR_SIZE and lsm use in llu_setattr_raw.
3110 Frequency : rare, only 2.6 kernels
3111 Description: Unusual file access patterns on the MDS may result in inode
3112 data being lost in very rare circumstances.
3113 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3114 code in the 2.6 kernel caused the inode and/or EA data not to
3115 be read from disk, causing single-file corruption.
3119 Frequency : rare, only 2.6 filesystems using extents
3120 Description: Heavy concurrent write and delete load may cause data corruption.
3121 Details : It was possible under high-load situations to have an extent
3122 metadata block in the block device cache from a just-unlinked
3123 file overwrite a newly-allocated data block. We now unmap any
3124 metadata buffers that alias just-allocated data blocks.
3128 Frequency : filesystems with default stripe_count larger than 77
3129 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3130 Details : lconf specifies an inode size of 4096 bytes when the default
3131 stripe_count is larger than 77. This conflicts with the default
3132 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3133 this case to avoid pinning too much memory for large EAs.
3135 ------------------------------------------------------------------------------
3137 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3142 Frequency : rare (extremely heavy IO load with hundreds of clients)
3144 Description: Client is evicted, gets IO error writing to file
3145 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3146 caused clients to be evicted for AST timeouts. The fixes in
3147 bug 5192 mean we no longer need to have such short AST timeouts
3148 so ldlm_timeout has been increased.
3151 Frequency : occasional during --force or --failover shutdown under load
3152 Bugzilla : 5949, 4834
3153 Description: Server oops/LBUG if stopped with --force or --failover under load
3154 Details : a collection of import/export refcount and cleanup ordering
3155 issues fixed for safer force cleanup
3158 Frequency : only filesystems larger than 120 OSTs
3159 Bugzilla : 5990, 6223
3160 Description: lfs getstripe would oops on a very large filesystem
3161 Details : lov_getconfig used kfree on vmalloc'd memory
3164 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3165 Bugzilla : 6242, 6243
3166 Description: reading from files that had been truncated to a non-zero size
3167 but never opened returned no data
3168 Details : ll_file_read() reads zeros from no-object files to EOF
3173 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3174 Details : There's little harm in aborting MDS/OSS recovery and letting it
3175 try again, so I removed the LASSERT and return an error instead.
3177 Severity : enhancement
3179 Description: New debugging infrastructure for tracking down data corruption
3180 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3181 (b) cover more of the client-side code path, and (c) try to narrow
3182 down where problems occurred
3186 Bugzilla : 3819, 4364, 4397, 6313
3187 Description: Racing close and eviction MDS could cause assertion in mds_close
3188 Details : It was possible to get multiple mfd references during close and
3189 client eviction, leading to one thread referencing a freed mfd.
3191 Severity: : enhancement
3192 Bugzilla : 3262, 6359
3193 Description: Attempts to reconnect to servers are now more aggressive.
3194 Details : This builds on the enhanced upcall-less recovery that was added
3195 in 1.4.2. When trying to reconnect to servers, clients will
3196 now try each server in the failover group every 10 seconds. By
3197 default, clients would previously try one server every 25 seconds.
3202 Description: After recovery, certain operations trigger a failed
3203 assertion on a client.
3204 Details : Failing over an mds, using lconf -d --failover, while a
3205 client was doing a readdir() call would cause the client to
3206 LBUG after recovery completed and the readdir() was resent.
3208 Severity : enhancement
3210 Description: Default groups are now added by lconf
3211 Details : You can now run lconf --group <servicename> without having to
3212 manually add groups with lmc.
3215 Frequency : occasional
3217 Description: Nodes with an elan id of 0 trigger a failed assertion
3220 Frequency : always when accessing e.g. tty/console device nodes
3222 Description: tty and some other devices nodes cannot be used on lustre
3223 Details : file's private_data field is used by device data and lustre
3224 values in there got lost. New field was added to struct file to
3225 store fs-specific private data.
3228 Frequency : when exporting Lustre via NFS
3230 Description: NFSD failed occasionally when looking up a path component
3231 Details : NFSD is looking up ".." which was broken in ext3 directories
3232 that had grown large enough to become hashed.
3235 Frequency : Clusters with multiple interfaces not on the same subnet
3237 Description: Nodes will repeatedly try to reconnect to an interface which it
3238 cannot reach and report an error to the log.
3239 Details : Extra peer list entries will be created by lconf with some peers
3240 unreachable. lconf now validates the peer before adding it.
3243 Frequency : Only if a default stripe is set on the filesystem root.
3245 Description: Setting a default stripe on the filesystem root prevented the
3246 filesystem from being remounted.
3247 Details : The client was sending extra request flags in the root getattr
3248 request and did not allocate a reply buffer for the dir EA.
3251 Frequency : occasional, higher if lots of files are accessed by one client
3252 Bugzilla : 6159, 6097
3253 Description: Client trips assertion regarding lsm mismatch/magic
3254 Details : While revalidating inodes the VFS looks up inodes with ifind()
3255 and in rare cases can find an inode that is being freed.
3256 The ll_test_inode() code will free the lsm during ifind()
3257 when it finds an existing inode and then the VFS later attaches
3258 this free lsm to a new inode.
3262 Bugzilla : 6422, 7030
3263 Description: MDS deadlock between mkdir and client eviction
3264 Details : Creating a new file via mkdir or mknod (starting a transaction
3265 and getting the ns lock) can deadlock with client eviction
3266 (gets ns lock and trying to finish a synchronous transaction).
3269 Frequency : occasional
3270 Description: While starting a server, the fsfilt_ext3 module could not be
3272 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3273 kernels. Previously, lconf would still use the ext3 name
3274 when trying to load modules. Now, it will correctly use
3275 ext3 on 2.4 and ldiskfs on 2.6.
3277 Severity : enhancement
3278 Description: The default stripe count has been changed to 1
3279 Details : The interpretation of the default stripe count (0, to lfs
3280 or lmc) has been changed to mean striping across a single
3281 OST, rather than all available. For general usage we have
3282 found a stripe count of 1 or 2 works best.
