1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6 2.6.16.46-0.14 (SLES 10)
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.16 - 2.6.22 vanilla (kernel.org)
11 * Due to recently discovered recovery problems, we do not recommend
12 using patchless RHEL 4 clients with this or any earlier release.
13 * Recommended e2fsprogs version: 1.39.cfs8
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 * Starting with this release, the ldiskfs backing filesystem required
16 by Lustre is now in its own package, lustre-ldiskfs. This package
17 should be installed. It is versioned separately from Lustre and
18 may be released separately in future.
21 Frequency : only for SLES9
23 Description: Routinely utilize latest Quadrics drivers in CFS releases
24 Details : Update patch qsnet-suse-2.6.patch.
27 Frequency : only for sles10
29 Description: Update patches for SLES 10 SP1 kernel.
30 Details : Update the patch vfs_intent-2.6-sles10.patch.
35 Description: Routinely utilize latest Quadrics drivers in CFS releases
36 Details : Update patch qsnet-rhel4-2.6.patch.
39 Frequency : at statup only
41 Description: mds_lov_synchronize race leads to various problems
42 Details : simultaneous MDT->OST connections at startup can cause the
43 sync to abort, leaving the OSC in a bad state.
45 Severity : enhancement
47 Description: add optional extra BUILD_VERSION info
48 Details : add a new environment variable (namely LUSTRE_VERS) which allows
49 to override the lustre version.
52 Frequency : 2.6.18 servers only
54 Description: ll_kern_mount() doesn't release the module reference
55 Details : The ldiskfs module reference count never drops down to 0
56 because ll_kern_mount() doesn't release the module reference.
61 Description: server LBUG when using old ost_num_threads parameter
62 Details : Accept the old ost_num_threads parameter but warn that it
63 is deprecated, and fix an off-by-one error that caused an LBUG.
68 Description: Transient SCSI error results in persistent IO issue
69 Details : iobuf->dr_error is not reinitialized to 0 between two
73 Frequency : sometimes when underlying device returns I/O errors
75 Description: OSTs not going read-only during write failures
76 Details : OSTs are not remounted read-only when the journal commit threads
77 get I/O errors because fsfilt_ext3 calls journal_start/stop()
78 instead of the ext3 wrappers.
82 Description: poor connect scaling with increasing client count
83 Details : Don't run filter_grant_sanity_check for more than 100 exports
84 to improve scaling for large numbers of clients.
87 Frequency : SLES10 only
89 Description: sanity-quota.sh quotacheck failed: rc = -22
90 Details : Quotas cannot be enabled on SLES10.
93 Frequency : liblustre clients only
95 Description: getdirentries does not give error when run on compute nodes
96 Details : getdirentries does not fail when the size specified as an argument
97 is too small to contain at least one entry
99 Severity : enhancement
101 Description: Add LNET router traceability for debug purposes
102 Details : If a checksum failure occurs with a router as part of the
103 IO path, the NID of the last router that forwarded the bulk data
104 is printed so it can be identified.
109 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
110 Details : Due to a race condition, liblustre clients were occasionally
113 Severity : enhancement
115 Description: lfs setstripe use optional parameters instead of postional
118 Severity : enhancement
120 Description: Nanosecond timestamp support for ldiskfs
121 Details : The on-disk ldiskfs filesystem has added support for nanosecond
122 resolution timestamps. There is not yet support for this at
123 the Lustre filesystem level.
126 Frequency : during server recovery
128 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
129 Details : request with rq_no_resend flag not awake l_wait_event if they get a
133 Frequency : nfs export on patchless client
135 Description: connectathon hang when test nfs export over patchless client
136 Details : Disconnected dentry cannot be found with lookup, so we do not need
137 to unhash it or make it invalid
140 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
141 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
142 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
147 Description: open req refcounting wrong on reconnect
148 Details : If reconnect happened between getting open reply from server and
149 call to mdc_set_replay_data in ll_file_open, we will schedule
150 replay for unreferenced request that we are about to free.
151 Subsequent close will crash in variety of ways.
152 Check that request is still eligible for replay in
153 mdc_set_replay_data().
158 Description: disable writes to filesystem when reading health_check file
159 Details : the default for reading the health_check proc file has changed
160 to NOT do a journal transaction and write to disk, because this
161 can cause reads of the /proc file to hang and block HA state
162 checking on a healthy but otherwise heavily loaded system. It
163 is possible to return to the previous behaviour during configure
164 with --enable-health-write.
166 Severity : enhancement
168 Description: 64-bit inode version
169 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
170 to the inode. This will be required for version-based recovery.
175 Description: MDS fails to start if a duplicate client export is detected
176 Details : in some rare cases it was possible for a client to connect to
177 an MDS multiple times. Upon recovery the MDS would detect this
178 and fail during startup. Handle this more gracefully.
180 Severity : enhancement
182 Description: Add -o localflock option to simulate old noflock
184 Details : This will achieve local-only flock/fcntl locks
190 Description: log_commit_thread vs filter_destroy race leads to crash
191 Details : Take import reference before releasing llog record semaphore
196 Description: Wrong request locking in request set processing
197 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
198 imp_delayed_list, in this place should be used imp_lock.
201 Frequency : when reconnection
203 Description: Grant Leak when osc reconnect to OST
204 Details : When osc reconnect ost, OST(filter) should check whether it
205 should grant more space to client by comparing fed_grant and
206 cl_avail_grant, and return the granted space to client instead
207 of "new granted" space, because client will call osc_init_grant
208 to update the client grant space info.
211 Frequency : when client reconnect to OST
213 Description: Grant Leak when osc do resend and replay bulk write
214 Details : When osc reconnect to OST, OST(filter)should clear grant info of
215 bulk write request, because the grant info while be sync between
216 OSC and OST when reconnect, and we should ignore the grant info
217 these of resend/replay write req.
222 Description: Grant space more than avaiable left space sometimes.
223 Details : When then OST is about to be full, if two bulk writing from
224 different clients came to OST. Accord the avaliable space of the
225 OST, the first req should be permitted, and the second one
226 should be denied by ENOSPC. But if the seconde arrived before
227 the first one is commited. The OST might wrongly permit second
228 writing, which will cause grant space > avaiable space.
231 Frequency : when client is evicted
233 Description: Grant might be wrongly erased when osc is evicted by OST
234 Details : when the import is evicted by server, it will fork another
235 thread ptlrpc_invalidate_import_thread to invalidate the
236 import, where the grant will be set to 0. While the original
237 thread will update the grant it got when connecting. So if
238 the former happened latter, the grant will be wrongly errased
239 because of this race.
244 Description: Checking Stale with correct fid
245 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
246 to check whether it is stale, because sometimes, we want the
247 enqueue happened anyway, and op_data.fid2 will not be initialized.
249 Severity : enhancement
251 Description: update patchless client
252 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
255 Frequency : only with 2.4 kernel
257 Description: random memory corruption
258 Details : size of struct ll_inode_info is to big for union inode.u and this
259 can be cause of random memory corruption.
264 Description: Memory leak in recovery
265 Details : Lov_mds_md was not free in an error handler in mds_create_object.
266 It should also check obd_fail before fsfilt_start, otherwise if
267 fsfilt_start return -EROFS,(failover mds during mds recovery).
268 then the req will return with repmsg->transno = 0 and rc = EROFS.
269 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
270 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
271 be freed no matter whether fsfilt_commit success or not.
274 Frequency : only with huge count clients
276 Description: Prevents from taking the superblock lock in llap_from_page for
278 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
279 for llap_from_page call in ll_removepage prevents from taking the
280 superblock lock for a soon died page.
285 Description: Not check open intent error before release open handle
286 Details : in some rare cases, the open intent error is not checked before
287 release open handle, which may cause
288 ASSERTION(open_req->rq_transno != 0), because it tries to release
289 the failed open handle.
294 Description: Set cat log bitmap only after create log success.
295 Details : in some rare cases, the cat log bitmap is set too early. and it
296 should be set only after create log success.
300 Description: Accessing a block bevice can re-enable I/O when Lustre is
301 tearing down a device.
302 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
307 Description: service threads may hog cpus when there are a lot of requests
308 Details : Insert cond_resched to give other threads a chance to use some CPU
313 Description: the cat log was not initialized in recovery
314 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
315 unlink log on mds will not be initialized until mds post
316 recovery. And also in mds post recovery, the unlink log will
317 initialization will be done asynchronausly, so there will be race
318 between add unlink log and unlink log initialization.
322 Description: brw_stats were being printed incorrectly
323 Details : brw_stats were being printed as log2 but all of them were not
324 recorded as log2. Also remove some code duplication arising from
325 filter_tally_{read,write}.
329 Frequency : rare, only in recovery.
330 Description: ASSERTION(req->rq_type != LI_POISON) failed
331 Details : imp_lock should be held while iterating over imp_sending_list for
332 prevent destroy request after get timeout in ptlrpc_queue_wait.
336 Description: replay-single.sh test 52 fails
337 Details : A lock's skiplist need to be cleanup when it being unlinked
338 from its resource list.
343 Description: Short directio read returns full requested size rather than
345 Details : Direct I/O operations should return actual amount of bytes
346 transferred rather than requested size.
348 Severity : enhancement
350 Description: metadata RPC reduction (e.g. for rm performance)
351 Details : decrease the amount of synchronous RPC between clients and servers
352 by canceling conflicing lock before the operation on the client
353 and packing thier handles into the main operation RPC to server.
355 Severity : enhancement
357 Description: add #ifdef HAVE_KERNEL_CONFIG_H
358 Details : kernels from 2.6.19 not need include linux/config.h, but add
359 include linux/autoconf.h in commpiler command line.
361 Severity : enhancement
363 Description: patchless client support for 2.6.22 kernel
364 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
365 removed. In this case we need drop using os depended interface to
366 kmem_cache and use cfs_mem_cache API.
371 Description: fix mal-formatted messages
372 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
376 Frequency : always in liblustre
377 Description: wrong IS_ERR implementation in liblustre.h
378 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
383 Description: Correct condition for output debug message.
384 Details : inode i_nlink equal zero is not enough for output message about
385 disk corruption, i_ctime and i_mode should be also checked.
389 Frequency : always in patchless client
390 Description: add configure check for truncate_complete_page
391 Details : improve checks for exported symbols. This allow run check without
392 sources, but with Module.symvers shipped with kernel distribution.
393 add check for truncate_complete_page used by patchless client.
398 Description: improve handling recoverable errors
399 Details : if request processig with error which can be recoverable on server
400 request should be resend, otherwise page released from cache and
405 Description: sanity.sh test_77h fails with "test_77h file compare failed"
406 Details : test_77h uses a file which was messed by other test case.
410 Description: Not Check whether lov_tgts is NULL in some lov functions
411 Details : Checking whether lov_tgts is NULL in some functions.
415 Description: replace obdo_alloc() with OBDO_ALLOC macro
416 Details : nothing special is done in obdo_alloc() function, and for
417 debugging purpose, it needs to be replaced with macros.
421 Description: bad return value and errno from fcntl call
422 Details : In liblustre API, errno should be a negative value if error
427 Description: ptlrpc_check_set() LBUG
428 Details : In case of positive reply from server and failed client bulk
429 callback after bulk transfer shouldn't LBUG, but process this
430 request as erroneous.
432 Severity : enhancement
434 Description: VFS operations stats tool.
435 Details : Tool which collects stats by tracking value written in pid,
436 ppid, gid and uses llstat to generate output to plot graph using
438 Updated lustre/utils/Makefile.am
439 Added lustre/utils/ltrack_stats.c
441 Severity : enhancement
443 Description: 2.6.18 server support (lustre 1.6.1)
444 Details : Support for 2.6.18 kernels on the server side.
449 Description: ASSERTION(imp->imp_conn_current) failed
450 Details : an assertion failure is hit if a client node boots and attempts to
451 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
454 Frequency : only for i686
456 Description: 1.4.11 RC1 build fails for RHEL 4, i686
457 Details : Fixed config variable for build.
462 Description: Updated patchess for new RHEL4 kernel
463 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
464 Updated patch jbd-stats-2.6.9.patch
465 Updated patch qsnet-rhel4-2.6.patch
466 Updated patch quota-deadlock-on-pagelock-core.patch
467 Updated patch vfs_intent-2.6-rhel4.patch
468 Updated patch vfs_races-2.6-rhel4.patch
469 Updated series file 2.6-rhel4-titech.series
470 Updated series file 2.6-rhel4.series
471 Updated kernel config files
476 Description: lquota slave complains LBUG when reconnecting with mds
478 Details : quota slave depends qctxt->lqc_import to send its quota request.
479 This pointer will be invalid if mds did failover or broke its
480 connect to osts, which leads to LBUG.
483 Frequency : when qunit size is too small(less than 20M)
485 Description: write is stopped by improper -EDQUOT
486 Details : If the master is busy and qunit size is small enough(let's say 1M),
487 the slave can not get quota from master on time, which will lead
488 slave to trigger a -EQUOTA to client.
493 Description: Deadlock during metadata tests
494 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
495 called with the per-dentry lock held.
498 Frequency : SLES9 only
500 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
501 Details : set CONFIG_NR_CPUS to 128 instead of 8.
503 Severity : enhancement
505 Description: remove fs_prep_san_write operation and related patches
506 Details : remove the ext3-san-jdike patches which are no longer useful.
511 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
512 Details : a disk corruption can cause the mballoc code to assert on a
513 double free or other extent corruptions. Handle these with
514 ext3_error() instead of assertions.
517 Frequency : only with mballoc3 code and deep extent trees
519 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
520 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
521 can cause the filesystem to be remounted read-only.
526 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
527 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
528 few external users of this ioctl number is lfsck's call to
529 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
530 Add the old ioctl number to the handler so both old and new
535 Description: parallel lock callbacks
536 Details : Instead of sending blocking and completion callbacks as separated
537 requests, adding them to a set and sending in parallel.
541 Description: Disable most debugging by default
542 Details : To improve performance, disable most logging (for debug purposes)
543 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
544 default, and HA includes fewer messages.
548 Frequency : only run on patchless client.
549 Description: use do_facet on sanity.sh for test handling recoverables errors
550 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
554 Description: hash tables for lists of nids, connections and uuids
555 Details : Hash tables noticeably help when a lot of clients connect to a
556 server, to faster identify duplicate connections or reconnects,
557 also to faster find export to evict in manual eviction case.
561 Description: Sometimes, when the server evict a client, and the client will
562 not be evicted as soon as possible.
563 Details : In enqueue req, the error was returned by intent, instead of
564 rq_status which make ptlrpc layer not detect this error, and
565 does not evict the client. So enqueue error should be returned
571 Description: Using wrong pointer in osc_brw_prep_request
572 Details : Access to array[-1] can produce panic if kernel compiled with
573 CONFIG_PAGE_ALLOC enabled
575 --------------------------------------------------------------------------------
577 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
582 Frequency : on some architectures
584 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
585 Details : Uninitialized flags sometimes cause configuration commands to
589 Frequency : patchless clients only
591 Description: missing __iget() symbol export
592 Details : The __iget() symbol export is missing. To avoid the need for
593 this on patchless clients the deathrow inode reaper is turned
594 off, and we depend on the VM to clean up old inodes. This
595 dependency was during via the fix for bug 12181.
600 Description: sanity.sh fail: test_52b
601 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
602 incorrect inode's flags to client.
604 --------------------------------------------------------------------------------
606 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
608 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
609 INTEROPERATE with older versions automatically. In many cases a
610 special upgrade step is needed. Please read the
611 user documentation before upgrading any part of a 1.4.x system.
612 * WARNING: Lustre configuration and startup changes are required with
613 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
615 * Support for kernels:
616 2.4.21-47.0.1.EL (RHEL 3)
618 2.6.9-42.0.10.EL (RHEL 4)
619 2.6.12.6 vanilla (kernel.org)
620 2.6.16.27-0.9 (SLES10)
621 * Client support for unpatched kernels:
622 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
623 2.6.16 - 2.6.19 vanilla (kernel.org)
624 2.6.9-42.0.8EL (RHEL 4)
625 * Recommended e2fsprogs version: 1.39.cfs6
626 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
629 Severity : enhancement
631 Description: Async OSC create to avoid the blocking unnecessarily.
632 Details : If a OST has no remain object, system will block on the creating
633 when need to create a new object on this OST. Now, ways use
634 pre-created objects when available, instead of blocking on an
635 empty osc while others are not empty. If we must block, we block
636 for the shortest possible period of time.