3284 Severity : enhancement
3285 Description: Add support for compiling against Cray portals.
3286 Details : Conditional compiling for some areas that are different
3290 Frequency : occasional
3291 Bugzilla : 6409, 6834
3292 Description: Creating files with an explicit stripe count may lead to
3293 a failed assertion on the MDS
3294 Details : If some OSTs are full or unavailable, creating files may
3295 trigger a failed assertion on the MDS. Now, Lustre will
3296 try to use other servers or return an error to the
3300 Frequency : occasional
3302 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3303 caused lock timeout during readahead (since 1.4.2).
3304 Details : Processes doing readahead might match a lock that hasn't been
3305 granted yet if there are overlapping and conflicting lock
3306 requests. The readahead process waits on ungranted lock
3307 (original lock is CBPENDING), while OST waits for that process
3308 to cancel CBPENDING read lock and eventually evicts client.
3310 Severity : enhancement
3312 Description: Initial enabling of flock support for clients
3313 Details : Implements fcntl advisory locking and file status functions.
3314 This feature is provided as an optional mount flag (default
3315 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3316 locking are implemented yet, and those that are are not guaranteed
3317 to be completely correct in production environments.
3318 mount -t lustre -o [flock|noflock] ...
3321 Frequency : occasional
3323 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3324 assertion in the ext3 JBD (journaling) layer.
3325 Details : The b_committed_data struct is protected by the big kernel lock
3326 in 2.4 kernels, serializing journal_commit_transaction() and
3327 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3328 access to this struct. In 2.6 kernels there is finer grained
3329 locking to improve SMP performance of the JBD layer.
3333 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3335 -----------------------------------------------------------------------------
3337 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3339 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3340 versions of Lustre. Please update all servers and clients to
3341 version 1.4.2 or later at the same time. You must also run
3342 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3343 to update the configuration logs.
3345 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3346 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3347 - swab llog records of type '0' so we get proper header size/idx (5861)
3348 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3349 - fix rename of one directory over another leaking an inode (5953)
3350 - avoid SetPageDirty on 2.6 (5981)
3351 - don't re-add just-being-destroyed locks to the waiting list (5653)
3352 - when creating new directories, inherit the parent's custom
3353 striping settings if present parent (3048)
3354 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3355 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3356 - don't hold client locks on temporary worklist from l_lru (5666)
3357 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3358 - automatically evict dead clients (5921)
3359 - Update file size properly in create+truncate+fstat case (6196)
3360 - Do not unhash mountpoint dentries, do not allow removal of
3362 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3363 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3364 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3365 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3366 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3367 - Correctly handle reads of files with no objects (6243)
3368 - lctl recover will also mark a device active if deactivate used (5933)
3370 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3371 - allow --write-conf on an MDS with different nettype than client (5619)
3372 - don't write config llogs to MDS for mounts not from that MDS (5617)
3373 - lconf should create multiple TCP connections from a client (5201)
3374 - init scripts are now turned off by default; run chkconfig --on
3375 lustre and chkconfig --on lustrefs to use them
3376 - upcalls are no longer needed for clients to recover to failover
3378 - add --abort-recovery option to lconf to abort recovery on device
3380 - add support for an arbitrary number of OSTs (3026)
3381 - Quota support protocol changes.
3382 - forward compatibility changes to wire structs (6007)
3383 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3384 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3385 - improved "lustre status" script
3386 - initialize blocksize for non-regular files (6062)
3387 - added --disable-server and --disable-client configure options (5782)
3388 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3389 - Vanilla 2.4.29 support
3390 - increase maximum number of obd devices to 520 (6242)
3391 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3392 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3393 SLES 9 2.6.5 kernels
3394 - limit stripes per file to 160 (the maximum EA size) (6093)
3396 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3399 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3400 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3401 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3402 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3403 - small ext3 extents cleanups and fixes (5733)
3404 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3405 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3406 - avoid needless client->OST connect, fix handle mismatch (5317)
3407 - fix DLM error path that led to out-of-sync client, long delays (5779)
3408 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3409 - fix several locking issues related to i_size (5492,5624,5654,5672)
3410 - don't move pending lock onto export if it is already evicted (5683)
3411 - fix kernel oops when creating .foo in unlinked directory (5548)
3412 - fix deadlock in obdfilter statistics vs. object create (5811)
3413 - use time_{before,after} to avoid timer jiffies wrap (5882)
3414 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3415 - Do not leak request if server was not able to process it (5154)
3416 - If mds_open unable to find parent dir, make that negative lookup(5154)
3417 - don't create new directories with extent-mapping (5909, 5936)
3419 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3420 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3421 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3422 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3423 - support for RHEL4 kernel on i686 (5773)
3424 - provide error messages when incompatible logs are encountered (5898)
3426 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3427 * version 1.4.0.10 (1.4.1 release candidate 1)
3429 - don't keep a lock reference when lock is not granted (4238)
3430 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3431 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3432 - drop import inflight refcount on signal_completed_replay error (5255)
3433 - unlock page after async write error during send (3677)
3434 - handle missing objects in filter_preprw_read properly (5265)
3435 - no transno return for symlink open, don't save no-trasno open (3440)
3436 - don't try to complete elan receive that already failed (4012)
3437 - free RPC server reply state on error (5406)
3438 - clean up thread from ptlrpc_start_thread() on error (5160)
3439 - readahead could read extra page into cache that wasn't ejected (5388)
3440 - prevent races in class_attach/setup/cleanup/detach (5260)
3441 - don't dereference de->d_inode after l_dput of de (5458)
3442 - use "int" for stripe value returned from lock_to_stripe (5544)
3443 - mballoc allocation and error-checking fixes in 2.6 (5504)
3444 - block device patches to fix I/O request sizes in 2.6 (5482)
3445 - look up hostnames for IB nals (5602)
3446 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3447 - don't start multiple acceptors for the same port (5277)
3448 - fix incorrect LASSERT in mds_getattr_name (5635)
3449 - export a proc file for general "ping" checking (5628)
3450 - fix "lfs check" to not block when the MDS is down (5628)
3452 - service request history (4965)
3453 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3454 - create an "evict_client" /proc entry on OSTs, like the MDS has
3455 - fix mount usage message, return errors per mount(8) (5168)
3456 - change grep [] to grep "[]" in tests so they work in more UMLs
3457 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3458 - remove ancient LOV_MAGIC_V0 EA support (5047)
3459 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3460 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3461 - allow readahead allocations to fail when low on memory (5383)
3462 - mmap locking landed again, after considerable improvement (2828)
3463 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3465 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3468 - send OST transaction number in read/write reply to free req (4966)
3469 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3470 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3471 - the watchdog thread now runs as interruptible (5246)
3472 - flock/lockf fixes (but it's still disabled, pending 5135)
3473 - don't use EXT3 constants in llite code (5094)
3474 - memory shortage at startup could cause assertion (5176)
3476 - reorganization of lov code
3477 - single portals codebase
3479 - add extents/mballoc support (5025)
3480 - direct I/O reads in the obdfilter (4048)
3481 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3483 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3486 - send OST transaction number in read/write reply to free req (4966)
3487 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3488 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3489 - the watchdog thread now runs as interruptible (5246)
3490 - handle missing objects in filter_preprw_read properly (5265)
3491 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3492 - drop import inflight refcount on signal_completed_replay error (5255)
3493 - unlock page after async write error during send (3677)
3494 - return original error code on reconstructed replies (3761)
3495 - no transno return for symlink open, don't save no-trasno open (3440)
3497 - add pid to ldlm debugging output (4922)
3498 - bump the watchdog timeouts -- we can't handle 30sec yet
3499 - extra debugging for orphan dentry/inode bug (5259)
3501 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3504 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3505 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3506 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3507 - fix NULL dereference in /proc/sys/portals/routes (4827)
3508 - allow failed mdc_close() operations to be interrupted (4561)
3509 - stop precreate on OST before MDS would time out on it (4778)
3510 - don't send partial-page writes before EOF from client (4410)
3511 - discard client grant for sub-page writes on large-page clients (4520)
3512 - don't free dentries not owned by NFS code, check generation (4806)
3513 - fix lsm leak if mds_create_objects() fails (4801)
3514 - limit debug_daemon file size, always print CERROR messages (4789)
3515 - use transno after validating reply (3892)
3516 - process timed out requests if import state changes (3754)
3517 - update mtime on OST during writes, return in glimpse (4829)
3518 - add mkfsoptions to LDAP (4679)
3519 - use ->max_readahead method instead of zapping global ra (5039)
3520 - don't interrupt __l_wait_event() during strace
3522 - add software watchdogs to catch hung threads quickly (4941)
3523 - make lustrefs init script start after nfs is mounted
3524 - fix CWARN/ERROR duplication (4930)
3525 - return async write errors to application if possible (2248)
3526 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3527 - print NAL number in %x format (4645)
3528 - update barely-supported suse-2.4.21-171 series (4842)
3529 - support for sles 9 %post scripts
3530 - support for building 2.6 kernel-source packages
3531 - support for sles km_* packages
3533 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3536 - ignore -ENOENT errors in osc_destroy (3639)
3537 - notify osc create thread that OSC is being cleaned up (4600)
3538 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3539 - reconstruct ost_handle() like mds_handle() (4657)
3540 - create a new thread to do import eviction to avoid deadlock (3969)
3541 - let lconf resolve symlinked-to devices (4629)
3542 - don't unlink "objects" from directory with default EA (4554)
3543 - hold socknal file ref over connect in case target is down (4394)
3544 - allow more than 32000 subdirectories in a single directory (3244)
3545 - fix blocks count for O_DIRECT writes (3751)
3546 - OST returns ENOSPC from object create when no space left (4539)
3547 - don't send truncate RPC if file size isn't changing (4410)
3548 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3549 - bind to privileged port in socknal and tcpnal (3689)
3551 - rate limit CERROR/CWARN console message to avoid overload (4519)
3552 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3553 - basic mmap support (3918)
3554 - kernel patch series update from b1_4 (4711)
3556 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3559 - avoid crash during MDS cleanup with OST shut down (2775)
3560 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3561 - don't use bad inodes on the MDS (3744)
3562 - dynamic object preallocation to improve recovery speed (4236)
3563 - don't hold spinlock over lock dumping or change debug flags (4401)
3564 - don't zero obd_dev when it is force cleaned (3651)
3565 - print grants to console if they go negative (4431)
3566 - "lctl deactivate" will stop automatic recovery attempts (3406)
3567 - look for existing locks in ldlm_handle_enqueue() (3764)
3568 - don't resolve lock handle twice in recovery avoiding race (4401)
3569 - revalidate should check working dir is a directory (4134)
3571 - don't always mark "slow" obdfilter messages as errors (4418)
3573 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3576 - don't close LustreDB during write_conf until it is done (3860)
3577 - fix typo in lconf for_each_profile (3821)
3578 - allow dumping logs from multiple threads at one time (3820)
3579 - don't allow multiple threads in OSC recovery (3812)
3580 - fix debug_size parameters (3864)
3581 - fix mds_postrecov to initialize import for llog ctxt (3121)
3582 - replace config semaphore with spinlock (3306)
3583 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3584 - don't allow enqueue to complete on a destroyed export (3822)
3585 - down write_lock before checking llog header bitmap (3825)
3586 - recover from lock replay timeout (3764)
3587 - up llog sem before sending rpc (3652)
3588 - reduce ns lock hold times when setting kms (3267)
3589 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3590 - fix NULL deref and obd_dev leak on setup error (3312)
3591 - replace some LBUG about llog ops with error handling (3841)
3592 - don't match INVALID dentries from d_lookup and spin (3784)
3593 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3594 - fix invalid assertion in ptlrpc_set_wait (3880)
3596 - add libwrap support for the TCP acceptor (3996)
3597 - add /proc/sys/portals/routes for non-root route listing (3994)
3598 - allow setting MDS UUID in .xml (2580)
3599 - print the stack of a process that LBUGs (4228)
3601 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3604 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3605 - make sure to unset replay flag from failed open requests (3440)
3606 - if default stripe count is 0, use OST count for inode size (3636)
3607 - update parent mtime/ctime on client for create/unlink (2611)