638 Severity : enhancement
640 Description: MountConf
641 Details : Lustre configuration is now managed via mkfs and mount
642 commands instead of lmc and lconf. New obd types (MGS, MGC)
643 are added for dynamic configuration management. See
644 https://mail.clusterfs.com/wikis/lustre/MountConf for
647 Severity : enhancement
649 Description: dynamic OST addition
650 Details : OSTs can now be added to a live filesystem
652 Severity : enhancement
654 Description: startup order invariance
655 Details : MDTs and OSTs can be started in any order. Clients only
656 require the MDT to complete startup.
658 Severity : enhancement
660 Description: parallel, asynchronous orphan cleanup
661 Details : orphan cleanup is now performed in separate threads for each
662 OST, allowing parallel non-blocking operation.
664 Severity : enhancement
666 Description: optimized stripe assignment
667 Details : stripe assignments are now made based on ost space available,
668 ost previous usage, and OSS previous usage, in order to try
669 to optimize storage space and networking resources.
671 Severity : enhancement
673 Description: Permanently set tunables
674 Details : All writable /proc/fs/lustre tunables can now be permanently
675 set on a per-server basis, at mkfs time or on a live
678 Severity : enhancement
680 Description: Lustre message v2
681 Details : Add lustre message format v2.
683 Severity : enhancement
685 Description: client OST exclusion list
686 Details : Clients can be started with a list of OSTs that should be
687 declared "inactive" for known non-responsive OSTs.
691 Description: ENOENT returned for valid filehandle during dbench.
692 Details : Check if a directory has children when invalidating dentries
693 associated with an inode during lock cancellation. This fixes
694 an incorrect ENOENT sometimes seen for valid filehandles during
698 Frequency : SFS test only (otherwise harmless)
700 Description: SPEC SFS validation failure on NFS v2 over lustre.
701 Details : Changes the blocksize for regular files to be 2x RPC size,
702 and not depend on stripe size.
704 Severity : enhancement
706 Description: fine-grained SMP locking inside DLM
707 Details : Improve DLM performance on SMP systems by removing the single
708 per-namespace lock and replace it with per-resource locks.
710 Severity : enhancement
712 Description: don't hold multiple extent locks at one time
713 Details : To avoid client eviction during large writes, locks are not
714 held on multiple stripes at one time or for very large writes.
715 Otherwise, clients can block waiting for a lock on a failed OST
716 while holding locks on other OSTs and be evicted.
718 Severity : enhancement
720 Description: Multiple MD RPCs in flight.
721 Details : Further unserialise some read-only MDT RPCs - learn about intents.
722 To avoid overly-overloading MDT, introduce a limit on number of
723 MDT RPCs in flight for a single client and add /proc controls
724 to adjust this limit.
726 Severity : enhancement
728 Description: client read/write statistics
729 Details : Add client read/write call usage stats for performance
730 analysis of user processes.
731 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
732 file access. extents_stats shows chunk size distribution.
733 extents_stats_per_process show chunk size distribution per
736 Severity : enhancement
738 Description: per-client statistics on server
739 Details : Add ldlm and operations statistics for each client in
740 /proc/fs/lustre/mds|obdfilter/*/exports/
742 Severity : enhancement
744 Description: improved MDT statistics
745 Details : Add detailed MDT operations statistics in
746 /proc/fs/lustre/mds/*/stats
748 Severity : enhancement
750 Description: VFS operations stats
751 Details : Add client VFS call stats, trackable by pid, ppid, or gid
752 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
757 Description: Fix client-side osc byte counters
758 Details : The osc read/write byte counters in
759 /proc/fs/lustre/osc/*/stats are now working
762 Frequency : always as root on SLES
764 Description: Failure of copying files with lustre special EAs.
765 Details : Client side always return success for setxattr call for lustre
766 special xattr (currently only "trusted.lov").
771 Description: Refcount LNET uuids
772 Details : The global LNET uuid list grew linearly with every startup;
773 refcount repeated list entries instead of always adding to
776 Severity : enhancement
778 Description: Dynamic service threads
779 Details : Within a small range, start extra service threads
780 automatically when the request queue builds up.
783 Frequency : mixed-endian client/server environments
785 Description: mixed-endian crashes
786 Details : The new msg_v2 system had some failures in mixed-endian
789 Severity : enhancement
791 Description: Easy OST removal
792 Details : OSTs can be permanently deactivated with e.g. 'lctl
793 conf_param lustre-OST0001.osc.active=0'
795 Severity : enhancement
797 Description: MGS proc entries
798 Details : Added basic proc entries for the MGS showing what filesystems
801 Severity : enhancement
803 Description: provide MGS failover
804 Details : Added config lock reacquisition after MGS server failover.
806 Severity : enhancement
808 Description: add Linux 2.4 support
809 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
813 Description: a large application tries to do I/O to the same resource and dies
815 Details : Check the req->rq_arrival time after the call to
816 ost_brw_lock_get(), but before we do anything about
817 processing it & sending the BULK transfer request. This
818 should help move old stale pending locks off the queue as
819 quickly as obd_timeout.
822 Frequency : when an incorrect nid is specified during startup
824 Description: ptlrpc connect to non-existant node causes kernel crash
825 Details : LNET can't be re-entered from an event callback, which
826 happened when we expire a message after the export has been
827 cleaned up. Instead, hand the zombie cleanup off to another
830 Severity : enhancement
832 Description: plain/inodebits lock performance improvement
833 Details : Grouping plain/inodebits in granted list by their request modes
834 and bits policy, thus improving the performance of search through
838 Frequency : only if OST filesystem is corrupted
840 Description: client incorrectly hits assertion in ptlrpc_replay_req()
841 Details : for a short time RPCs with bulk IO are in the replay list,
842 but replay of bulk IOs is unimplemented. If the OST filesystem
843 is corrupted due to disk cache incoherency and then replay is
844 started it is possible to trip an assertion. Avoid putting
845 committed RPCs into the replay list at all to avoid this issue.
848 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
850 Description: System hang on startup
851 Details : This bug allowed the liblustre (e.g. catamount) client to
852 return to the app before handling all startup RPCs. This
853 could leave the node unresponsive to lustre network traffic
854 and manifested as a server ptllnd timeout.
856 Severity : enhancement
858 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
859 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
860 boolean to control whether to print peer debug info when a
861 client's RPC times out.
864 Frequency : only for kernels with patches from Lustre below 1.4.3
866 Description: Remove old rdonly API
867 Details : Remove old rdonly API which unused from at least lustre 1.4.3
870 Frequency : only for devices with external journals
872 Description: Set external device read-only also
873 Details : During a commanded failover stop, we set the disk device
874 read-only while the server shuts down. We now also set any
875 external journal device read-only at the same time.
878 Frequency : when upgrading from 1.4 while trying to change parameters
880 Description: The wrong (new) MDC name was used when setting parameters for
881 upgraded MDT's. Also allows changing of OSC (and MDC)
882 parameters if --writeconf is specified at tunefs upgrade time.
885 Frequency : when setting specific ost indicies
887 Description: QOS code breaks on skipped indicies
888 Details : Add checks for missing OST indicies in the QOS code, so OSTs
889 created with --index need not be sequential.
891 Severity : enhancement
893 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
894 Details : The uninit_groups feature works in conjunction with the kernel
895 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
896 up the pass1 processing of e2fsck. This is a read-only feature
897 in ldiskfs2 only, so older kernels and current ldiskfs cannot
898 mount filesystems that have had this feature enabled.
900 Severity : enhancement
902 Description: Improve multi-block allocation algorithm to avoid fragmentation
903 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
904 allocation locality and avoid filesystem fragmentation.
906 ------------------------------------------------------------------------------
908 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
910 * Support for kernels:
911 2.4.21-47.0.1.EL (RHEL 3)
913 2.6.9-42.0.10.EL (RHEL 4)
914 2.6.12.6 vanilla (kernel.org)
915 2.6.16.27-0.9 (SLES 10)
916 * Recommended e2fsprogs version: 1.39.cfs5
918 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
922 Frequency : occasional, depends on client load and configuration
923 Bugzilla : 12181, 12203
924 Description: data loss for recently-modified files
926 Details : In some cases it is possible that recently written or created
927 files may not be written to disk in a timely manner (this should
928 normally be within 30s unless client IO load is very high).
929 The problem appears as zero-length files or files that are a
930 multiple of 1MB in size after a client crash or client eviction
931 that are missing data at the end of the file.
933 This problem is more likely to be hit on clients where files are
934 repeatedly created and unlinked in the same directory, clients
935 have a large amount of RAM, have many CPUs, the filesystem has
936 many OSTs, the clients are rebooted frequently, and/or the files
937 are not accessed by other nodes after being written.
939 The presence of the problem can be detected by looking at
940 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
941 smaller than the second (nr_unused) then dirty files will not
942 be flushed automatically to disk. "sync; sleep 10" should be
943 run several times on the node before unmounting it to update
944 Lustre (this is also safe to run on nodes without this problem).
946 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
947 that can cause this same problem, so customers using that kernel
948 also need to update the kernel in addition to Lustre. In order
949 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
951 It is normal that files written just before a client crash (less
952 than 30s) may not yet have been flushed to disk, even for local
956 Frequency : frequent on thin XT3 nodes
958 Description: UUID collision on thin XT3 Linux nodes
959 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
960 randomly, since we relied on an insufficiently-seeded PRNG.
965 Description: OSS hangs after "All ost request buffers busy"
966 Details : A deadlock between quota and journal operations caused OSS
967 hangs after printing "All ost request buffers busy."
970 Frequency : always on liblustre builds
972 Description: Cleanup compiler warnings on liblustre
975 Frequency : always on liblustre builds on XT3
977 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
981 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
982 > 32000 subdirectories
983 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
984 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
985 subdirectory count crosses 32000. This will aid e2fsck to
986 correctly handle more than 32000 subdirectories.
989 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
991 Description: System hang on startup
992 Details : This bug allowed the liblustre (e.g. catamount) client to
993 return to the app before handling all startup RPCs. This
994 could leave the node unresponsive to lustre network traffic
995 and manifested as a server ptllnd timeout.
997 Severity : enhancement
999 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1000 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1001 boolean to control whether to print peer debug info when a
1002 client's RPC times out.
1007 Description: make O_SYNC working on 2.6 kernels
1008 Details : 2.6 kernels use different method for mark pages for write,
1009 so need add a code to lustre for O_SYNC work.
1014 Description: Failure to close file and release space on NFS
1015 Details : Put inode details into lock acquired in ll_intent_file_open.
1016 Use mdc_intent_lock in ll_intent_open to properly
1017 detect all kind of errors unhandled by mdc_enqueue.
1022 Description: proc file read during shutdown sometimes raced obd removal,
1024 Details : Add lock to prevent obd access after proc file removal.
1027 Frequency : Only for files larger than 4GB on 32-bit clients.
1029 Description: improperly doing page alignment of locks
1030 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1031 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1036 Description: RPCs being resent when they shouldn't be
1037 Details : Some RPCs that should not be resent are being resent. This
1038 can cause inconsistencies in the RPC state machine. Do not
1039 resend such requests.
1042 Frequency : rare, only with NFS export
1044 Description: Crash on NFS re-export node
1045 Details : under very unusual load conditions an assertion is hit in
1046 ll_intent_file_open()
1049 Frequency : only if OST filesystem is corrupted
1051 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1052 Details : for a short time RPCs with bulk IO are in the replay list,
1053 but replay of bulk IOs is unimplemented. If the OST filesystem
1054 is corrupted due to disk cache incoherency and then replay is
1055 started it is possible to trip an assertion. Avoid putting
1056 committed RPCs into the replay list at all to avoid this issue.
1061 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1062 Details : Large single O_DIRECT read and write calls can fail to allocate
1063 a sufficiently large buffer to process the request. In case of
1064 allocation failure the allocation is retried with a smaller
1065 buffer and broken into smaller requests.
1067 Severity : enhancement
1069 Description: Add -o localflock option to simulate old noflock behaviour.
1070 Details : This will achieve local-only flock/fcntl locks coherentness.
1075 Description: versioning check is incomplete
1076 Details : Checking the version difference of client vs. server, report
1077 error if the gap is too big.
1082 Description: add support PG_writeback bit
1083 Details : add support for PG_writeback bit for Lustre, for more carefull
1084 work with page cache in 2.6 kernel. This also fix some deadlocks
1085 and remove hack for work O_SYNC with 2.6 kernel.
1087 Severity : enhancement
1089 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1090 Details : The uninit_groups feature works in conjunction with the kernel
1091 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1092 up the pass1 processing of e2fsck. This is a read-only feature
1093 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1094 mount filesystems that have had this feature enabled.
1096 Severity : enhancement
1098 Description: Improve multi-block allocation algorithm to avoid fragmentation
1099 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1100 allocation locality and avoid filesystem fragmentation.
1102 ------------------------------------------------------------------------------
1104 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1106 * Support for kernels:
1107 2.6.9-42.0.3.EL (RHEL 4)
1108 2.6.5-7.276 (SLES 9)
1109 2.4.21-47.0.1.EL (RHEL 3)
1110 2.6.12.6 vanilla (kernel.org)
1111 2.6.16.21-0.8 (SLES10)
1112 * Recommended e2fsprogs version: 1.39.cfs2-0
1114 * The backwards-compatible /proc/sys/portals symlink has been removed
1115 in this release. Before upgrading, please ensure that you change
1116 any configuration scripts or /etc/sysctl.conf files that access
1117 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1118 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1119 in advance of the upgrade on any system running Lustre 1.4.6 or
1120 newer, since /proc/sys/lnet was added in that version.
1121 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1125 Frequency : only when quota is used
1127 Description: avoid scanning export list for quota master
1128 Details : Change the algorithms to avoid scanning export list in order
1129 to improve the efficiency.
1132 Frequency : MDS failover only, very rarely
1134 Description: "went back in time" messages on mds failover
1135 Details : The greatest transno may be lost when the current operation
1136 finishes with an error (transno==0) and the client's last_rcvd
1137 record is over-written. Save the greatest transno in the
1138 mds_last_transno for this case.
1141 Frequency : always for specific kernels and striping counts
1143 Description: client may get "Matching packet too big" without ACL support
1144 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1145 when trying to access files in certain configurations. The
1146 clients should in fact be denied when mounting because they do
1147 not understand ACLs.
1150 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1152 Description: many clients connecting with IO in progress causes connect timeouts
1153 Details : Avoid synchronous journal commits to avoid delays caused by many
1154 clients connecting/disconnecting when bulk IO is in progress.
1155 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1156 OST_IO_PORTAL to avoid delays behind potentially many pending
1160 Frequency : occasionally with multiple writers to a single file
1162 Description: shared writes to file may result in wrong size reported by stat()
1163 Details : Allow growing of kms when extent lock is cancelled
1166 Frequency : always with random mmap IO to multi-striped file
1168 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1169 Details : Only if the hole is at the end of OST object so that kms is too
1170 small. Fix is to increase kms accordingly in ll_nopage.
1173 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1175 Description: writes to a missing object would leak memory on the OST
1176 Details : If there is an inconsistency between the MDS and OST filesystems,
1177 such that the MDS references an object that doesn't exist, writes
1178 to that object will leak memory due to incorrect cleanup in the
1179 error handling path, eventually running out of memory on the OST.
1184 Description: Creating too long symlink causes lustre errors
1185 Details : Check symlink and name lengths before sending requests to MDS.
1188 Frequency : only if flock is enabled (not on by default)
1190 Description: posix locks not released on fd closure on 2.6.9+
1191 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1192 kernels, this caused such locks not to be released on fd close and
1193 then assertions on fs unmount about still used locks.
1196 Frequency : MDS failover only, very rarely
1198 Description: clients may get ASSERTION(granted_lock != NULL)
1199 Details : When request was taking a long time, and a client was resending
1200 a getattr by name lock request. The were multiple lock requests
1201 with the same client lock handle and
1202 mds_getattr_name->fixup_handle_for_resent_request found one of the
1203 lock handles but later failed with ASSERTION(granted_lock != NULL).
1208 Description: handle->h_buffer_credits > 0, assertion failure
1209 Details : h_buffer_credits is zero after truncate, causing assertion
1210 failure. This patch extends the transaction or creates a new
1214 Frequency : NFS re-export or patchless client
1215 Bugzilla : 11179, 10796
1216 Description: Crash on NFS re-export node (__d_move)
1217 Details : We do not want to hash the dentry if we don't have a lock.
1218 But if this dentry is later used in d_move, we'd hit uninitialised
1219 list head d_hash, so we just do this to init d_hash field but
1220 leave dentry unhashed.
1223 Frequency : NFS re-export or patchless client
1225 Description: NFS exports has problem with symbolic link
1226 Details : lustre client didn't properly install dentry when re-exported
1227 to NFS or running patchless client.
1230 Frequency : NFS re-export or patchless client
1232 Description: Various nfs/patchless fixes.
1233 Details : fixes reuse disconected alias for lookup process - this fixes
1234 warning "find_exported_dentry: npd != pd",
1235 fix permission error with open files at nfs.
1236 fix apply umask when do revalidate.
1239 Frequency : occasional
1241 Description: Crash on NFS re-export node
1242 Details : calling clear_page() on the wrong pointer triggered oops in
1243 generic_mapping_read().
1246 Frequency : rarely, using O_DIRECT IO
1248 Description: unaligned directio crashes client with LASSERT
1249 Details : check for unaligned buffers before trying any requests.
1252 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1254 Description: stale data returned from RAID cache
1255 Details : If only a small amount of IO is done to the RAID device before
1256 reading it again it is possible to get stale data from the RAID
1257 cache instead of reading it from disk.
1260 Frequency : always for sles10 kernel
1262 Description: sles10 support
1263 Details : ll_follow_link: compile fixes and using of nd_set_link
1264 under newer kernels.
1267 Frequency : depends on arch, kernel and compiler version, always on sles10
1270 Description: recursive or deep enough symlinks cause stack overflow
1271 Details : getting rid of large stack-allocated variable in
1275 Frequency : depends on hardware
1277 Description: lustre write performance loss in the SLES10 kernel
1278 Details : the performance loss is caused by using of write barriers in the
1279 ext3 code. The SLES10 kernel turns barrier support on by
1280 default. The fix is to undo that change for ldiskfs.
1282 ------------------------------------------------------------------------------
1284 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1286 * Support for kernels:
1287 2.6.9-42.0.3EL (RHEL 4)
1288 2.6.5-7.276 (SLES 9)
1289 2.4.21-47.0.1.EL (RHEL 3)
1290 2.6.12.6 vanilla (kernel.org)
1294 Frequency : quota enabled and large files being deleted
1296 Description: releasing more than 4GB of quota at once hangs OST
1297 Details : If a user deletes more than 4GB of files on a single OST it
1298 will cause the OST to spin in an infinite loop. Release
1299 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1304 Description: statfs data retrieved from /proc may be stale or zero
1305 Details : When reading per-device statfs data from /proc, in the
1306 {kbytes,files}_{total,free,avail} files, it may appear
1307 as zero or be out of date.
1310 Frequency : systems with MD RAID1 external journal devices
1312 Description: lconf's call to blkid is confused by RAID1 journal devices
1313 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1314 of returning all block devices that match the journal UUID.
1317 Frequency : always, for aggregate stripe size over 4GB
1319 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1320 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1321 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1324 Frequency : always if "lfs find" used on a local file/directory
1326 Description: "lfs find" segfaults if used on a local file/directory
1327 Details : The case where a directory component was not specified wasn't
1328 handled correctly. Handle this properly.
1331 Frequency : always on ppc64
1333 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1334 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1335 perform endianness conversion.
1338 Frequency : rarely (truncate to non-zero file size after write under load)
1339 Bugzilla : 10730, 10687
1340 Description: Files padded with zeros to next 4K multiple
1341 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1342 it is possible that files that are truncated to a non-zero size
1343 immediately after being written are filled with zero bytes beyond
1344 the truncated size. No file data is lost.
1346 Severity : enhancement
1348 Description: Allow recovery/failover for liblustre clients.
1349 Details : liblustre clients were unaware of failover configurations until
1352 Severity : enhancement
1354 Description: user file locks should fail when not mounting with flock option
1355 Details : Set up an error-returning stub in ll_file_operations.lock field
1356 to prevent incorrect behaviour when client is mounted without
1357 flock option. Also, set up properly f_op->flock field for
1361 Frequency : always on ia64
1363 Description: "lfs df" loops on printing out MDS statfs information
1364 Details : The obd_ioctl_data was not initialized and in some systems
1365 this caused a failure during the ioctl that did not return
1366 an error. Initialize the struct and return an error on failure.
1369 Frequency : SLES 9 only
1371 Description: Error of copying files with lustre special EAs as root
1372 Details : Client side always return success for setxattr call for lustre
1373 special xattr (currently only "trusted.lov").
1376 Frequency : rarely on clusters with both ia64+i386 clients
1378 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1379 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1380 clients are doing concurrent writes to the same file it is
1381 possible that the ia64 clients may LASSERT because the OST
1382 extent locks are not PAGE_SIZE aligned. Ensure that grown
1383 locks are always aligned on the request boundary.
1386 Frequency : specific use, occasional
1388 Description: Overwriting in use executable truncates on-disk binary image
1389 Details : If one node attempts to overwrite an executable in use by
1390 another node, we now correctly return ETXTBSY instead of
1391 truncating the file.
1393 Severity : enhancement
1395 Description: Async OSC create to avoid the blocking unnecessarily.
1396 Details : If a OST has no remain object, system will block on the creating
1397 when need to create a new object on this OST. Now, ways use
1398 pre-created objects when available, instead of blocking on an
1399 empty osc while others are not empty. If we must block, we block
1400 for the shortest possible period of time.
1405 Description: chmod on Lustre root is propagated to other clients
1406 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1407 invalid by the follow_mount time.
1410 Frequency : liblustre clients only
1412 Description: Race in 'instant cancel' lock handling could lead to such locks
1413 never to be granted in case of SMP MDS
1414 Details : Do not destroy not yet granted but cbpending locks in
1418 Frequency : replay/resend of open
1420 Description: non null lock assetion failure in mds_intent_policy
1421 Details : Trying to replay/resend lockless open requests resulted in
1422 mds_open() returning 0 with no lock. Now it sets a flag if
1423 a lock is going to be returned.
1425 Severity : enhancement
1427 Description: Checksum enhancements
1428 Details : New checksum enhancements allow for resending RPCs that failed
1431 Severity : enhancement
1433 Description: Tunables on number of dirty pages in cacche
1434 Details : Allow to set limit on number of dirty pages cached.
1439 Description: client crash on unmount - lock still has references
1440 Details : In some error handling cases it was possible to leak a lock
1441 reference on a client while accessing a file. This was not
1442 harmful to the client during operation, but would cause the
1443 client to crash when the filesystem is unmounted.
1446 Frequency : specific case, rare
1448 Description: ETXTBSY on mds though file not in use by client
1449 Details : ETXTBSY is no longer incorrectly returned when attempting to
1450 chmod or chown a directory that the user previously tried to
1451 execute or a currently-executing binary.
1454 Frequency : extremely rare except on liblustre-based clients
1456 Description: Lustre space not freed when files are deleted
1457 Details : Clean up open-unlinked files after client eviction. Previously
1458 the unlink was skipped and the files remained as orphans.
1463 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1464 Details : In some cases it was possible to send a blocking callback to
1465 a client doing a glimpse, even though that client didn't get
1466 a lock granted. When the glimpse lock is cancelled on the OST
1467 the freed lock is left on the waiting list and corrupted the list.
1470 Frequency : all core dumps
1472 Description: Broke core dumps to lustre
1473 Details : Negative dentry may be unhashed if parent does not have UPDATE
1474 lock, but some callers, e.g. do_coredump, expect dentry to be
1475 hashed after successful create, hash it in ll_create_it.
1477 ------------------------------------------------------------------------------
1479 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1481 * Support for kernels:
1482 2.6.9-42.0.2.EL (RHEL 4)
1483 2.6.5-7.276 (SLES 9)
1484 2.4.21-40.EL (RHEL 3)
1485 2.6.12.6 vanilla (kernel.org)
1489 Frequency : always on RHEL 3
1491 Description: Number of open files grows over time
1492 Details : The number of open files grows over time, whether or not
1493 Lustre is started. This was due to a filp leak introduced
1494 by one of our kernel patches.
1496 ------------------------------------------------------------------------------
1498 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
1500 * Support for kernels:
1501 2.6.9-42.EL (RHEL 4)
1502 2.6.5-7.267 (SLES 9)
1503 2.4.21-40.EL (RHEL 3)
1504 2.6.12.6 vanilla (kernel.org)
1509 Bugzilla : 5719, 9635, 9792, 9684
1510 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1511 Details : If a server is under heavy load and cannot reply to new
1512 connection requests before the client resends the (re)connect,
1513 the connection handling code can behave badly if two service
1514 threads are concurrently handing separate (re)connections from
1515 the same client. Add better locking to the connection handling
1516 code, and ensure that only a single connection will be processed
1517 for a given client UUID, even if the lock is dropped.
1519 Severity : enhancement
1521 Description: add TCP zero-copy support to kernel
1522 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1523 sends if the hardware supports scatter-gather and checksumming.
1524 This allows socklnd to do client-write and server-read more
1525 efficiently and reduce CPU utilization from skbuf copying.
1528 Frequency : only if NFS exporting from client
1530 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1531 Details : If setting attributes on a file created under NFS that had
1532 never been opened it would be possible to oops the client
1533 if the file had no objects.
1537 Bugzilla : 9326, 10402, 10897
1538 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1539 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1540 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1541 before ptlrpcd_wake() was called and the request was freed.
1544 Frequency : always for liblustre
1546 Description: liblustre client does MDS+OSTs setattr RPC for each write
1547 Details : When doing a write from a liblustre client, the client
1548 incorrectly issued an RPC to the MDS and each OST the file was
1549 striped over in order to update the timestamps. When writing
1550 with small chunks and many clients this could overwhelm the MDS
1551 with RPCs. In all cases it would slow down the write because
1552 these RPCs are unnecessary.
1554 Severity : enhancement
1556 Description: allow number of MDS service threads to be changed at module load
1557 Details : It is now possible to change the number of MDS service threads
1558 running. Adding "options mds mds_num_threads={N}" to the MDS's
1559 /etc/modprobe.conf will set the number of threads for the next
1560 time Lustre is restarted (assuming the "mds" module is also
1561 reloaded at that time). The default number of threads will
1562 stay the same, 32 for most systems.
1567 Description: OST crash if filesystem is unformatted or corrupt
1568 Details : If an OST is started on a device that has never been formatted
1569 or if the filesystem is corrupt and cannot even mount then the
1570 error handling cleanup routines would dereference a NULL pointer.
1575 Description: NULL pointer deref in llap_from_page.
1576 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1577 incorrect) mapping if the page was truncated/reclaimed while it was
1578 searched for. Check for this condition and skip such pages when
1579 doing readahead. Introduce extra check to llap_from_page() to
1580 verify page->mapping->host is non-NULL (so page is not anonymous).
1583 Frequency : Sometimes when using sys_sendfile
1585 Description: "page not covered by a lock" warnings from ll_readpage
1586 Details : sendfile called ll_readpage without right page locks present.
1587 Now we introduced ll_file_sendfile that does necessary locking
1588 around call to generic_file_sendfile() much like we do in
1592 Frequency : with certain MDS communication failures at client mount time
1594 Description: NULL pointer deref after failed client mount
1595 Details : a client connection request may delayed by the network layer
1596 and not be sent until after the PTLRPC layer has timed out the
1597 request. If the client fails the mount immediately it will try
1598 to clean up before the network times out the request. Add a
1599 reference from the request import to the obd device and delay
1600 the cleanup until the network drops the request.
1603 Frequency : occasionally during client (re)connect
1605 Description: assertion failure during client (re)connect
1606 Details : processing a client connection request may be delayed by the
1607 client or server longer than the client connect timeout. This
1608 causes the client to resend the connection request. If the
1609 original connection request is replied in this interval, the
1610 client may trip an assertion failure in ptlrpc_connect_interpret()
1611 which thought it would be the only running connect process.
1614 Frequency : only with obd_echo servers and clients that are rebooted
1616 Description: kernel BUG accessing uninitialized data structure
1617 Details : When running an obd_echo server it did not start the ping_evictor
1618 thread, and when a client was evicted an uninitialized data
1619 structure was accessed. Start the ping_evictor in the RPC
1620 service startup instead of the OBD startup.
1622 Severity : enhancement
1623 Bugzilla : 10193 (patchless)
1624 Description: Remove dependency on various unexported kernel interfaces.
1625 Details : No longer need reparent_to_init, exit_mm, exit_files,
1626 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1629 Frequency : rare (only users of deprecated and unsupported LDAP config)
1631 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1632 Details : LDAP apparently contains 'lustreName' attributes instead of
1633 'name'. A simple remapping of the name is sufficient.
1636 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1638 Description: waiting_locks_callback trips kernel BUG if client is evicted
1639 Details : Running with the dump_on_timeout debug flag turned on makes
1640 it possible that the waiting_locks_callback() can try to dump
1641 the Lustre kernel debug logs from an interrupt handler. Defer
1642 this log dumping to the expired_lock_main() thread.
1644 Severity : enhancement
1646 Description: Support NFS exporting on 2.6 kernels.
1647 Details : Implement non-rawops metadata methods for NFS server to use without
1648 changing NFS server code.
1651 Frequency : very rare (synthetic metadata workload only)
1653 Description: two racing renames might cause an MDS thread to deadlock
1654 Details : Running the "racer" program may cause one MDS thread to rename
1655 a file from being the source of a rename to being the target of
1656 a rename at exactly the same time that another thread is doing
1657 so, and the second thread has already enqueued these locks after
1658 doing a lookup of the target and is trying to relock them in
1659 order. Ensure that we don't try to re-lock the same resource.
1662 Frequency : only very large systems with liblustre clients
1664 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1665 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1666 all OSTs in parallel. This allows the eviction of stale liblustre
1667 clients to proceed much faster than if they were done in series,
1668 and also offers similar improvements for other set_info RPCs.
1673 Description: excessive CPU usage during initial read phase on client
1674 Details : During the initial read phase on a client, it would agressively
1675 retry readahead on the file, consuming too much CPU and impacting
1676 performance (since 1.4.5.8). Improve the readahead algorithm
1677 to avoid this, and also improve some other common cases (read
1678 of small files in particular, where "small" is files smaller than
1679 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1684 Description: MDS crash when receiving packet with unknown intent.
1685 Details : Do not LBUG in unknown intent case, just return -EFAULT
1687 Severity : enhancement
1688 Bugzilla : 9293, 9385
1689 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1690 Details : Do not serialize getattr (non-intent version) and statfs.
1693 Frequency : occasional, when OST network is overloaded/intermittent
1695 Description: client evicted by OST after bulk IO timeout
1696 Details : If a client sends a bulk IO request (read or write) the OST
1697 may evict the client if it is unresposive to its data GET/PUT
1698 request. This is incorrect if the network is overloaded (takes
1699 too long to transfer the RPC data) or dropped the OST GET/PUT
1700 request. There is no need to evict the client at all, since
1701 the pinger and/or lock callbacks will handle this, and the
1702 client can restart the bulk request.
1705 Frequency : Always when mmapping file with no objects
1707 Description: client crashes when mmapping file with no objects
1708 Details : Check that we actually have objects in a file before doing any
1709 operations on objects in ll_vm_open, ll_vm_close and
1715 Description: Request leak when working with deleted CWD
1716 Details : Introduce advanced request refcount tracking for requests
1717 referenced from lustre intent.