3608 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3609 - free recovery state on server during a forced cleanup (3571)
3610 - unregister_reply for resent reqs (3063)
3611 - loop back devices mounting and status check on 2.6 (3563)
3612 - fix resource-creation race that can provoke i_size == 0 (3513)
3613 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3614 - more debugging for page-accounting assertion (3746)
3615 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3616 - avoid deadlock after precreation failure (3758)
3617 - fix race and lock order deadlock in orphan handling (3450, 3750)
3618 - add validity checks when grabbing inodes from l_ast_data (3599)
3620 - add /proc/.../recovery_status to obdfilter (3428)
3621 - lightweight CDEBUG infrastructure, debug daemon (3668)
3622 - change default OSC RPC parameters to be better on small clusters
3623 - turn off OST read cache for files smaller than 32MB
3624 - install man pages and include them in rpms (3100)
3625 - add new init script for (un)mounting lustre filesystems (2593)
3626 - run chkconfig in %post for init scripts (3701)
3627 - drop scimac NAL (unmaintained)
3629 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3632 - clean kiobufs before and after use (3485)
3633 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3634 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3635 - change init script to fail more clearly if not run as root (1528)
3636 - allow clients to reconnect during replay (1742)
3637 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3638 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3639 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3640 - fix cond_resched (3554)
3641 - only evict unfinished clients after recovery (3515)
3642 - allow bulk resend, prevent data loss (3570)
3643 - dynamic ptlrpc request buffer allocation (2102)
3644 - don't allow unlinking open directory if it isn't empty (2904)
3645 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3646 - remove extraneous obd dereference causing LASSERT failure (3334)
3647 - don't use get_cycles() when creating temp. files on the mds (3156)
3648 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3649 - handle EEXIST for set-stripe, set proper directory name (3336)
3651 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3652 - fix ksocknal_fmb_callback() error messages (2918)
3654 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3657 - don't copy lvb into (possibly NULL) reply on error (2983)
3658 - don't deref dentry after dput, don't free lvb on error (2922)
3659 - use the kms to determine writeback rpc length (2947)
3660 - increment oti_logcookies when osc is inactive (2948)
3661 - update client's i_blocks count via lvb messages (2543)
3662 - handle intent open/close of special files properly (1557)
3663 - mount MDS with errors=remount-ro, like obdfilter (2009)
3664 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3665 - don't use cancelling-locks' kms values (2947)
3666 - use highest lock extent for kms, not last one (2925)
3667 - don't dereference ERR_PTR() dentry in error handling path (3107)
3668 - fix thread race in portals_debug_dumplog() (3122)
3669 - create lprocfs device entries at setup instead of at attach (1519)
3670 - common AST error handler, don't evict client on completion race (3145)
3671 - zero nameidata in detach_mnt in 2.6 (3118)
3672 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3673 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3674 - initialize RPC timeout timer earlier for 2.6 (3219)
3675 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3676 - print nal/nid for unknown nid (3258)
3677 - additional checks for oscc recovery before doing precreate (3284)
3678 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3679 - don't crash in mdc_close for bad permissions on open (3285)
3680 - zero i_rdev for non-device files (3147)
3681 - clear page->private before handing to FS, better assertion (3119)
3682 - tune the read pipeline (3236)
3683 - fix incorrect decref of invalidated dentry (2350)
3684 - provide read-ahead stats and refine rpc in flight stats (3328)
3685 - don't hold journal transaction open across create RPC (3313)
3686 - update atime on MDS at close time (3265)
3687 - close LDAP connection when recovering to avoid server load (3315)
3688 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3689 - don't leak open file on MDS after open resend (3325)
3690 - serialize filter_precreate and filter_destroy_precreated (3329)
3691 - loop device shouldn't call sync_dev() for nul device (3092)
3692 - clear page cache after eviction (2766)
3693 - resynchronize MDS->OST in background (2824)
3694 - refuse to mount the same filesystem twice on same mountpoint (3394)
3695 - allow llmount to create routes for mounting behind routers (3320)
3696 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3697 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3698 - fix rare problem with rename on htree directories (3417)
3700 - allow default OST striping configuration per directory (1414)
3701 - fix compilation for qswnal for 2.6 kernels (3125)
3702 - increase maximum number of MDS request buffers for large systems
3703 - change liblustreapi to be useful for external progs like lfsck (3098)
3704 - increase local configuration timeout for slow disks (3353)
3705 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3707 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3710 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3711 - don't overwrite extent policy data in reply if lock was blocked (2901)
3712 - drop filter export grants atomically with removal from device (2663)
3713 - del obd_self_export from work_list in class_disconnect_exports (2908)
3714 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3715 - swab reply message in mdc_close, other PPC fixes (2464)
3716 - fix destroying of named logs (2325)
3717 - overwrite old logs when running lconf --write_conf (2264)
3718 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3719 - fix race in target_handle_connect (2898)
3720 - mds_reint_create() should take same inode create lock (2926)
3721 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3722 - don't close files for self_export to avoid uninitialized obd (2936)
3723 - allow MDS with the same name as client node (2939)
3724 - hold dentry reference for closed log files for unlink (2325)
3725 - reserve space for all logs during transactions (2059)
3726 - don't evict page beyond end of stripe extent (2925)
3727 - don't oops on a deleted current working directory (2399)
3728 - handle hard links to targets without a parent properly (2517)
3729 - don't dereference NULL lock when racing during eviction (2867)
3730 - don't grow lock extents when lots of conflicting locks (2919)
3732 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3735 - account for cache space usage on clients to avoid data loss (974)
3736 - lfsck support in lustre kernel code (2349)
3737 - reduce journal credits needed for BRW writes (2370)
3738 - orphan handling to avoid losing space on client/server crashes
3739 - ptlrpcd can be blocked, stopping ALL progress (2477)
3740 - use lock value blocks to assist in proper KMS, faster stat (1021)
3741 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3742 - recovery for initial connections (2355)
3743 - fixes for mds_cleanup_orphans (1934)
3744 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3745 - block all file creations until orphan recovery completes (1901)
3746 - client remove rq_connection from request struct (2423)
3747 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3748 - recovery timer race (2670)
3749 - mdc_close recovey bug (2532)
3750 - ptlrpc cleanup bug (2710)
3751 - mds timeout on local locks (2588)
3752 - namespace lock held during RPCs (2431)
3753 - handle interrupted sync write properly (2503)
3754 - don't try to handle a message that hasn't been replied to (2699)
3755 - client assert failure during cleanup after abort recovery (2701)