1719 Severity : Enhancement
1721 Description: Cache open file handles on client.
1722 Details : MDS now will return special lock along with openhandle, if
1723 requested and client is allowed to hold openhandle, even if unused,
1724 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1725 closing files for every read/write openration.
1727 Severity : Enhancement
1729 Description: Cache open negative dentries on client when possible.
1730 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1731 negative dentries on lock revocation.
1736 Description: Remounting a client read-only wasn't possible with a zconf mount
1737 Details : It wasn't possible to remount a client read-only with llmount.
1739 Severity : enhancement
1740 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1741 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1742 MPICH is included to add Lustre-specific ADIO interfaces.
1743 This is based closely on the UFS ADIO layer and only differs
1744 in file creation, in order to allow the OST striping to be set.
1745 This is user-contributed code and not supported by CFS.
1750 Description: extended inode attributes (immutable, append-only) work improperly
1751 when 2.4 and 2.6 kernels are used on client/server or vice versa
1752 Details : Introduce kernel-independent values for these flags.
1754 Severity : enhancement
1757 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1758 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1759 granularity. Now, Fractional MB granularity is be supported,
1760 this is very useful for low memory system.
1762 Severity : enhancement
1764 Description: Getattr by fid
1765 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1766 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1767 have locks and avoids getattr rpc requests for it on every stat.
1770 Frequency : Always, for filesystems larger than 2TB
1772 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1773 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1774 filesystems larger than 2TB to be mounted with mballoc enabled.
1777 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1779 Description: filesystem corruption for non-standard kernels and very large OSTs
1780 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1781 filesystems larger than 2TB is mounted then the kernel will
1782 silently corrupt the start of the filesystem. CONFIG_LBD is
1783 enabled for all CFS-supported kernels, but the possibility of
1784 this happening with a modified kernel config exists.
1786 Severity : enhancement
1788 Description: add client O_DIRECT support for 2.6 kernels
1789 Details : It is now possible to do O_DIRECT reads and writes to files
1790 in the Lustre client mountpoint on 2.6 kernel clients.
1792 Severity : enhancement
1794 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1795 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1796 parallel, not waiting for response from every OST before sending
1797 a rpc to the next OST.
1802 Description: setattr vs write race when updating file timestamps
1803 Details : Client processes that update a file timestamp into the past
1804 right after writing to the file (e.g. tar) it is possible that
1805 the updated file modification time can be reset to the current
1806 time due to a race between processing the setattr and write RPC.
1808 Severity : enhancement
1810 Description: Bring 'lfs find' closer in line with regular Linux find.
1811 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1812 -print0 options and obtains all the needed info through the lustre
1815 Severity : enhancement
1817 Description: support up to 1024 configured devices on one node
1818 Details : change obd_dev array from statically allocated to dynamically
1819 allocated structs as they are first used to reduce memory usage
1824 Description: Flush dirty partially truncated pages during truncate
1825 Details : Immediatelly flush partially truncated pages in filter_setattr,
1826 this way we completely avoid having any pages in page cache on OST
1827 and can retire ugly workarounds during writes to flush such pages.
1832 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1833 Details : protect inode from truncation within vfs_unlink() context
1834 just take a reference before calling vfs_unlink() and release it
1835 when parent's i_sem is free.
1838 Frequency : always, if extents are used on OSTs
1840 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1841 Details : OSTs using the extents format would not zero the high 16 bits of
1842 the index physical block number. This is not a problem for any
1843 OST filesystems smaller than 16TB, and no kernels support ext3
1844 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1845 new/modified files) and can be fixed for existing filesystems
1846 with e2fsprogs-1.39-cfs1.
1851 Description: import connection selection may be incorrect if timer wraps
1852 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1853 import connections to be ignored if the 32-bit jiffies counter
1854 wraps. Use a 64-bit jiffies counter.
1857 Frequency : during server recovery
1859 Description: crash after server is denying duplicate export
1860 Details : If clients are resending connect requests to the server, the
1861 server refuses to allow a client to connect multiple times.
1862 Fixed a bug in the handling of this case.
1865 Frequency : very large clusters immediately after boot
1867 Description: LNET request buffers exhausted under heavy short-term load
1868 Details : If a large number of client requests are generated on a service
1869 that has previously never seen so many requests it is possible
1870 that the request buffer growth cannot keep up with the spike in
1871 demand. Instead of dropping incoming requests, they are held in
1872 the LND until the RPC service can accept more requests.
1875 Frequency : Sometimes during replay
1877 Description: Assertion failure in ll_local_open after replay.
1878 Details : If replay happened on an open request reply before we were able
1879 to set replay handler, reply will become not swabbed tripping the
1880 assertion in ll_local_open. Now we set the handler right after
1881 recognising of open request
1884 Frequency : very rare
1886 Description: kernel reports "badness in vsnprintf"
1887 Details : Reading from the "recovery_status" /proc file in small chunks
1888 may cause a negative length in lprocfs_obd_rd_recovery_status()
1889 call to vsnprintf() (which is otherwise harmless). Exit early
1890 if there is no more space in the output buffer.
1892 Severity : enhancement
1894 Description: clear OBD RPC statistics by writing to them
1895 Details : It is now possible to clear the OBD RPC statistics by writing
1896 to the "stats" file.
1901 Description: Client mtime is not the same on different clients after utimes
1902 Details : In some cases, the client was using the utimes() syscall on
1903 a file cached on another node. The clients now validate the
1904 ctime from the MDS + OSTs to determine which one is right.
1909 Description: Inability to activate failout mode
1910 Details : lconf script incorrectly assumed that in python string's numeric
1911 value is used in comparisons.
1914 Frequency : always with multiple stripes per file
1916 Description: Inefficient object allocation for mutli-stripe files
1917 Details : When selecting which OSTs to stripe files over, for files with
1918 a stripe count that divides evenly into the number of OSTs,
1919 the MDS is always picking the same starting OST for each file.
1920 Return the OST selection heuristic to the original design.
1925 Description: mount failures may take full timeout to return an error
1926 Details : Under some heavy load conditions it is possible that a
1927 failed mount can wait for the full obd_timeout interval,
1928 possibly several minutes, before reporting an error.
1929 Instead return an error as soon as the status is known.
1931 ------------------------------------------------------------------------------
1933 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1935 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1936 INTEROPERATE with older versions automatically. Please read the
1937 user documentation before upgrading any part of a live system.
1938 * WARNING: Lustre networking configuration changes are required with
1939 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1942 * Support for kernels:
1943 2.6.9-22.0.2.EL (RHEL 4)
1944 2.6.5-7.244 (SLES 9)
1945 2.6.12.6 vanilla (kernel.org)
1948 Severity : enhancement
1949 Bugzilla : 7981/8208
1950 Description: Introduced Lustre Networking (LNET)
1951 Details : LNET is new networking infrastructure for Lustre, it includes
1952 a reorganized network configuration mode (see the user
1953 documentation for full details) as well as support for routing
1954 between different network fabrics. Lustre Networking Devices
1955 (LNDs) for the supported network fabrics have also been
1956 created for this new infrastructure.
1958 Severity : enhancement
1959 Description: Introduced Access control lists
1960 Details : clients can set ACLs on files and directories in order to have
1961 more fine-grained permissions than the standard Unix UGO+RWX.
1962 The MDS must be started with the "-o acl" mount option.
1964 Severity : enhancement
1965 Description: Introduced filesystem quotas
1966 Details : Administrators may now establish per-user quotas on the
1969 Severity : enhancement
1971 Description: Configuration change for the XT3
1972 The PTLLND is now used to run Lustre over Portals on the XT3
1973 The configure option(s) --with-cray-portals are no longer used.
1974 Rather --with-portals=<path-to-portals-includes> is used to
1975 enable building on the XT3. In addition to enable XT3 specific
1976 features the option --enable-cray-xt3 must be used.
1981 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1982 Details : A race between allocating a new llcd and re-getting the llcd_lock
1983 allowed another thread to grab newly-allocated llcd.
1985 Severity : enhancement
1987 Description: 2.6 OST async journal commit and locking fix to improve performance
1988 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1989 2.6 kernels are now async as they already were in 2.4 kernels so
1990 that they can commit concurrently with the network bulk transfer.
1991 For block-allocated files the filter allocation semaphore is held
1992 to avoid filesystem fragmentation during allocation. BKL lock
1993 removed for 2.6 xattr operations where it is no longer needed.
1998 Description: lconf incorrectly determined whether two IP networks could talk
1999 Details : In some more complicated routing and multiple-network
2000 configurations, lconf will avoid trying to make a network
2001 connection to a disjoint part of the IP space. It was doing the
2002 math incorrectly for one set of cases.
2007 Description: Fix for potential infinite loop processing records in an llog.
2008 Details : If an llog record is corrupted/zeroed, it is possible to loop
2009 forever in llog_process(). Validate the llog record length
2010 and skip the remainder of the block on error.
2013 Frequency : occasional (liblustre only)
2015 Description: liblustre could not open files whose last component is a symlink
2016 Details : sysio_path_walk() would incorrectly pass the open intent to
2017 intermediate path components.
2020 Frequency : rare (liblustre only with non-standard tuning)
2021 Bugzilla : 7201 (7350)
2022 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2023 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2024 after it has been released, only do this for referenced locks
2026 Severity : enhancement
2028 Description: specifying an (invalid) directory default stripe_size of -1
2029 would reset the directory default striping
2030 Details : stripe_size -1 was used internally to signal directory stripe
2031 removal, now use "all default" to signal dir stripe removal
2032 as a directory striping of "all default" is not useful
2035 Frequency : common for large clusters running liblustre clients
2037 Description: doing an ls when liblustre clients are running is slow
2038 Details : sending a glimpse AST to a liblustre client waits for every AST
2039 to time out, as liblustre clients will not respond. Since they
2040 cannot cache data we refresh the OST lock LVB from disk instead.
2042 Severity : enhancement
2044 Description: doing an ls at the same time as file IO can be slow
2045 Details : enqueue and other "small" requests can be blocked behind many
2046 large IO requests. Create a new OST IO portal for non-IO
2047 requests so they can be processed faster.
2050 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2052 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2053 created file with mode 000
2054 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2055 allow an ftruncate() as a non-root user to a file with mode 000.
2056 Lustre now respects this flag to disable mode checks when
2057 truncating a file owned by the user
2060 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2063 Description: Revoking locks from clients that went dead or catatonic might take
2065 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2066 cancellation of such locks instant on servers without waiting for
2067 any reply from clients. Clients drops these locks when cancel
2068 notification from server is received without replying.
2071 Frequency : liblustre-only, when liblustre client dies or becomes busy
2073 Description: Doing ls on Linux clients can take a long time with active
2075 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2076 granting such locks to it in the first place if possible. Locks
2077 are taken by proxy on the OST during the read or write and
2078 dropped immediately afterward. Add connect flags handling, do
2079 not grant locks to liblustre clients for glimpse ASTs.
2081 Severity : enhancement
2083 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2084 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2085 sequential reads and behaves badly with applications doing random
2086 reads. Improve it by reading ahead at least read region, and
2087 avoiding excessive large RPC for small reads.
2089 Severity : enhancement
2091 Description: Creating more than 1000 files for a single job may cause a load
2092 imbalance on the OSTs if there are also a large number of OSTs.
2093 Details : qos_prep_create() uses an OST index reseed value that is an
2094 even multiple of the number of available OSTs so that if the
2095 reseed happens in the middle of the object allocation it will
2096 still utilize the OSTs as uniformly as possible.
2101 Description: OST or MDS may oops in ping_evictor_main()
2102 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2103 but doesn't restart at beginning of obd_exports_timed list
2106 Severity : enhancement
2108 Description: improve by-nid export eviction on the MDS and OST
2109 Details : allow multiple exports with the same NID to be evicted at one
2110 time without re-searching the exports list.
2113 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2115 Description: MDS may oops in groups_free()
2116 Details : in rare race conditions a newly allocated group_info struct is
2117 freed again, and this can be NULL. The 2.4 compatibility code
2118 for groups_free() checked for a NULL pointer, but 2.6 did not.
2121 Frequency : common for liblustre clients doing little filesystem IO
2122 Bugzilla : 9352, 7313
2123 Description: server may evict liblustre clients accessing contended locks
2124 Details : if a client is granted a lock or receives a completion AST
2125 with a blocking AST already set it would not reply to the AST
2126 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2129 Frequency : lfs setstripe, only systems with more than 160 OSTs
2131 Description: unable to set striping with a starting offset beyond OST 160
2132 Details : llapi_create_file() incorrectly limited the starting stripe
2133 index to the maximum single-file stripe count.
2136 Frequency : LDAP users only
2138 Description: lconf did not handle in-kernel recovery with LDAP properly
2139 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2141 Severity : enhancement
2143 Description: bind OST threads to NUMA nodes to improve performance
2144 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2145 node and do their allocations there to localize memory access
2147 Severity : enhancement
2149 Description: llmount can determine client NID directly from Myrinet (GM)
2150 Details : the client NID code from gmnalnid was moved directly into
2151 llmount, removing the need to use this or specifying the
2152 client NID explicitly when mounting GM clients with zeroconf
2155 Frequency : if client is started with down MDS
2157 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2158 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2159 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2160 ptlrpc_import_delayed_req() because we didn't check if the
2161 request was interrupted, and we also didn't break out of the
2162 event loop if there was no timeout
2167 Description: data loss during non-page-aligned writes to a single file from
2168 both multiple nodes and multiple threads on one node at same time
2169 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2170 inconsistency led to false short-reads, that were cached and later
2171 used by ->prepare_write() to fill in partially written page,
2172 leading to data loss.
2175 Frequency : always, if lconf --abort_recovery used
2177 Description: lconf --abort_recovery fails with 'Operation not supported'
2178 Details : lconf was attempting to abort recovery on the MDT device and not
2181 Severity : enhancement
2183 Description: remove cleanup logs
2184 Details : replace lconf-generated cleanup logs with lustre internal
2185 cleanup routines. Eliminates the need for client-cleanup and
2188 Severity : enhancement
2190 Description: add support for EAs (user and system) on lustre filesystems
2191 Details : it is now possible to store extended attributes in the Lustre
2192 client filesystem, and with the user_xattr mount option it
2193 is possible to allow users to store EAs on their files also
2195 Severity : enhancement
2197 Description: Add possibility (config option) to show minimal available OST free
2199 Details : When compiled with --enable-mindf configure option, statfs(2)
2200 (and so, df) will return least minimal free space available from
2201 all OSTs as amount of free space on FS, instead of summary of
2202 free spaces of all OSTs.
2204 Severity : enhancement
2206 Description: do not expand extent locks acquired on OST-side
2207 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2208 by server: they are not cached anyway.
2211 Frequency : when mmap is used/binaries executed from Lustre
2213 Description: Unmmap pages before throwing them away from read cache.
2214 Details : llap_shrink cache now attempts to unmap pages before discarding
2215 them (if unmapping failed - do not discard). SLES9 kernel has
2216 extra checks that trigger if this unmapping is not done first.
2221 Description: lconf didn't resolve symlinks before checking to see whether a
2222 given mountpoint was already in use
2225 Frequency : when migrating failover services
2226 Bugzilla : 6395, 9514
2227 Description: When migrating a subset of services from a node (e.g. failback
2228 from a failover service node) the remaining services would
2229 time out and evict clients.
2230 Details : lconf --force (implied by --failover) sets the global obd_timeout
2231 to 5 seconds in order to quickly disconnect, but this caused
2232 other RPCs to time out too quickly. Do not change the global
2233 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2235 Severity : enhancement
2236 Frequency : if MDS is started with down OST
2237 Bugzilla : 9439,5706
2238 Description: Allow startup/shutdown of an MDS without depending on the
2239 availability of the OSTs.
2240 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2241 Add appropriate locking and lov-osc refcounts for safe
2242 cleaning. Add osc abort_inflight calls in case the
2243 synchronize never started.