3756 - leak mdc device after failed mount (2712)
3757 - ptlrpc_check_set allows timedout requests to complete (2714)
3758 - wait for inflight reqs when ptlrpcd finishes (2710)
3759 - make sure unregistered services are removed from the srv_list
3760 - reset bulk XID's when resending them (caught by 1138 test)
3761 - unregister_bulk after timeout
3762 - fix lconf error (2694)
3763 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3764 - readahead locks pages, leaves pending causing memory pressure (2673)
3765 - increase OST request buffers to 4096 on large machines (2729)
3766 - fix up permission of existing directories in simple_mkdir (2661)
3767 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3768 - don't assign transno to errored transactions (2742)
3769 - don't delete objects on OST if given a bogus objid from MDS (2751)
3770 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3771 - if rq_no_resend, then timeout request after recovery (2432)
3772 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3773 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3774 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3775 - MDS recovery shouldn't delete objects that it has given out (2730)
3776 - if enqueue arrives after completion, don't clobber LVB (2819)
3777 - don't unlock pages twice when trigger_group_io returns error (2814)
3778 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3779 - don't write pages to disk if there was an error (1450)
3780 - don't ping imports that have recovery disabled (2676)
3781 - take buffered bytes into account when balancing socknal conn (2817)
3782 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3783 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3784 - remove little-endian swabbing of llog records (1987)
3785 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3786 - retry reposting request buffers if they fail (1191)
3787 - grow extent at grant time to avoid granting a revoked lock (2809)
3788 - lock revoke doesn't evict page if covered by a second lock (2765)
3789 - disable VM readahead to avoid reading outside lock extents (2805)
3791 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3792 - updated kernel patches for hp-2.4.20 kernel (2681)
3794 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3797 - fix truncated write corruption (2366)
3798 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3800 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3801 - ensure len(uuid) < 37 in lmc (1171)
3802 - fix ia64 OOPS in llog_test (2255)
3803 - zero end of page at obdfilter for partial page writes (2648)
3804 - don't leave stale dentries around after renames (bug 2428)
3805 - fix timeouts when evicting a client with a single lock held (2642)
3806 - set deadline for the initial HELLO message to drain (2634)
3807 - print out dotted-quad IP addresses in the socknal (2302)
3809 - additional debugging for MDS client eviction problem (2443)
3810 - fix mkfsoptions support for osts (2603, 2604)
3812 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3815 - add series for the vanilla 2.6.0 kernel
3816 - add series for the vanilla 2.4.24 kernel
3817 - add series for a cray x86/64 UL kernel drop
3818 - fix xattr patches for the vanilla 2.4.19 series
3820 - generate true UUIDs in lmc (1171)
3821 - have portals stack dumping break in UML (2466)
3822 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3823 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3824 - wake callback waiting threads on client eviction (2460)
3825 - Add --ptldebug and --subsystem to lmc (1719)
3826 - update assertion to allow safe interrupt allocation
3827 - set rq_no_resend for cancel requests (2432)
3828 - recalculate ptlrpcd timeout after resend (2494)
3829 - call vfs_rmdir when removing pending directories (2368)
3830 - fix renaming a file to itself (2429)
3831 - lmc creates a default one-stripe lov (2454)
3832 - expand procfs space to handle large clusters (2326)
3833 - increase UML stack to avoid overflow
3834 - update lconf's list of debug and subsystem masks
3835 - fix lfs find --obd (2510)
3836 - /proc tunable for disabling filter read caching (2591)
3837 - stop rpm packages from altering slapd.conf (2301)
3838 - disable nagle in the socknal under 0conf (2578)
3839 - choose mds inode size based on stripe count (2572)
3840 - fix kernel-source rpm problems (2516)
3842 - add --disable-doc to avoid pdf generation (2421)
3843 - update documentation, tests, type-os, comments
3844 - avoid format warnings on ia64
3845 - remove the TOE NAL
3846 - tiny code cleanups by removing unused fields
3848 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3851 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3852 - remove the most blatant lies from BUILDING file (2371)
3853 - change default debug level to reasonable production setting
3854 - reduce client side cache size to reduce cache flush time
3855 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3856 - make TCP zerocopy and pinger support enabled by default (2476)
3857 - sync writes completed after process exits caused crashes (2319)
3858 - maintain correct mount count on the MDS (2356)
3859 - backout 1557, because 2316 wasn't really fixed
3860 - better file I/O statistics gathering in /proc
3861 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3862 - another kernel patch to fix zero-copy TCP function export
3863 - don't take duplicate lock when processing re-sent getattr (2420)
3864 - lctl uses obd_self_export instead of creating new conn (2353)
3865 - MDS/OST recovery case which requires object creation asserted (2425)
3866 - move lfs from /usr/sbin to /usr/bin in packages
3867 - fix race between mds_client_add and mds_client_free (2417)
3868 - use kmalloc instead of slabs in portals (2430)
3869 - don't create duplicate records when a failover MDS is present (2442)
3870 - remove unnecessary mount age check (2332)
3871 - don't remove directory inodes from locks prematurely (2451)
3872 - don't break if MDS service name is the same as hostname (2103)
3873 - fix races in client write RPC generation when cache full (2482)
3875 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3878 - remove now-unused request->rq_obd (278)
3879 - if an allocation fails, print out how much memory we've used (1933)
3880 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3881 - add the "configurable stack size" patch to most series files (1256)
3882 - ability to write large log records, for 100+ OST configs (2306)
3883 - fix NULL deref when filter_prep fails (2314)
3884 - fix operator precedence error in filter_sync
3885 - dynamic allocation of socknal TX descriptors (2315)
3886 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3887 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3888 - ensure that CWARN messages in Portals always get to the syslog
3889 - __init/__exit are not for prototype decls (ldlm_init/exit)
3890 - x86-64 compile warning fixes
3891 - fix gateway LMC keyword conflict (2318)
3892 - fix MDS lock inversions in getattr/reint paths (1844)
3893 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3894 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3895 - DLM race condition prevented some lock evictions (2328)
3896 - ENOMEM detection and retry on socknal sends (2230)
3897 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3898 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3899 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3900 - better error messages when a client is rejected during recovery (1505)
3901 - avoid cancelling locks which were never granted, after failure (2330)
3902 - fix i_sem/journal inversion in mds_client_add (2333)
3903 - fix truncate/getattr lock cycle deadlock (2334)
3904 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3905 - fix two rare exit paths which could leak an l_lock() ref (2321)
3906 - fencepost error in MDS/OST orphan recovery (2226)
3907 - make log record alignment 8 bytes (1988)