2246 Frequency : occasional (Cray XT3 only)
2248 Description: root not authorized to access files in CRAY_PORTALS environment
2249 Details : The client process capabilities were not honoured on the MDS in
2250 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2251 been accessed by an authorized user then root was able to access
2252 the file on the local client also. The root user capabilities
2253 are now allowed on the MDS, as this environment has secure UID.
2256 Frequency : occasional
2258 Description: ldiskfs "too long searching" message happens too often
2259 Details : A debugging message (otherwise harmless) prints too often on
2260 the OST console. This has been reduced to only happen when
2261 there are fragmentation problems on the filesystem.
2266 Description: Division by zero in statfs when all OSCs are inactive
2267 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2268 lov_statfs divides by value returned by lov_get_stripecnt().
2272 Bugzilla : 9489, 3273
2273 Description: First write from each client to each OST was only 4kB in size,
2274 to initialize client writeback cache, which caused sub-optimal
2275 RPCs and poor layout on disk for the first writen file.
2276 Details : Clients now request an initial cache grant at (re)connect time
2277 and so that they can start streaming writes to the cache right
2278 away and always do full-sized RPCs if there is enough data.
2279 If the OST is rebooted the client also re-establishes its grant
2280 so that client cached writes will be honoured under the grant.
2285 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2286 Details : Now I/O RPCs go to different portal number and (presumably) fast
2287 lock requests (and glimses) and other RPCs get their own service
2288 threads pool that should be able to service those RPCs
2291 Severity : enhancement
2293 Description: Ability to exchange lustre version between client and servers and
2294 issue warnings at client side if client is too old. Also for
2295 liblustre clients there is ability to refuse connection of too old
2297 Details : New 'version' field is added to connect data structure that is
2298 filled with version info. That info is later checked by server and
2302 Frequency : rare, liblustre only.
2303 Bugzilla : 9296, 9581
2304 Description: Two simultaneous writes from liblustre at offset within same page
2305 might proceed at the same time overwriting eachother with stale
2307 Details : I/O lock withing llu_file_prwv was released too early, before data
2308 actually was hitting the wire. Extended lock-holding time until
2309 server acknowledges receiving data.
2312 Frequency : extremely rare. Never observed in practice.
2314 Description: avoid generating lustre_handle cookie of 0.
2315 Details : class_handle_hash() generates handle cookies by incrementing
2316 global counter, and can hit 0 occasionaly (this is unlikely, but
2317 not impossible, because initial value of cookie counter is
2318 selected randonly). Value of 0 is used as a sentinel meaning
2319 "unassigned handle" --- avoid it. Also coalesce two critical
2320 sections in this function into one.
2322 Severity : enhancement
2324 Description: allow liblustre clients to delegate truncate locking to OST
2325 Details : To avoid overhead of locking, liblustre client instructs OST to
2326 take extent lock in ost_punch() on client's behalf. New connection
2327 flag is added to handle backward compatibility.
2329 Severity : enhancement
2330 Bugzilla : 4928, 7341, 9758
2331 Description: allow number of OST service threads to be specified
2332 Details : a module parameter allows the number of OST service threads
2333 to be specified via "options ost ost_num_threads={N}" in the
2334 OSS's /etc/modules.conf or /etc/modprobe.conf.
2338 Bugzilla : 6146, 9635, 9895
2339 Description: servers crash with bad pointer in target_handle_connect()
2340 Details : In rare cases when a client is reconnecting it was possible that
2341 the connection request was the last reference for that export.
2342 We would temporarily drop the export reference and get a new
2343 one, but this may have been the last reference and the export
2344 was just destroyed. Get new reference before dropping old one.
2346 Severity : enhancement
2347 Frequency : if client is started with failover MDS
2349 Description: Allow multiple MDS hostnames in the mount command
2350 Details : Try to read the configuration from all specified MDS
2351 hostnames during a client mount in case the "primary"
2354 Severity : enhancement
2356 Description: Stop sending data to evicted clients as soon as possible.
2357 Details : Check if the client we are about to send or are sending data to
2358 was evicted already. (Check is done every second of waiting,
2359 for which l_wait_event interface was extended to allow checking
2360 of exit condition at specified intervals).
2363 Frequency : rare, normally only when NFS exporting is done from client
2365 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2367 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2370 Frequency : always, if srand() is called before liblustre initialization
2372 Description: Liblustre uses system PRNG disturbing its usage by user application
2373 Details : Introduce internal to lustre fast and high-quality PRNG for
2374 lustre usage and make liblustre and some other places in generic
2375 lustre code to use it.
2377 Severity : enhancement
2378 Bugzilla : 9477, 9557, 9870
2379 Description: Verify that the MDS configuration logs are updated when xml is
2380 Details : Check if the .xml configuration logs are newer than the config
2381 logs stored on the MDS and report an error if this is the case.
2382 Request --write-conf, or allow starting with --old_conf.
2384 Severity : enhancement
2386 Description: Handle symlinks in the path when checking if Lustre is mounted.
2387 Details : Resolve intermediate symlinks when checking if a client has
2388 mounted a filesystem to avoid duplicate client mounts.
2393 Description: lconf can hit an error exception but still return success.
2394 Details : The lconf command catches the Command error exception at the top
2395 level script context and will exit with the associated exit
2396 status, but doesn't ensure that this exit status is non-zero.
2401 Description: failure of ptlrpc thread startup can cause oops
2402 Details : Starting a ptlrpc service thread can fail if there are a large
2403 number of threads or the server memory is very fragmented.
2404 Handle this without oopsing.
2407 Frequency : always, only if liblustre and non-default acceptor port was used
2409 Description: liblustre cannot connect to servers with non-default acceptor port
2410 Details : tcpnal_set_default_params() was not called and was therefore
2411 ignoring the environment varaible TCPNAL_PORT, as well as other
2412 TCPNAL_ environment variables
2417 Description: two objects could be created on the same OST for a single file
2418 Details : If an OST is down, in some cases it was possible to create two
2419 objects on a single OST for a single file. No problems other
2420 than potential performance impact and spurious error messages.
2424 Bugzilla : 5681, 9562
2425 Description: Client may oops in ll_unhash_aliases
2426 Details : Client dcache may become inconsistent in race condition.
2427 In some cases "getcwd" can fail if the current directory is
2433 Description: Inode refcounting problems in NFS export code
2434 Details : link_raw functions used to call d_instantiate without obtaining
2435 extra inode reference first.
2439 Bugzilla : 9942, 9903
2440 Description: Referencing freed requests leading to crash, memleaks with NFS.
2441 Details : We used to require that call to ll_revalidate_it was always
2442 followed by ll_lookup_it. Also with revalidate_special() it is
2443 possible to call ll_revalidate_it() twice for the same dentry
2444 even if first occurence returned success. This fix changes semantic
2445 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2446 reference on a request referred from the intent.
2447 ll_intent_release() then releases such a request.
2450 Frequency : rare, normally benchmark loads only
2452 Description: unlinked inodes were kept in memory on the client
2453 Details : If a client is repeatedly creating and unlinking files it
2454 can accumulate a lot of stale inodes in the inode slab cache.
2455 If there is no other client load running this can cause the
2456 client node to run out of memory. Instead flush old inodes
2457 from client cache that have the same inode number as a new inode.
2460 Frequency : SLES9 2.6.5 kernel and long filenames only
2461 Bugzilla : 9969, 10379
2462 Description: utime reports stale NFS file handle
2463 Details : SLES9 uses out-of-dentry names in some cases, which confused
2464 the lustre dentry revalidation. Change it to always use the
2468 Frequency : rare, unless heavy write-truncate concurrency is continuous
2469 Bugzilla : 4180, 6984, 7171, 9963, 9331
2470 Description: OST becomes very slow and/or deadlocked during object unlink
2471 Details : filter_destroy() was holding onto the parent directory lock
2472 while truncating+unlinking objects. For very large objects this
2473 may block other threads for a long time and slow overall OST
2474 responsiveness. It may also be possible to get a lock ordering
2475 deadlock in this case, or run out of journal credits because of
2476 the combined truncate+unlink. Solution is to do object truncate
2477 first in one transaction without parent lock, and then do the
2478 final unlink in a new transaction with the parent lock. This
2479 reduces the lock hold time dramatically.
2482 Frequency : rare, 2.4 kernels only
2484 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2485 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2486 before unmounting their filesystem in order to avoid deadlock.
2487 The kernel_locked() function in 2.4 kernels only checks whether
2488 the kernel lock is held, not whether it is this process that is
2489 holding it as 2.6 kernels do.
2494 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2495 Details : The client ptlrpc code may be trying to reconnect to a down
2496 server before a previous connection attempt has timed out.
2497 Increase the reconnect interval to be longer than the connection
2498 timeout interval to avoid sending duplicate connections to
2502 Frequency : echo_client brw_test command
2504 Description: fix echo_client to work with OST preallocated code
2505 Details : OST preallocation code (5137) didn't take echo_client IO path
2506 into account: echo_client calls filter methods outside of any
2507 OST thread and, hence, there is no per-thread preallocated
2508 pages and buffers to use. Solution: hijack pga pages for IO. As
2509 a byproduct, this avoids unnecessary data copying.
2513 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2514 Description: Client can oops in mdc_commit_close() after open replay
2515 Details : It was possible for the MDS to return an open request with no
2516 transaction number in mds_finish_transno() if the client was
2517 evicted, but without actually returning an error. Clients
2518 would later try to replay that open and may trip an assertion
2519 Simplify the client close codepath, and always return an error
2520 from the MDS in case the open is not successful.
2523 Frequency : rare, 2.6 OSTs only
2525 Description: OST may deadlock under high load on fragmented files
2526 Details : If there was a heavy load and highly-fragmented OST filesystems
2527 it was possible to have all the OST threads deadlock waiting on
2528 allocation of biovecs, because the biovecs were not released
2529 until the entire RPC IO was completed. Instead, release biovecs
2530 as soon as they are complete to ensure forward IO progress.
2532 Severity : enhancement
2534 Description: Support for specifying external journal device at mount
2535 Details : If an OST or MDS device is formatted with an external journal
2536 device, this device major/minor is stored in the ext3 superblock
2537 and may not be valid for failover. Allow detecting and
2538 specifying the external journal at mount time.
2543 Description: Mounting an MDS with pending unlinked files may cause oops
2544 Details : target_finish_recovery() calls mds_postrecov() which returned
2545 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2546 considers this an error and immediately begins cleaning up the
2547 lov, just after starting the mds_lov process
2549 Severity : enhancement
2551 Description: Implement 'lfs df' to report actual free space on per-OST basis
2552 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2553 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2554 usage of objects; '-h' to report in human readable format.
2556 ------------------------------------------------------------------------------
2558 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
2565 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2566 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2567 at startup during orphan unlinking. Instead, skip buddy update
2568 before setup, as it will be regenerated after recovery is complete.
2573 Description: If an OST is inactive, its locks might reference stale inodes.
2574 Details : lov_change_cbdata() must iterate over all namespaces, even if
2575 they are inactive to clear inode references from the lock.
2577 Severity : enhancement
2578 Frequency : occasional, if non-standard max_dirty_mb used
2580 Description: Client will block write RPCs if not enough grant
2581 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2582 then the client will block writes while waiting for another RPC
2583 to complete instead of consuming its dirty limit. With change
2584 we get improved performance when max_dirty_mb is small.
2586 Severity : enhancement
2587 Bugzilla : 3389, 6253
2588 Description: Add support for supplementary groups on the MDS.
2589 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2590 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2591 lookups for user supplementary groups into a cache.
2595 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2596 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2597 for liblustre O_EXCL is enforced by MDS.
2601 Description: readdir never returns NULL in liblustre.
2602 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2603 dentry in struct dirent.
2607 Frequency : liblustre only, depends on application IO pattern
2608 Description: liblustre clients evicted if not contacting servers
2609 Details : Don't put liblustre clients into the ping_evictor list, so
2610 they will not be evicted by the pinger ever.
2612 Severity : enhancement
2614 Description: Add ability to evict clients by NID from MDS.
2615 Details : By echoing "nid:$NID" string into
2616 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2617 $NID would be instantly evicted from this MDS and from all active
2618 OSTs connected to it.
2622 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2623 Details : lookup_it_finish() used to query file size from OSTs that was not
2628 Description: service threads change working directory to that of init
2629 Details : Starting lustre service threads may pin the working directory
2630 of the parent thread, making that filesystem busy. Threads
2631 now change to the working directory of init to avoid this.
2635 Frequency : during shutdown only
2636 Description: shutdown with a failed MDS or OST can cause unmount to hang
2637 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2642 Frequency : 2.6 only
2643 Description: chmod/chown may include an extra supplementary group
2644 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2645 supplementary group and if none is specified this is used.
2648 Bugzilla : 5479 (6816)
2650 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2651 Details : If lookup is in progress on a file that is unlinked we might try
2652 to revalidate the inode and fail in revalidate after lookup is
2653 complete and ll_file_open() enqueues the open again but
2654 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2657 Frequency : always, if lconf --abort_recovery used
2659 Description: lconf --abort_recovery fails with 'Operation not supported'
2660 Details : lconf was attempting to abort recovery on the MDT device and not
2663 ------------------------------------------------------------------------------
2665 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2670 Frequency : rare (only unsupported configurations with a node running as an
2672 Bugzilla : 6514, 5137
2673 Description: Mounting a Lustre file system on a node running as an OST could
2675 Details : OSTs now preallocates memory needed to write out data at
2676 startup, instead of when needed, to avoid having to
2677 allocate memory in possibly low memory situations.
2678 Specifically, if the file system is mounted on on OST,
2679 memory pressure could force it to try to write out data,
2680 which it needed to allocate memory to do. Due to the low
2681 memory, it would be unable to do so and the node would
2682 become unresponsive.
2684 Severity : enhancement
2686 Description: Addition of lconf --service command line option
2687 Details : lconf now accepts a '--service <arg>' option, which is
2688 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2690 Severity : enhancement
2692 Description: Failover mode is now the default for OSTs.
2693 Details : By default, OSTs will now run in failover mode. To return to
2694 the old behaviour, add '--failout' to the lmc line for OSTs.
2696 Severity : enhancement
2698 Description: Health checks are now provided for MDS and OSTs
2699 Details : Additional detailed health check information on MSD and OSTs
2700 is now provided through the procfs health_check value.
2703 Frequency : occasional, depends on IO load
2705 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2706 after numerous create/delete cycles
2707 Details : The ext3 inode allocation policy would not allocate new inodes
2708 very well on the OSTs because there are no new directories
2709 being created. Instead we look for groups with free space if
2710 the parent directories are nearly full.
2715 Description: Network or server problems during mount may cause partially
2716 mounted clients instead of returning an error.
2717 Details : The config llog parsing code may overwrite the error return
2718 code during mount error handling, returning success instead
2724 Description: MDS can fail to allocate large reply buffers
2725 Details : After long uptimes the MDS can fail to allocate large reply
2726 buffers (e.g. zconf client mount config records) due to memory
2727 fragmentation or consumption by the buffer cache. Preallocate
2728 some large reply buffers so that these replies can be sent even
2729 under memory pressure.
2733 Frequency : rare (liblustre)
2734 Description: fsx running with liblustre complained that using truncate() to
2735 extend the file doesn't work. This patch corrects that issue.
2736 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2737 ATTR_SIZE and lsm use in llu_setattr_raw.
2741 Frequency : rare, only 2.6 kernels
2742 Description: Unusual file access patterns on the MDS may result in inode
2743 data being lost in very rare circumstances.
2744 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2745 code in the 2.6 kernel caused the inode and/or EA data not to
2746 be read from disk, causing single-file corruption.
2750 Frequency : rare, only 2.6 filesystems using extents
2751 Description: Heavy concurrent write and delete load may cause data corruption.
2752 Details : It was possible under high-load situations to have an extent
2753 metadata block in the block device cache from a just-unlinked
2754 file overwrite a newly-allocated data block. We now unmap any
2755 metadata buffers that alias just-allocated data blocks.
2759 Frequency : filesystems with default stripe_count larger than 77
2760 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2761 Details : lconf specifies an inode size of 4096 bytes when the default
2762 stripe_count is larger than 77. This conflicts with the default
2763 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2764 this case to avoid pinning too much memory for large EAs.
2766 ------------------------------------------------------------------------------
2768 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2773 Frequency : rare (extremely heavy IO load with hundreds of clients)
2775 Description: Client is evicted, gets IO error writing to file
2776 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2777 caused clients to be evicted for AST timeouts. The fixes in
2778 bug 5192 mean we no longer need to have such short AST timeouts
2779 so ldlm_timeout has been increased.
2782 Frequency : occasional during --force or --failover shutdown under load
2783 Bugzilla : 5949, 4834
2784 Description: Server oops/LBUG if stopped with --force or --failover under load
2785 Details : a collection of import/export refcount and cleanup ordering
2786 issues fixed for safer force cleanup
2789 Frequency : only filesystems larger than 120 OSTs
2790 Bugzilla : 5990, 6223
2791 Description: lfs getstripe would oops on a very large filesystem
2792 Details : lov_getconfig used kfree on vmalloc'd memory
2795 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2796 Bugzilla : 6242, 6243
2797 Description: reading from files that had been truncated to a non-zero size
2798 but never opened returned no data
2799 Details : ll_file_read() reads zeros from no-object files to EOF
2804 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2805 Details : There's little harm in aborting MDS/OSS recovery and letting it
2806 try again, so I removed the LASSERT and return an error instead.
2808 Severity : enhancement
2810 Description: New debugging infrastructure for tracking down data corruption
2811 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2812 (b) cover more of the client-side code path, and (c) try to narrow
2813 down where problems occurred
2817 Bugzilla : 3819, 4364, 4397, 6313
2818 Description: Racing close and eviction MDS could cause assertion in mds_close
2819 Details : It was possible to get multiple mfd references during close and
2820 client eviction, leading to one thread referencing a freed mfd.
2822 Severity: : enhancement
2823 Bugzilla : 3262, 6359
2824 Description: Attempts to reconnect to servers are now more aggressive.
2825 Details : This builds on the enhanced upcall-less recovery that was added
2826 in 1.4.2. When trying to reconnect to servers, clients will
2827 now try each server in the failover group every 10 seconds. By
2828 default, clients would previously try one server every 25 seconds.
2833 Description: After recovery, certain operations trigger a failed
2834 assertion on a client.
2835 Details : Failing over an mds, using lconf -d --failover, while a
2836 client was doing a readdir() call would cause the client to
2837 LBUG after recovery completed and the readdir() was resent.
2839 Severity : enhancement
2841 Description: Default groups are now added by lconf
2842 Details : You can now run lconf --group <servicename> without having to
2843 manually add groups with lmc.
2846 Frequency : occasional
2848 Description: Nodes with an elan id of 0 trigger a failed assertion
2851 Frequency : always when accessing e.g. tty/console device nodes
2853 Description: tty and some other devices nodes cannot be used on lustre
2854 Details : file's private_data field is used by device data and lustre
2855 values in there got lost. New field was added to struct file to
2856 store fs-specific private data.
2859 Frequency : when exporting Lustre via NFS
2861 Description: NFSD failed occasionally when looking up a path component
2862 Details : NFSD is looking up ".." which was broken in ext3 directories
2863 that had grown large enough to become hashed.
2866 Frequency : Clusters with multiple interfaces not on the same subnet
2868 Description: Nodes will repeatedly try to reconnect to an interface which it
2869 cannot reach and report an error to the log.
2870 Details : Extra peer list entries will be created by lconf with some peers
2871 unreachable. lconf now validates the peer before adding it.
2874 Frequency : Only if a default stripe is set on the filesystem root.
2876 Description: Setting a default stripe on the filesystem root prevented the
2877 filesystem from being remounted.
2878 Details : The client was sending extra request flags in the root getattr
2879 request and did not allocate a reply buffer for the dir EA.
2882 Frequency : occasional, higher if lots of files are accessed by one client
2883 Bugzilla : 6159, 6097
2884 Description: Client trips assertion regarding lsm mismatch/magic
2885 Details : While revalidating inodes the VFS looks up inodes with ifind()
2886 and in rare cases can find an inode that is being freed.
2887 The ll_test_inode() code will free the lsm during ifind()
2888 when it finds an existing inode and then the VFS later attaches
2889 this free lsm to a new inode.
2893 Bugzilla : 6422, 7030
2894 Description: MDS deadlock between mkdir and client eviction
2895 Details : Creating a new file via mkdir or mknod (starting a transaction
2896 and getting the ns lock) can deadlock with client eviction
2897 (gets ns lock and trying to finish a synchronous transaction).
2900 Frequency : occasional
2901 Description: While starting a server, the fsfilt_ext3 module could not be
2903 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2904 kernels. Previously, lconf would still use the ext3 name
2905 when trying to load modules. Now, it will correctly use
2906 ext3 on 2.4 and ldiskfs on 2.6.
2908 Severity : enhancement
2909 Description: The default stripe count has been changed to 1
2910 Details : The interpretation of the default stripe count (0, to lfs
2911 or lmc) has been changed to mean striping across a single
2912 OST, rather than all available. For general usage we have
2913 found a stripe count of 1 or 2 works best.
2915 Severity : enhancement
2916 Description: Add support for compiling against Cray portals.
2917 Details : Conditional compiling for some areas that are different
2921 Frequency : occasional
2922 Bugzilla : 6409, 6834
2923 Description: Creating files with an explicit stripe count may lead to
2924 a failed assertion on the MDS
2925 Details : If some OSTs are full or unavailable, creating files may
2926 trigger a failed assertion on the MDS. Now, Lustre will
2927 try to use other servers or return an error to the
2931 Frequency : occasional
2933 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2934 caused lock timeout during readahead (since 1.4.2).
2935 Details : Processes doing readahead might match a lock that hasn't been
2936 granted yet if there are overlapping and conflicting lock
2937 requests. The readahead process waits on ungranted lock
2938 (original lock is CBPENDING), while OST waits for that process
2939 to cancel CBPENDING read lock and eventually evicts client.
2941 Severity : enhancement
2943 Description: Initial enabling of flock support for clients
2944 Details : Implements fcntl advisory locking and file status functions.
2945 This feature is provided as an optional mount flag (default
2946 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2947 locking are implemented yet, and those that are are not guaranteed
2948 to be completely correct in production environments.
2949 mount -t lustre -o [flock|noflock] ...
2952 Frequency : occasional
2954 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2955 assertion in the ext3 JBD (journaling) layer.
2956 Details : The b_committed_data struct is protected by the big kernel lock
2957 in 2.4 kernels, serializing journal_commit_transaction() and
2958 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2959 access to this struct. In 2.6 kernels there is finer grained
2960 locking to improve SMP performance of the JBD layer.
2964 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2966 -----------------------------------------------------------------------------
2968 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2970 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2971 versions of Lustre. Please update all servers and clients to
2972 version 1.4.2 or later at the same time. You must also run
2973 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2974 to update the configuration logs.
2976 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2977 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2978 - swab llog records of type '0' so we get proper header size/idx (5861)
2979 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2980 - fix rename of one directory over another leaking an inode (5953)
2981 - avoid SetPageDirty on 2.6 (5981)
2982 - don't re-add just-being-destroyed locks to the waiting list (5653)
2983 - when creating new directories, inherit the parent's custom
2984 striping settings if present parent (3048)
2985 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2986 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2987 - don't hold client locks on temporary worklist from l_lru (5666)
2988 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2989 - automatically evict dead clients (5921)
2990 - Update file size properly in create+truncate+fstat case (6196)
2991 - Do not unhash mountpoint dentries, do not allow removal of
2993 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2994 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2995 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2996 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2997 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2998 - Correctly handle reads of files with no objects (6243)
2999 - lctl recover will also mark a device active if deactivate used (5933)
3001 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3002 - allow --write-conf on an MDS with different nettype than client (5619)
3003 - don't write config llogs to MDS for mounts not from that MDS (5617)
3004 - lconf should create multiple TCP connections from a client (5201)
3005 - init scripts are now turned off by default; run chkconfig --on
3006 lustre and chkconfig --on lustrefs to use them
3007 - upcalls are no longer needed for clients to recover to failover
3009 - add --abort-recovery option to lconf to abort recovery on device
3011 - add support for an arbitrary number of OSTs (3026)
3012 - Quota support protocol changes.
3013 - forward compatibility changes to wire structs (6007)
3014 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3015 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3016 - improved "lustre status" script
3017 - initialize blocksize for non-regular files (6062)
3018 - added --disable-server and --disable-client configure options (5782)
3019 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3020 - Vanilla 2.4.29 support
3021 - increase maximum number of obd devices to 520 (6242)
3022 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3023 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3024 SLES 9 2.6.5 kernels
3025 - limit stripes per file to 160 (the maximum EA size) (6093)
3027 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3030 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3031 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3032 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3033 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3034 - small ext3 extents cleanups and fixes (5733)
3035 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3036 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3037 - avoid needless client->OST connect, fix handle mismatch (5317)
3038 - fix DLM error path that led to out-of-sync client, long delays (5779)
3039 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3040 - fix several locking issues related to i_size (5492,5624,5654,5672)
3041 - don't move pending lock onto export if it is already evicted (5683)
3042 - fix kernel oops when creating .foo in unlinked directory (5548)
3043 - fix deadlock in obdfilter statistics vs. object create (5811)
3044 - use time_{before,after} to avoid timer jiffies wrap (5882)
3045 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3046 - Do not leak request if server was not able to process it (5154)
3047 - If mds_open unable to find parent dir, make that negative lookup(5154)
3048 - don't create new directories with extent-mapping (5909, 5936)
3050 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3051 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3052 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3053 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3054 - support for RHEL4 kernel on i686 (5773)
3055 - provide error messages when incompatible logs are encountered (5898)
3057 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3058 * version 1.4.0.10 (1.4.1 release candidate 1)
3060 - don't keep a lock reference when lock is not granted (4238)
3061 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3062 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3063 - drop import inflight refcount on signal_completed_replay error (5255)
3064 - unlock page after async write error during send (3677)
3065 - handle missing objects in filter_preprw_read properly (5265)
3066 - no transno return for symlink open, don't save no-trasno open (3440)
3067 - don't try to complete elan receive that already failed (4012)
3068 - free RPC server reply state on error (5406)
3069 - clean up thread from ptlrpc_start_thread() on error (5160)
3070 - readahead could read extra page into cache that wasn't ejected (5388)
3071 - prevent races in class_attach/setup/cleanup/detach (5260)
3072 - don't dereference de->d_inode after l_dput of de (5458)
3073 - use "int" for stripe value returned from lock_to_stripe (5544)
3074 - mballoc allocation and error-checking fixes in 2.6 (5504)
3075 - block device patches to fix I/O request sizes in 2.6 (5482)
3076 - look up hostnames for IB nals (5602)
3077 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3078 - don't start multiple acceptors for the same port (5277)
3079 - fix incorrect LASSERT in mds_getattr_name (5635)
3080 - export a proc file for general "ping" checking (5628)
3081 - fix "lfs check" to not block when the MDS is down (5628)
3083 - service request history (4965)
3084 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3085 - create an "evict_client" /proc entry on OSTs, like the MDS has
3086 - fix mount usage message, return errors per mount(8) (5168)
3087 - change grep [] to grep "[]" in tests so they work in more UMLs
3088 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3089 - remove ancient LOV_MAGIC_V0 EA support (5047)
3090 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3091 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3092 - allow readahead allocations to fail when low on memory (5383)
3093 - mmap locking landed again, after considerable improvement (2828)
3094 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3096 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3099 - send OST transaction number in read/write reply to free req (4966)
3100 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3101 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3102 - the watchdog thread now runs as interruptible (5246)
3103 - flock/lockf fixes (but it's still disabled, pending 5135)
3104 - don't use EXT3 constants in llite code (5094)
3105 - memory shortage at startup could cause assertion (5176)
3107 - reorganization of lov code
3108 - single portals codebase
3110 - add extents/mballoc support (5025)
3111 - direct I/O reads in the obdfilter (4048)
3112 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3114 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3117 - send OST transaction number in read/write reply to free req (4966)
3118 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3119 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3120 - the watchdog thread now runs as interruptible (5246)
3121 - handle missing objects in filter_preprw_read properly (5265)
3122 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3123 - drop import inflight refcount on signal_completed_replay error (5255)
3124 - unlock page after async write error during send (3677)
3125 - return original error code on reconstructed replies (3761)
3126 - no transno return for symlink open, don't save no-trasno open (3440)
3128 - add pid to ldlm debugging output (4922)
3129 - bump the watchdog timeouts -- we can't handle 30sec yet
3130 - extra debugging for orphan dentry/inode bug (5259)
3132 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3135 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3136 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3137 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3138 - fix NULL dereference in /proc/sys/portals/routes (4827)
3139 - allow failed mdc_close() operations to be interrupted (4561)
3140 - stop precreate on OST before MDS would time out on it (4778)
3141 - don't send partial-page writes before EOF from client (4410)
3142 - discard client grant for sub-page writes on large-page clients (4520)
3143 - don't free dentries not owned by NFS code, check generation (4806)
3144 - fix lsm leak if mds_create_objects() fails (4801)
3145 - limit debug_daemon file size, always print CERROR messages (4789)
3146 - use transno after validating reply (3892)
3147 - process timed out requests if import state changes (3754)
3148 - update mtime on OST during writes, return in glimpse (4829)
3149 - add mkfsoptions to LDAP (4679)
3150 - use ->max_readahead method instead of zapping global ra (5039)
3151 - don't interrupt __l_wait_event() during strace
3153 - add software watchdogs to catch hung threads quickly (4941)
3154 - make lustrefs init script start after nfs is mounted
3155 - fix CWARN/ERROR duplication (4930)
3156 - return async write errors to application if possible (2248)
3157 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3158 - print NAL number in %x format (4645)
3159 - update barely-supported suse-2.4.21-171 series (4842)
3160 - support for sles 9 %post scripts
3161 - support for building 2.6 kernel-source packages
3162 - support for sles km_* packages
3164 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3167 - ignore -ENOENT errors in osc_destroy (3639)
3168 - notify osc create thread that OSC is being cleaned up (4600)
3169 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3170 - reconstruct ost_handle() like mds_handle() (4657)
3171 - create a new thread to do import eviction to avoid deadlock (3969)
3172 - let lconf resolve symlinked-to devices (4629)
3173 - don't unlink "objects" from directory with default EA (4554)
3174 - hold socknal file ref over connect in case target is down (4394)
3175 - allow more than 32000 subdirectories in a single directory (3244)
3176 - fix blocks count for O_DIRECT writes (3751)
3177 - OST returns ENOSPC from object create when no space left (4539)
3178 - don't send truncate RPC if file size isn't changing (4410)
3179 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3180 - bind to privileged port in socknal and tcpnal (3689)
3182 - rate limit CERROR/CWARN console message to avoid overload (4519)
3183 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3184 - basic mmap support (3918)
3185 - kernel patch series update from b1_4 (4711)
3187 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3190 - avoid crash during MDS cleanup with OST shut down (2775)
3191 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3192 - don't use bad inodes on the MDS (3744)
3193 - dynamic object preallocation to improve recovery speed (4236)
3194 - don't hold spinlock over lock dumping or change debug flags (4401)
3195 - don't zero obd_dev when it is force cleaned (3651)
3196 - print grants to console if they go negative (4431)
3197 - "lctl deactivate" will stop automatic recovery attempts (3406)
3198 - look for existing locks in ldlm_handle_enqueue() (3764)
3199 - don't resolve lock handle twice in recovery avoiding race (4401)
3200 - revalidate should check working dir is a directory (4134)
3202 - don't always mark "slow" obdfilter messages as errors (4418)
3204 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3207 - don't close LustreDB during write_conf until it is done (3860)
3208 - fix typo in lconf for_each_profile (3821)
3209 - allow dumping logs from multiple threads at one time (3820)
3210 - don't allow multiple threads in OSC recovery (3812)
3211 - fix debug_size parameters (3864)
3212 - fix mds_postrecov to initialize import for llog ctxt (3121)
3213 - replace config semaphore with spinlock (3306)
3214 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3215 - don't allow enqueue to complete on a destroyed export (3822)
3216 - down write_lock before checking llog header bitmap (3825)
3217 - recover from lock replay timeout (3764)
3218 - up llog sem before sending rpc (3652)
3219 - reduce ns lock hold times when setting kms (3267)
3220 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3221 - fix NULL deref and obd_dev leak on setup error (3312)
3222 - replace some LBUG about llog ops with error handling (3841)
3223 - don't match INVALID dentries from d_lookup and spin (3784)
3224 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3225 - fix invalid assertion in ptlrpc_set_wait (3880)
3227 - add libwrap support for the TCP acceptor (3996)
3228 - add /proc/sys/portals/routes for non-root route listing (3994)
3229 - allow setting MDS UUID in .xml (2580)
3230 - print the stack of a process that LBUGs (4228)
3232 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3235 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3236 - make sure to unset replay flag from failed open requests (3440)
3237 - if default stripe count is 0, use OST count for inode size (3636)
3238 - update parent mtime/ctime on client for create/unlink (2611)
3239 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3240 - free recovery state on server during a forced cleanup (3571)
3241 - unregister_reply for resent reqs (3063)
3242 - loop back devices mounting and status check on 2.6 (3563)
3243 - fix resource-creation race that can provoke i_size == 0 (3513)
3244 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3245 - more debugging for page-accounting assertion (3746)
3246 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3247 - avoid deadlock after precreation failure (3758)
3248 - fix race and lock order deadlock in orphan handling (3450, 3750)
3249 - add validity checks when grabbing inodes from l_ast_data (3599)
3251 - add /proc/.../recovery_status to obdfilter (3428)
3252 - lightweight CDEBUG infrastructure, debug daemon (3668)
3253 - change default OSC RPC parameters to be better on small clusters
3254 - turn off OST read cache for files smaller than 32MB