3908 - lstripe now fails when requested offset > ost_count (2237)
3909 - ensure that all kernel series have a complete list.h (1607)
3910 - fix crashes in special-file operations (2316)
3911 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3913 - allow configurable automake binary, for testing new versions
3914 - small update to the lfs documentation
3916 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3918 * fix negative export reference count in fsfilt_sync (2312)
3920 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3921 * release candidate 0.9.1
3923 - orphans are moved into the PENDING directory for possible recovery
3924 - replayed opens now open by fid for orphan/rename safety (1042)
3925 - last close of an orphan inode generates a transno (683)
3926 - chdir() and mount() now pin the directory entry (1020)
3927 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3928 - discard very old requests without processing them (1502)
3929 - remove some common, well-understood CERRORs (1505)
3930 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3931 - clear "grant" flags in OST replies until OST grant code lands (1644)
3932 - fix read performance by not clobbering i_blksize on client (1598)
3933 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3934 - make LRU size a /proc tunable, clears locks when reduced (707)
3935 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3936 - server locks take references on exports now (1558)
3937 - build fixes for 2.4.20-rh trees (1663)
3938 - return an error from lov_create if all OSCs are inactive (1751)
3939 - fix import levels when a reconnect happens without a timeout (1597)
3940 - exit early from mds_open if we get a lookup error (1749)
3941 - partial page read at EOF wouldn't wait for disk before sending (1642)
3942 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3943 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3944 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3945 - write/truncate lock inversion (1639)
3946 - Don't auto-load obdclass, portals modules during cleanup (1495)
3947 - fix timestamps from jumping to "now" (1763)
3948 - extra journal assertions (1648)
3949 - add an extra multiunlink test (1771)
3950 - fix read_record/write_record API (1776)
3951 - fix leak of offset_extent, possible incorrect i_size later (1772)
3952 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3953 - Debugging for the kqswnal_get_idle_tx problems (1820)
3954 - Allow recovery to be attempted multiple times (1536)
3955 - Write out MDS last_rcvd file after it is first created (1600)
3956 - Fix tx_descriptor leak in failed transmit situations (1827)
3957 - ext3 journaling fixes for assertion failure after IO error (1871)
3958 - class_export_put() on freed export after completion AST error (1896)
3959 - Fix revalidate looping in VFS (1322)
3960 - Don't access a freed export during MDS_REINT timeout (1521)
3961 - Add open-unlink recovery support on the MDS (1673,1764)
3962 - Return an error if no MDS data was read from last_rcvd (1946)
3963 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3964 - Don't setup a disk device that doesn't match exported UUID (317)
3965 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3966 - avoid committing NULL handle in force close
3967 - local.sh is now a one-stripe LOV configuration
3968 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3969 - fix ext3 htree duplicate directory entry corruption (1516)
3970 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3971 - update to new LOV EA format (2097)
3972 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3974 2003-06-15 Phil Schwan <phil@clusterfs.com>
3977 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3978 - per-import/export recovery handling (958, 931, 959)
3979 - multiple last-rcvd slots, for serving multiple FSes (949)
3980 - connections are again shared between multiple imp/exports (963, 964)
3981 - "umount -f" would hang if any requests needed to be sent (393, 978)
3982 - avoid pinning large req buffer by copying for queued messages (989)
3983 - add "uuid" to "lctl device" command to help upcalls (991)
3984 - "open" RPCs with transnos would confuse recovery counters (1037)
3985 - do proper endian conversion of all wire messages (288, 340, 891)
3986 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3987 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3988 - fix ptlrpc_connection leak in target_handle_connect (1174)
3989 - fix import refcounting bug in OST and MDS cleanup (1134)
3990 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3991 - fix very unlikely obd_types race condition (501)
3992 - remove osc_open hack for echo_client (1187)
3993 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3994 - a failure in read_inode2 leads to deadlock (1139)
3995 - cancel ack-locks as soon as transaction is committed (1072)
3996 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3997 - make sure to commitrw after any preprw to avoid deadlock (1162)
3998 - failing to execute a file in a lustre FS would lock inode (1203)
3999 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4000 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4001 - obd_uuid2tgt would walk off the end of the list (1255)
4002 - on IA64 the osc would give portals incorrect bulk size (1258)
4003 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4004 - fix lock inversion caused by new llite matching code (1282)
4005 - limit the number of dirty pages on a client to 10MB (1286)
4006 - timed out locks were not being corrected cancelled (1289)
4007 - fix O_DIRECT above 4GB on IA-32 (1292)
4008 * major user-visible changes
4009 - fail out/fail over policy now controlled by the upcall (993)
4011 - add OBD_PING to check server availability and failure (954)
4012 - lustre messages are now sent in sending host order (288, 340, 891)
4013 - add eadatalen to MDS getattr reply (340)
4014 - OST read replies may contain second buffer, with per-page status (593)
4016 2003-03-11 Phil Schwan <phil@clusterfs.com>
4019 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4020 - failed open()s could cause deadlock; fixed (867, 869)
4021 - stop cancelling OST locks when files are closed (481)
4022 - overlapping XID spaces caused network corruption (851, 853)
4023 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4024 - setattr_raw added, to do single-RPC, server-side setattrs
4025 - lmc/lconf syntax change for OST UUIDs
4026 - fix crashy race condition between ptlrpc_free_req and osc_close
4027 - don't use request in mdc_enqueue if we hit a timeout (889)
4028 - don't set the inode i_size for regular files from the MDS (896)
4029 - handle out of order completion AST (842)
4030 - don't LBUG if a lock request times out after receiving AST (913)
4031 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4032 - if a bad lock AST arrives, send an error instead of dropping entirely
4033 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4034 - fix leak in bulk IO when only partially completed (899, 900, 926)
4035 - fix O_DIRECT for ia64 (55)
4036 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4037 - C-z after timeout could hang a process forever; fixed (977)
4039 - client-side I/O cache (678, 924, 929, 941, 970)
4041 - READPAGE and SETATTRs which don't take server-side locks get
4044 2003-02-11 Phil Schwan <phil@clusterfs.com>
4047 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4048 - Fix fsfilt_extN_readpage() to read a full page of directory
4049 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4050 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4051 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4052 - Most intent operations were removed, in favour of a new RPC mode
4053 that does a single RPC to the server and bypasses most of the VFS
4054 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4055 - Aggressively cancel local locks on DLM servers
4056 - mds_reint_unlink sends EA to the client if it's the last nlink.
4057 client uses that EA to unlink OST objects.
4058 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4059 - recursive symlinks were fixed (439)
4060 - fixed NULL deref in DEBUG_REQ
4061 - filter_update_lastobjid no longer calls sync, which annoyed extN
4062 - fixed multi-client small-writes to a single file problem (445)
4063 - fixed mtime updates during file writes (607)
4064 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4065 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4066 - fixed Linux OST size reporting problem (444, 656)
4067 - OST now updates object mtime with writes or setattr (607, 619)
4068 - client verifies file size before zeroing page past EOF (445)
4069 - OST now writes last allocated objid to disk with allocation (108)
4070 - LOV on echo now works (409)
4072 - mds_reint_unlink sends a new buffer, with the EA included. this
4073 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4074 set if a regular file was being unlinked, and it was the last link
4075 - use PtlGet from the target for bulk writes (315)
4076 - OST now updates object mtime with writes or setattr (607, 619)
4077 - LDLM now has a grant-time callback to revalidate locked items, if
4079 - Many MDS operations were reorganized to combat race conditions
4081 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4082 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4084 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4087 - Fully reactivate OST imports after reconnection (512, others)
4088 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4089 - More graceful error handling for truncating on dead OST (515)
4090 - Don't error out unless we're actually accessing dead stripes (474)
4091 - Fix garbage sizes when stripes are missing (410)
4092 - LRU counters were broken, causing constant lock purge (433, 432)
4093 - garbage on read from stripes with failed OSTs (441)
4094 - mark OSCs as active before reconnecting during recovery (438)
4095 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4096 - lfind did not preserve OST order in output (443)
4097 - symlinks cause hung clients, incorrect data (439)
4098 - stop dereferencing request after dropping refcount (457)
4099 - don't LASSERT(spin_is_locked) on non-SMP (455)
4100 - fixes for many rename() bugs
4101 - fstat didn't correctly synchronize attributes (399)
4102 - server must handle lock cancellation during blocking AST prep (487)
4103 - bulk descriptors were free()d too soon (511)
4104 - fix paths in lconf, which would load incorrect modules (451, 507)
4105 - fix confusing lconf 'host not found' error message (386)
4106 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4107 - fix race condition in mdc_blocking_ast() for inode access (526)
4108 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4109 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4110 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4111 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4112 - fix lstripe in several ways (kernel side) (527)
4113 - fix request leak in ldlm_cli_enqueue (262)
4114 - incorrect OSC was marked inactive after OST failure
4115 - call mds_fs_cleanup before unmounting filesystem (524)
4116 - fix races between taking ns_lock and ldlm_lock_change_resource
4117 - fix races updating LOV export open file list
4118 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4119 - fix recovery NULL deref in ldlm_cli_cancel_unused
4120 - fix some DLM races by using new hash table for lock handles (419)
4121 - permit the client to specify desired inodes, at replay
4122 - duplicate requests when we queue them for replay reintegration
4123 - fix last_rcvd offset calculation
4124 - sync after each recovered transaction, so we always make progress
4125 - never, not always, ERESTART requests without transnos
4126 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4128 - skip replay if the MDS says that the client is already connected
4129 - don't check for a recovery-enabled export to match lctl's UUID
4130 - don't INC_USE_COUNT for phantom exports
4131 - don't crash when cleaning up phantom exports (567)
4132 - don't double-finish or set replay data for errored mdc_open requests
4133 - abort requests when they time out, so we don't get old replies
4134 - send/receive replies for AST messages again
4135 - if the client says that it doesn't have the lock, cancel it on the
4137 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4138 mark it as destroyed, it will all work out when decref is called
4139 - fix module use counts (22, 581)
4141 - ASTs now expect a reply (server cancels lock on error reply)
4143 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4146 - fix many simultaneous client startup (392)
4147 - fix dentry->d_it clobbering
4148 - credentials weren't being shipped for readdir/getattr operations
4149 - remove invalid assertions triggered during some concurrent MD
4151 - proper Lustre versions added (336, 389)
4152 - fix memory leak for create error case (398)
4153 - fix LOV locking bug that would get cli/srv out of sync
4154 - fix echo client over LOV (409)
4155 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4156 - fix double-O_EXCL intent crash (424)
4157 - avoid sending multiple lock CANCELs (352)
4159 - MDS can do multi-client recovery (modulo bugs in new code)
4161 - many updates, edits, cleanups
4163 2002-11-18 Phil Schwan <phil@clusterfs.com>
4166 - fix null d_it dereference (346)
4167 - fix full OST/dbench hang (333)
4168 - fix permission problem with file removal (286)
4169 - fix removal of OSCs from LOV when they fail
4170 - fix NULL deref during bulk timeout (214)
4171 - fix problems related to multiple filesystems on one MDS (241)
4172 - fixed serious subtle metadata locking bugs
4173 - free locks on clients when inodes are removed due to memory
4175 - fix inode pointer in lock data (285)
4176 - partial support for multiple MDS on a single host (241)
4177 - data locks weren't cancelled at clear_inode time (290, 311)
4178 - intent locks could lead to unbounded lock growth (205)
4179 - added a maximum lock count, an LRU list, and a flusher
4180 - fix multiple rename (365)
4181 - properly abstracted the echo client
4182 - OSC locked 1 byte too many; fixed
4183 - rewrote brw callback code:
4184 - fixed recovery bugs related to LOVs (306)
4185 - fixed too-many-pages-in-one-write crash (191)
4186 - fixed (again) crash in sync_io_timeout (214)
4187 - probably fixed callback-related race (385)
4189 - Add capability to MDS protocol
4190 - LDLM cancellations and callbacks on different portals
4192 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4195 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4196 - defer open object destruction to close time (601981, 138)