3255 - install man pages and include them in rpms (3100)
3256 - add new init script for (un)mounting lustre filesystems (2593)
3257 - run chkconfig in %post for init scripts (3701)
3258 - drop scimac NAL (unmaintained)
3260 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3263 - clean kiobufs before and after use (3485)
3264 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3265 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3266 - change init script to fail more clearly if not run as root (1528)
3267 - allow clients to reconnect during replay (1742)
3268 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3269 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3270 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3271 - fix cond_resched (3554)
3272 - only evict unfinished clients after recovery (3515)
3273 - allow bulk resend, prevent data loss (3570)
3274 - dynamic ptlrpc request buffer allocation (2102)
3275 - don't allow unlinking open directory if it isn't empty (2904)
3276 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3277 - remove extraneous obd dereference causing LASSERT failure (3334)
3278 - don't use get_cycles() when creating temp. files on the mds (3156)
3279 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3280 - handle EEXIST for set-stripe, set proper directory name (3336)
3282 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3283 - fix ksocknal_fmb_callback() error messages (2918)
3285 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3288 - don't copy lvb into (possibly NULL) reply on error (2983)
3289 - don't deref dentry after dput, don't free lvb on error (2922)
3290 - use the kms to determine writeback rpc length (2947)
3291 - increment oti_logcookies when osc is inactive (2948)
3292 - update client's i_blocks count via lvb messages (2543)
3293 - handle intent open/close of special files properly (1557)
3294 - mount MDS with errors=remount-ro, like obdfilter (2009)
3295 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3296 - don't use cancelling-locks' kms values (2947)
3297 - use highest lock extent for kms, not last one (2925)
3298 - don't dereference ERR_PTR() dentry in error handling path (3107)
3299 - fix thread race in portals_debug_dumplog() (3122)
3300 - create lprocfs device entries at setup instead of at attach (1519)
3301 - common AST error handler, don't evict client on completion race (3145)
3302 - zero nameidata in detach_mnt in 2.6 (3118)
3303 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3304 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3305 - initialize RPC timeout timer earlier for 2.6 (3219)
3306 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3307 - print nal/nid for unknown nid (3258)
3308 - additional checks for oscc recovery before doing precreate (3284)
3309 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3310 - don't crash in mdc_close for bad permissions on open (3285)
3311 - zero i_rdev for non-device files (3147)
3312 - clear page->private before handing to FS, better assertion (3119)
3313 - tune the read pipeline (3236)
3314 - fix incorrect decref of invalidated dentry (2350)
3315 - provide read-ahead stats and refine rpc in flight stats (3328)
3316 - don't hold journal transaction open across create RPC (3313)
3317 - update atime on MDS at close time (3265)
3318 - close LDAP connection when recovering to avoid server load (3315)
3319 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3320 - don't leak open file on MDS after open resend (3325)
3321 - serialize filter_precreate and filter_destroy_precreated (3329)
3322 - loop device shouldn't call sync_dev() for nul device (3092)
3323 - clear page cache after eviction (2766)
3324 - resynchronize MDS->OST in background (2824)
3325 - refuse to mount the same filesystem twice on same mountpoint (3394)
3326 - allow llmount to create routes for mounting behind routers (3320)
3327 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3328 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3329 - fix rare problem with rename on htree directories (3417)
3331 - allow default OST striping configuration per directory (1414)
3332 - fix compilation for qswnal for 2.6 kernels (3125)
3333 - increase maximum number of MDS request buffers for large systems
3334 - change liblustreapi to be useful for external progs like lfsck (3098)
3335 - increase local configuration timeout for slow disks (3353)
3336 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3338 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3341 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3342 - don't overwrite extent policy data in reply if lock was blocked (2901)
3343 - drop filter export grants atomically with removal from device (2663)
3344 - del obd_self_export from work_list in class_disconnect_exports (2908)
3345 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3346 - swab reply message in mdc_close, other PPC fixes (2464)
3347 - fix destroying of named logs (2325)
3348 - overwrite old logs when running lconf --write_conf (2264)
3349 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3350 - fix race in target_handle_connect (2898)
3351 - mds_reint_create() should take same inode create lock (2926)
3352 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3353 - don't close files for self_export to avoid uninitialized obd (2936)
3354 - allow MDS with the same name as client node (2939)
3355 - hold dentry reference for closed log files for unlink (2325)
3356 - reserve space for all logs during transactions (2059)
3357 - don't evict page beyond end of stripe extent (2925)
3358 - don't oops on a deleted current working directory (2399)
3359 - handle hard links to targets without a parent properly (2517)
3360 - don't dereference NULL lock when racing during eviction (2867)
3361 - don't grow lock extents when lots of conflicting locks (2919)
3363 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3366 - account for cache space usage on clients to avoid data loss (974)
3367 - lfsck support in lustre kernel code (2349)
3368 - reduce journal credits needed for BRW writes (2370)
3369 - orphan handling to avoid losing space on client/server crashes
3370 - ptlrpcd can be blocked, stopping ALL progress (2477)
3371 - use lock value blocks to assist in proper KMS, faster stat (1021)
3372 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3373 - recovery for initial connections (2355)
3374 - fixes for mds_cleanup_orphans (1934)
3375 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3376 - block all file creations until orphan recovery completes (1901)
3377 - client remove rq_connection from request struct (2423)
3378 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3379 - recovery timer race (2670)
3380 - mdc_close recovey bug (2532)
3381 - ptlrpc cleanup bug (2710)
3382 - mds timeout on local locks (2588)
3383 - namespace lock held during RPCs (2431)
3384 - handle interrupted sync write properly (2503)
3385 - don't try to handle a message that hasn't been replied to (2699)
3386 - client assert failure during cleanup after abort recovery (2701)
3387 - leak mdc device after failed mount (2712)
3388 - ptlrpc_check_set allows timedout requests to complete (2714)
3389 - wait for inflight reqs when ptlrpcd finishes (2710)
3390 - make sure unregistered services are removed from the srv_list
3391 - reset bulk XID's when resending them (caught by 1138 test)
3392 - unregister_bulk after timeout
3393 - fix lconf error (2694)
3394 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3395 - readahead locks pages, leaves pending causing memory pressure (2673)
3396 - increase OST request buffers to 4096 on large machines (2729)
3397 - fix up permission of existing directories in simple_mkdir (2661)
3398 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3399 - don't assign transno to errored transactions (2742)
3400 - don't delete objects on OST if given a bogus objid from MDS (2751)
3401 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3402 - if rq_no_resend, then timeout request after recovery (2432)
3403 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3404 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3405 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3406 - MDS recovery shouldn't delete objects that it has given out (2730)
3407 - if enqueue arrives after completion, don't clobber LVB (2819)
3408 - don't unlock pages twice when trigger_group_io returns error (2814)
3409 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3410 - don't write pages to disk if there was an error (1450)
3411 - don't ping imports that have recovery disabled (2676)
3412 - take buffered bytes into account when balancing socknal conn (2817)
3413 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3414 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3415 - remove little-endian swabbing of llog records (1987)
3416 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3417 - retry reposting request buffers if they fail (1191)
3418 - grow extent at grant time to avoid granting a revoked lock (2809)
3419 - lock revoke doesn't evict page if covered by a second lock (2765)
3420 - disable VM readahead to avoid reading outside lock extents (2805)
3422 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3423 - updated kernel patches for hp-2.4.20 kernel (2681)
3425 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3428 - fix truncated write corruption (2366)
3429 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3431 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3432 - ensure len(uuid) < 37 in lmc (1171)
3433 - fix ia64 OOPS in llog_test (2255)
3434 - zero end of page at obdfilter for partial page writes (2648)
3435 - don't leave stale dentries around after renames (bug 2428)
3436 - fix timeouts when evicting a client with a single lock held (2642)
3437 - set deadline for the initial HELLO message to drain (2634)
3438 - print out dotted-quad IP addresses in the socknal (2302)
3440 - additional debugging for MDS client eviction problem (2443)
3441 - fix mkfsoptions support for osts (2603, 2604)
3443 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3446 - add series for the vanilla 2.6.0 kernel
3447 - add series for the vanilla 2.4.24 kernel
3448 - add series for a cray x86/64 UL kernel drop
3449 - fix xattr patches for the vanilla 2.4.19 series
3451 - generate true UUIDs in lmc (1171)
3452 - have portals stack dumping break in UML (2466)
3453 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3454 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3455 - wake callback waiting threads on client eviction (2460)
3456 - Add --ptldebug and --subsystem to lmc (1719)
3457 - update assertion to allow safe interrupt allocation
3458 - set rq_no_resend for cancel requests (2432)
3459 - recalculate ptlrpcd timeout after resend (2494)
3460 - call vfs_rmdir when removing pending directories (2368)
3461 - fix renaming a file to itself (2429)
3462 - lmc creates a default one-stripe lov (2454)
3463 - expand procfs space to handle large clusters (2326)
3464 - increase UML stack to avoid overflow
3465 - update lconf's list of debug and subsystem masks
3466 - fix lfs find --obd (2510)
3467 - /proc tunable for disabling filter read caching (2591)
3468 - stop rpm packages from altering slapd.conf (2301)
3469 - disable nagle in the socknal under 0conf (2578)
3470 - choose mds inode size based on stripe count (2572)
3471 - fix kernel-source rpm problems (2516)
3473 - add --disable-doc to avoid pdf generation (2421)
3474 - update documentation, tests, type-os, comments
3475 - avoid format warnings on ia64
3476 - remove the TOE NAL
3477 - tiny code cleanups by removing unused fields
3479 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3482 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3483 - remove the most blatant lies from BUILDING file (2371)
3484 - change default debug level to reasonable production setting
3485 - reduce client side cache size to reduce cache flush time
3486 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3487 - make TCP zerocopy and pinger support enabled by default (2476)
3488 - sync writes completed after process exits caused crashes (2319)
3489 - maintain correct mount count on the MDS (2356)
3490 - backout 1557, because 2316 wasn't really fixed
3491 - better file I/O statistics gathering in /proc
3492 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3493 - another kernel patch to fix zero-copy TCP function export
3494 - don't take duplicate lock when processing re-sent getattr (2420)
3495 - lctl uses obd_self_export instead of creating new conn (2353)
3496 - MDS/OST recovery case which requires object creation asserted (2425)
3497 - move lfs from /usr/sbin to /usr/bin in packages
3498 - fix race between mds_client_add and mds_client_free (2417)
3499 - use kmalloc instead of slabs in portals (2430)
3500 - don't create duplicate records when a failover MDS is present (2442)
3501 - remove unnecessary mount age check (2332)
3502 - don't remove directory inodes from locks prematurely (2451)
3503 - don't break if MDS service name is the same as hostname (2103)
3504 - fix races in client write RPC generation when cache full (2482)
3506 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3509 - remove now-unused request->rq_obd (278)
3510 - if an allocation fails, print out how much memory we've used (1933)
3511 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3512 - add the "configurable stack size" patch to most series files (1256)
3513 - ability to write large log records, for 100+ OST configs (2306)
3514 - fix NULL deref when filter_prep fails (2314)
3515 - fix operator precedence error in filter_sync
3516 - dynamic allocation of socknal TX descriptors (2315)
3517 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3518 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3519 - ensure that CWARN messages in Portals always get to the syslog
3520 - __init/__exit are not for prototype decls (ldlm_init/exit)
3521 - x86-64 compile warning fixes
3522 - fix gateway LMC keyword conflict (2318)
3523 - fix MDS lock inversions in getattr/reint paths (1844)
3524 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3525 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3526 - DLM race condition prevented some lock evictions (2328)
3527 - ENOMEM detection and retry on socknal sends (2230)
3528 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3529 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3530 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3531 - better error messages when a client is rejected during recovery (1505)
3532 - avoid cancelling locks which were never granted, after failure (2330)
3533 - fix i_sem/journal inversion in mds_client_add (2333)
3534 - fix truncate/getattr lock cycle deadlock (2334)
3535 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3536 - fix two rare exit paths which could leak an l_lock() ref (2321)
3537 - fencepost error in MDS/OST orphan recovery (2226)
3538 - make log record alignment 8 bytes (1988)
3539 - lstripe now fails when requested offset > ost_count (2237)
3540 - ensure that all kernel series have a complete list.h (1607)
3541 - fix crashes in special-file operations (2316)
3542 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3544 - allow configurable automake binary, for testing new versions
3545 - small update to the lfs documentation
3547 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3549 * fix negative export reference count in fsfilt_sync (2312)
3551 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3552 * release candidate 0.9.1
3554 - orphans are moved into the PENDING directory for possible recovery
3555 - replayed opens now open by fid for orphan/rename safety (1042)
3556 - last close of an orphan inode generates a transno (683)
3557 - chdir() and mount() now pin the directory entry (1020)
3558 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3559 - discard very old requests without processing them (1502)
3560 - remove some common, well-understood CERRORs (1505)
3561 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3562 - clear "grant" flags in OST replies until OST grant code lands (1644)
3563 - fix read performance by not clobbering i_blksize on client (1598)
3564 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3565 - make LRU size a /proc tunable, clears locks when reduced (707)
3566 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3567 - server locks take references on exports now (1558)
3568 - build fixes for 2.4.20-rh trees (1663)
3569 - return an error from lov_create if all OSCs are inactive (1751)
3570 - fix import levels when a reconnect happens without a timeout (1597)
3571 - exit early from mds_open if we get a lookup error (1749)
3572 - partial page read at EOF wouldn't wait for disk before sending (1642)
3573 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3574 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3575 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3576 - write/truncate lock inversion (1639)
3577 - Don't auto-load obdclass, portals modules during cleanup (1495)
3578 - fix timestamps from jumping to "now" (1763)
3579 - extra journal assertions (1648)
3580 - add an extra multiunlink test (1771)
3581 - fix read_record/write_record API (1776)
3582 - fix leak of offset_extent, possible incorrect i_size later (1772)
3583 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3584 - Debugging for the kqswnal_get_idle_tx problems (1820)
3585 - Allow recovery to be attempted multiple times (1536)
3586 - Write out MDS last_rcvd file after it is first created (1600)
3587 - Fix tx_descriptor leak in failed transmit situations (1827)
3588 - ext3 journaling fixes for assertion failure after IO error (1871)
3589 - class_export_put() on freed export after completion AST error (1896)
3590 - Fix revalidate looping in VFS (1322)
3591 - Don't access a freed export during MDS_REINT timeout (1521)
3592 - Add open-unlink recovery support on the MDS (1673,1764)
3593 - Return an error if no MDS data was read from last_rcvd (1946)
3594 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3595 - Don't setup a disk device that doesn't match exported UUID (317)
3596 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3597 - avoid committing NULL handle in force close
3598 - local.sh is now a one-stripe LOV configuration
3599 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3600 - fix ext3 htree duplicate directory entry corruption (1516)
3601 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3602 - update to new LOV EA format (2097)
3603 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3605 2003-06-15 Phil Schwan <phil@clusterfs.com>
3608 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3609 - per-import/export recovery handling (958, 931, 959)
3610 - multiple last-rcvd slots, for serving multiple FSes (949)
3611 - connections are again shared between multiple imp/exports (963, 964)
3612 - "umount -f" would hang if any requests needed to be sent (393, 978)
3613 - avoid pinning large req buffer by copying for queued messages (989)
3614 - add "uuid" to "lctl device" command to help upcalls (991)
3615 - "open" RPCs with transnos would confuse recovery counters (1037)
3616 - do proper endian conversion of all wire messages (288, 340, 891)
3617 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3618 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3619 - fix ptlrpc_connection leak in target_handle_connect (1174)
3620 - fix import refcounting bug in OST and MDS cleanup (1134)
3621 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3622 - fix very unlikely obd_types race condition (501)
3623 - remove osc_open hack for echo_client (1187)
3624 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3625 - a failure in read_inode2 leads to deadlock (1139)
3626 - cancel ack-locks as soon as transaction is committed (1072)
3627 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3628 - make sure to commitrw after any preprw to avoid deadlock (1162)
3629 - failing to execute a file in a lustre FS would lock inode (1203)
3630 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3631 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3632 - obd_uuid2tgt would walk off the end of the list (1255)
3633 - on IA64 the osc would give portals incorrect bulk size (1258)
3634 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3635 - fix lock inversion caused by new llite matching code (1282)
3636 - limit the number of dirty pages on a client to 10MB (1286)
3637 - timed out locks were not being corrected cancelled (1289)
3638 - fix O_DIRECT above 4GB on IA-32 (1292)
3639 * major user-visible changes
3640 - fail out/fail over policy now controlled by the upcall (993)
3642 - add OBD_PING to check server availability and failure (954)
3643 - lustre messages are now sent in sending host order (288, 340, 891)
3644 - add eadatalen to MDS getattr reply (340)
3645 - OST read replies may contain second buffer, with per-page status (593)
3647 2003-03-11 Phil Schwan <phil@clusterfs.com>
3650 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3651 - failed open()s could cause deadlock; fixed (867, 869)
3652 - stop cancelling OST locks when files are closed (481)
3653 - overlapping XID spaces caused network corruption (851, 853)
3654 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3655 - setattr_raw added, to do single-RPC, server-side setattrs
3656 - lmc/lconf syntax change for OST UUIDs
3657 - fix crashy race condition between ptlrpc_free_req and osc_close
3658 - don't use request in mdc_enqueue if we hit a timeout (889)
3659 - don't set the inode i_size for regular files from the MDS (896)
3660 - handle out of order completion AST (842)
3661 - don't LBUG if a lock request times out after receiving AST (913)
3662 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3663 - if a bad lock AST arrives, send an error instead of dropping entirely
3664 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3665 - fix leak in bulk IO when only partially completed (899, 900, 926)
3666 - fix O_DIRECT for ia64 (55)
3667 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3668 - C-z after timeout could hang a process forever; fixed (977)
3670 - client-side I/O cache (678, 924, 929, 941, 970)
3672 - READPAGE and SETATTRs which don't take server-side locks get
3675 2003-02-11 Phil Schwan <phil@clusterfs.com>
3678 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3679 - Fix fsfilt_extN_readpage() to read a full page of directory
3680 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3681 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3682 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3683 - Most intent operations were removed, in favour of a new RPC mode
3684 that does a single RPC to the server and bypasses most of the VFS
3685 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3686 - Aggressively cancel local locks on DLM servers
3687 - mds_reint_unlink sends EA to the client if it's the last nlink.
3688 client uses that EA to unlink OST objects.
3689 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3690 - recursive symlinks were fixed (439)
3691 - fixed NULL deref in DEBUG_REQ
3692 - filter_update_lastobjid no longer calls sync, which annoyed extN
3693 - fixed multi-client small-writes to a single file problem (445)
3694 - fixed mtime updates during file writes (607)
3695 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3696 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3697 - fixed Linux OST size reporting problem (444, 656)
3698 - OST now updates object mtime with writes or setattr (607, 619)
3699 - client verifies file size before zeroing page past EOF (445)
3700 - OST now writes last allocated objid to disk with allocation (108)
3701 - LOV on echo now works (409)
3703 - mds_reint_unlink sends a new buffer, with the EA included. this
3704 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3705 set if a regular file was being unlinked, and it was the last link
3706 - use PtlGet from the target for bulk writes (315)
3707 - OST now updates object mtime with writes or setattr (607, 619)
3708 - LDLM now has a grant-time callback to revalidate locked items, if
3710 - Many MDS operations were reorganized to combat race conditions
3712 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3713 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3715 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3718 - Fully reactivate OST imports after reconnection (512, others)
3719 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3720 - More graceful error handling for truncating on dead OST (515)
3721 - Don't error out unless we're actually accessing dead stripes (474)
3722 - Fix garbage sizes when stripes are missing (410)
3723 - LRU counters were broken, causing constant lock purge (433, 432)
3724 - garbage on read from stripes with failed OSTs (441)
3725 - mark OSCs as active before reconnecting during recovery (438)
3726 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3727 - lfind did not preserve OST order in output (443)
3728 - symlinks cause hung clients, incorrect data (439)
3729 - stop dereferencing request after dropping refcount (457)
3730 - don't LASSERT(spin_is_locked) on non-SMP (455)
3731 - fixes for many rename() bugs
3732 - fstat didn't correctly synchronize attributes (399)
3733 - server must handle lock cancellation during blocking AST prep (487)
3734 - bulk descriptors were free()d too soon (511)
3735 - fix paths in lconf, which would load incorrect modules (451, 507)
3736 - fix confusing lconf 'host not found' error message (386)
3737 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3738 - fix race condition in mdc_blocking_ast() for inode access (526)
3739 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3740 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3741 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3742 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3743 - fix lstripe in several ways (kernel side) (527)
3744 - fix request leak in ldlm_cli_enqueue (262)
3745 - incorrect OSC was marked inactive after OST failure
3746 - call mds_fs_cleanup before unmounting filesystem (524)
3747 - fix races between taking ns_lock and ldlm_lock_change_resource
3748 - fix races updating LOV export open file list
3749 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3750 - fix recovery NULL deref in ldlm_cli_cancel_unused
3751 - fix some DLM races by using new hash table for lock handles (419)
3752 - permit the client to specify desired inodes, at replay
3753 - duplicate requests when we queue them for replay reintegration
3754 - fix last_rcvd offset calculation
3755 - sync after each recovered transaction, so we always make progress
3756 - never, not always, ERESTART requests without transnos
3757 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3759 - skip replay if the MDS says that the client is already connected
3760 - don't check for a recovery-enabled export to match lctl's UUID
3761 - don't INC_USE_COUNT for phantom exports
3762 - don't crash when cleaning up phantom exports (567)
3763 - don't double-finish or set replay data for errored mdc_open requests
3764 - abort requests when they time out, so we don't get old replies
3765 - send/receive replies for AST messages again
3766 - if the client says that it doesn't have the lock, cancel it on the
3768 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3769 mark it as destroyed, it will all work out when decref is called
3770 - fix module use counts (22, 581)
3772 - ASTs now expect a reply (server cancels lock on error reply)
3774 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3777 - fix many simultaneous client startup (392)
3778 - fix dentry->d_it clobbering
3779 - credentials weren't being shipped for readdir/getattr operations
3780 - remove invalid assertions triggered during some concurrent MD
3782 - proper Lustre versions added (336, 389)
3783 - fix memory leak for create error case (398)
3784 - fix LOV locking bug that would get cli/srv out of sync
3785 - fix echo client over LOV (409)
3786 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3787 - fix double-O_EXCL intent crash (424)
3788 - avoid sending multiple lock CANCELs (352)
3790 - MDS can do multi-client recovery (modulo bugs in new code)
3792 - many updates, edits, cleanups
3794 2002-11-18 Phil Schwan <phil@clusterfs.com>
3797 - fix null d_it dereference (346)
3798 - fix full OST/dbench hang (333)
3799 - fix permission problem with file removal (286)
3800 - fix removal of OSCs from LOV when they fail
3801 - fix NULL deref during bulk timeout (214)
3802 - fix problems related to multiple filesystems on one MDS (241)
3803 - fixed serious subtle metadata locking bugs
3804 - free locks on clients when inodes are removed due to memory
3806 - fix inode pointer in lock data (285)
3807 - partial support for multiple MDS on a single host (241)
3808 - data locks weren't cancelled at clear_inode time (290, 311)
3809 - intent locks could lead to unbounded lock growth (205)
3810 - added a maximum lock count, an LRU list, and a flusher
3811 - fix multiple rename (365)
3812 - properly abstracted the echo client
3813 - OSC locked 1 byte too many; fixed
3814 - rewrote brw callback code:
3815 - fixed recovery bugs related to LOVs (306)
3816 - fixed too-many-pages-in-one-write crash (191)
3817 - fixed (again) crash in sync_io_timeout (214)
3818 - probably fixed callback-related race (385)
3820 - Add capability to MDS protocol
3821 - LDLM cancellations and callbacks on different portals
3823 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3826 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3827 - defer open object destruction to close time (601981, 138)
3828 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3829 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3830 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3831 - don't call obd_setattr in ll_file_release for destroyed objects
3832 * protocol change to lustre_msg: move |version| and add |flags|
3833 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3834 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3835 * added replay of create, unlink, link and rename operations during
3836 MDS failover; recovery should be much more robust now
3837 * remove failed OSCs from LOVs (only lov_create uses this so far)
3838 * the lustre-HOWTO was brought (more) up to date (582544)
3840 2002-10-23 Phil Schwan <phil@clusterfs.com>
3843 - in-use dentries weren't being reused properly (617851)
3844 - prevent multiple LDLM setup (599178)
3845 - fix LOV size calculations for truncate (617853)
3846 - fix client handling of MDS intent errors (POSIX)
3847 - fix permission bug in lovstripe.c test (624321)
3848 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3849 - truncate past end of file could corrupt data
3850 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3851 - a race in recovery could return ETIMEDOUT to apps (623947)
3852 - building outside the source directory was fixed
3853 * the lustre-HOWTO was brought (more) up to date (582544)
3854 * major progress was made on recovery functionality
3856 2002-10-10 Phil Schwan <phil@clusterfs.com>
3859 - recovery deadlock fix
3860 - rm -rf causes LBUG fix (617817)
3861 - file open by multiple tasks fix (618962)
3862 - directory permissions bugs (602707 and 620007)
3863 - journal_stop fixed with locking (611313)
3864 - O_APPEND failures resolved (618273, perhaps 614459)
3865 - lconf PATH fix (619770)
3866 - IA64 build fix (621450)
3867 - RPC buffer sizes scale with amount of memory
3869 2002-10-01 Phil Schwan <phil@clusterfs.com>
3872 - locks would be cancelled without throwing away data pages,
3873 resulting in inconsistent data (605627)
3874 - inode attributes were not always being refreshed (605627, 612449)
3875 - lconf now continues to cleanup after lctl reports an error
3876 - MDS now enforces user permissions (602707)
3877 - lprocfs cleanup fixed, but not yet enabled (614157)
3878 - fixed infinite server hang, should a client not respond to an AST
3879 - avoid going into recovery if user calls readlink() with a buffer
3880 that's too small (613941)
3881 - AST RPCs no longer require replies (614867) -- this may be changed
3882 - don't crash server if client sends an IOV that's too big (611336)
3883 - fixed lock conversion deadlock (611892)
3884 - fixed the following of symlinks (614622)
3885 * recovery: the server can remove locks from a client that dies, other
3886 clients can make progress
3887 * more extN patch fixes
3888 * compile-time configurable ptlrpc buffer allocations
3890 - collaborative read cache document
3891 - Lustre Lite Performance CDR document-in-progress
3893 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3896 - fix typo in patch-2.4.18
3898 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3901 - clear ptlrpc request each time in handle_incoming_request()
3902 - unlink of files now destroys the object on the OST
3904 2002-09-19 Peter Braam <braam@clusterfs.com>
3906 * add hard link support
3907 * change obdfile creation method
3908 * kernel patch changed
3910 2002-09-19 Peter Braam <braam@clusterfs.com>
3913 - stack overflow bug in extN fixed
3915 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3917 * documentation updates
3918 - add man pages for config tools
3919 - update tests/README to describe testing with new config tools
3920 - finish metadata API descriptions
3921 * bug fixes and cleanups
3922 - statfs workaround for 16TB limit
3923 - LOV stripe allocation improved, can stripe on subset of OSTs
3924 - LOV file size/IO offset was wrong for files > 4GB in size
3925 - object EA data was being dropped, caused files to be unreadable
3926 - memory overflow with non-LOV OST caused memory corruption
3927 - fixed regression tests to work with new config tools, obdfilter
3928 - fixed bug when directory size became larger than 1 block
3929 - fixed bug (for single client case) when PWD was deleted
3930 - invalidate local directory pages when doing intent-based ops
3931 - avoid LDLM oops when lock callback contained bad data
3933 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3935 * documentation updates
3936 * bug fixes and cleanups
3937 - configuration tools
3940 - 64-bit compile warnings
3941 - 64-bit internal statfs data
3943 * test_brw on persistent OST devices
3945 * lprocfs (disabled)
3947 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3949 * documentation updates
3950 * bug fixes and cleanups
3951 * configuration tools
3953 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3956 * many small fixes to 0_5_4
3957 * io/network handling
3958 * thinkos in MDS operations
3960 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3963 * crucial basic fixes to 0.5.3
3964 * IOR, Iozone work over Elan
3967 2002-08-07 Phil Schwan <phil@clusterfs.com>
3968 * version 0_5_3, our first alpha
3969 * we use the new Portals iovs
3970 * documentation updates
3971 * bug fixes and cleanups
3972 * small changes in the DLM wire protocol
3974 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3975 * version 0_5_1 with some initial stability,
3976 * locking on MD and file I/O.
3977 * documentation updates
3978 * several bug fixes since 0.5.0
3979 * small changes in wire protocol
3981 2002-07-18 Phil Schwan <phil@clusterfs.com>
3983 * delivered as Lustre Light Alpha
3984 * fixed a crash after handling invalid MDS requests
3985 * fixed directory pages for architectures with non-4k pages sizes
3987 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3988 * release version v0_4_4
3989 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3991 * Fixes a number of interruption problems with OST operations.
3992 * Update documentation for portals header changes
3993 * Move all wire protocol structs/defines to lustre_idl.h
3994 * Fixes symlink length bug.
3995 * Add tcpdump to repository.
3997 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3998 * release version v0_4_3
3999 * Fixes statfs for inodes on extN.
4000 * Fixes bug in runtests which would delete /etc/hosts.
4001 * Use 64-bit object IDs wherever possible (not into VFS though)
4002 Remove ost_get_info, which is unused by lustre, and out of date.
4004 2002-07-03 Peter Braam <braam@clusterfs.com>
4005 * release version v0_4_2 Fixes a lookup error (type not passed)
4006 * move forward to head of Portals
4007 * move forward to latest Lustre kernel
4009 2002-06-25 Peter Braam <braam@clusterfs.com>
4010 * release version v0_4_1. Hopefully stable on single node use.