4197 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4198 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4199 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4200 - don't call obd_setattr in ll_file_release for destroyed objects
4201 * protocol change to lustre_msg: move |version| and add |flags|
4202 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4203 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4204 * added replay of create, unlink, link and rename operations during
4205 MDS failover; recovery should be much more robust now
4206 * remove failed OSCs from LOVs (only lov_create uses this so far)
4207 * the lustre-HOWTO was brought (more) up to date (582544)
4209 2002-10-23 Phil Schwan <phil@clusterfs.com>
4212 - in-use dentries weren't being reused properly (617851)
4213 - prevent multiple LDLM setup (599178)
4214 - fix LOV size calculations for truncate (617853)
4215 - fix client handling of MDS intent errors (POSIX)
4216 - fix permission bug in lovstripe.c test (624321)
4217 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4218 - truncate past end of file could corrupt data
4219 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4220 - a race in recovery could return ETIMEDOUT to apps (623947)
4221 - building outside the source directory was fixed
4222 * the lustre-HOWTO was brought (more) up to date (582544)
4223 * major progress was made on recovery functionality
4225 2002-10-10 Phil Schwan <phil@clusterfs.com>
4228 - recovery deadlock fix
4229 - rm -rf causes LBUG fix (617817)
4230 - file open by multiple tasks fix (618962)
4231 - directory permissions bugs (602707 and 620007)
4232 - journal_stop fixed with locking (611313)
4233 - O_APPEND failures resolved (618273, perhaps 614459)
4234 - lconf PATH fix (619770)
4235 - IA64 build fix (621450)
4236 - RPC buffer sizes scale with amount of memory
4238 2002-10-01 Phil Schwan <phil@clusterfs.com>
4241 - locks would be cancelled without throwing away data pages,
4242 resulting in inconsistent data (605627)
4243 - inode attributes were not always being refreshed (605627, 612449)
4244 - lconf now continues to cleanup after lctl reports an error
4245 - MDS now enforces user permissions (602707)
4246 - lprocfs cleanup fixed, but not yet enabled (614157)
4247 - fixed infinite server hang, should a client not respond to an AST
4248 - avoid going into recovery if user calls readlink() with a buffer
4249 that's too small (613941)
4250 - AST RPCs no longer require replies (614867) -- this may be changed
4251 - don't crash server if client sends an IOV that's too big (611336)
4252 - fixed lock conversion deadlock (611892)
4253 - fixed the following of symlinks (614622)
4254 * recovery: the server can remove locks from a client that dies, other
4255 clients can make progress
4256 * more extN patch fixes
4257 * compile-time configurable ptlrpc buffer allocations
4259 - collaborative read cache document
4260 - Lustre Lite Performance CDR document-in-progress
4262 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4265 - fix typo in patch-2.4.18
4267 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4270 - clear ptlrpc request each time in handle_incoming_request()
4271 - unlink of files now destroys the object on the OST
4273 2002-09-19 Peter Braam <braam@clusterfs.com>
4275 * add hard link support
4276 * change obdfile creation method
4277 * kernel patch changed
4279 2002-09-19 Peter Braam <braam@clusterfs.com>
4282 - stack overflow bug in extN fixed
4284 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4286 * documentation updates
4287 - add man pages for config tools
4288 - update tests/README to describe testing with new config tools
4289 - finish metadata API descriptions
4290 * bug fixes and cleanups
4291 - statfs workaround for 16TB limit
4292 - LOV stripe allocation improved, can stripe on subset of OSTs
4293 - LOV file size/IO offset was wrong for files > 4GB in size
4294 - object EA data was being dropped, caused files to be unreadable
4295 - memory overflow with non-LOV OST caused memory corruption
4296 - fixed regression tests to work with new config tools, obdfilter
4297 - fixed bug when directory size became larger than 1 block
4298 - fixed bug (for single client case) when PWD was deleted
4299 - invalidate local directory pages when doing intent-based ops
4300 - avoid LDLM oops when lock callback contained bad data
4302 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4304 * documentation updates
4305 * bug fixes and cleanups
4306 - configuration tools
4309 - 64-bit compile warnings
4310 - 64-bit internal statfs data
4312 * test_brw on persistent OST devices
4314 * lprocfs (disabled)
4316 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4318 * documentation updates
4319 * bug fixes and cleanups
4320 * configuration tools
4322 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4325 * many small fixes to 0_5_4
4326 * io/network handling
4327 * thinkos in MDS operations
4329 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4332 * crucial basic fixes to 0.5.3
4333 * IOR, Iozone work over Elan
4336 2002-08-07 Phil Schwan <phil@clusterfs.com>
4337 * version 0_5_3, our first alpha
4338 * we use the new Portals iovs
4339 * documentation updates
4340 * bug fixes and cleanups
4341 * small changes in the DLM wire protocol
4343 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4344 * version 0_5_1 with some initial stability,
4345 * locking on MD and file I/O.
4346 * documentation updates
4347 * several bug fixes since 0.5.0
4348 * small changes in wire protocol
4350 2002-07-18 Phil Schwan <phil@clusterfs.com>
4352 * delivered as Lustre Light Alpha
4353 * fixed a crash after handling invalid MDS requests
4354 * fixed directory pages for architectures with non-4k pages sizes
4356 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4357 * release version v0_4_4
4358 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4360 * Fixes a number of interruption problems with OST operations.
4361 * Update documentation for portals header changes
4362 * Move all wire protocol structs/defines to lustre_idl.h
4363 * Fixes symlink length bug.
4364 * Add tcpdump to repository.
4366 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4367 * release version v0_4_3
4368 * Fixes statfs for inodes on extN.
4369 * Fixes bug in runtests which would delete /etc/hosts.
4370 * Use 64-bit object IDs wherever possible (not into VFS though)
4371 Remove ost_get_info, which is unused by lustre, and out of date.
4373 2002-07-03 Peter Braam <braam@clusterfs.com>
4374 * release version v0_4_2 Fixes a lookup error (type not passed)
4375 * move forward to head of Portals
4376 * move forward to latest Lustre kernel
4378 2002-06-25 Peter Braam <braam@clusterfs.com>
4379 * release version v0_4_1. Hopefully stable on single node use.