1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5 2.6.9-55.0.2.EL (RHEL 4),
6 2.6.16.46-0.14 (SLES 10),
7 2.6.18-8.1.10.el5 (RHEL 5),
8 2.6.18.8 vanilla (kernel.org)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.16 - 2.6.22 vanilla (kernel.org)
12 * Due to recently discovered recovery problems, we do not recommend
13 using patchless RHEL 4 clients with this or any earlier release.
14 * Recommended e2fsprogs version: 1.40.2-cfs1
15 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Severity : enhancement
19 Description: eliminate client locks in face of contention
20 Details : file contention detection and lockless i/o implementation
25 Description: Remove client patches from SLES 10 kernel.
26 Details : This causes SLES 10 clients to behave as patchless clients
27 even on a Lustre-patched (server) kernel.
31 Description: use i_size_read and i_size_write in 2.6 port
32 Details : replace inode->i_size access with i_size_read/write()
35 Frequency : when removing large files
37 Description: scheduling issue during removal of large Lustre files
38 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
39 It causes scheduling issues when removing large files (17TB in the
44 Description: 1.4.11 Can't handle directories with stripe set and extended acls
45 Details : Impossible (EPROTO is returned) to access a directory that has a
46 non-default striping and ACLs.
49 Frequency : only on ppc
51 Description: /proc/fs/lustre/devices broken on ppc
52 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
53 We should make sure the type of 'index' is loff_t and then cast
54 explicitly as needed below. Do not assign an explicitly cast
58 Frequency : only for rhel5
60 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
61 Details : Modify the target file & which_kernel.
64 Frequency : if the uninit_groups feature is enabled on ldiskfs
66 Description: e2fsck reports "invalid unused inodes count"
67 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
68 feature and only a single inode is created in a group then the
69 "bg_unused_inodes" count is incorrectly updated. Creating a
70 second inode in that group would update it correctly.
74 Description: buffer overruns could theoretically occur
75 Details : llapi_semantic_traverse() modifies the "path" argument by
76 appending values to the end of the origin string, and a buffer
77 overrun may occur. Adding buffer overrun check in liblustreapi.
79 --------------------------------------------------------------------------------
81 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
83 * Support for kernels:
85 2.6.9-55.0.2.EL (RHEL 4),
86 2.6.16.46-0.14 (SLES 10),
87 2.6.18-8.1.8.el5 (RHEL 5),
88 2.6.18.8 vanilla (kernel.org)
89 * Client support for unpatched kernels:
90 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
91 2.6.16 - 2.6.22 vanilla (kernel.org)
92 * Due to recently discovered recovery problems, we do not recommend
93 using patchless RHEL 4 clients with this or any earlier release.
94 * Recommended e2fsprogs version: 1.40.2-cfs1
95 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
99 Description: Fix errors in lfs documentation
100 Details : Fixes man pages
102 Severity : enhancement
104 Description: Adaptive timeouts
105 Details : RPC timeouts adapt to changing server load and network
106 conditions to reduce resend attempts and improve recovery time.
108 Severity : enhancement
110 Description: llapi_file_create() does not allow some changes
111 Details : add llapi_file_open() that allows specifying the file creation
112 mode and open flags, and also returns an open file handle.
114 Severity : enhancement
116 Description: df doesn't work properly if diskfs blocksize != 4K
117 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
120 Frequency : always on directories with default striping set
122 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
123 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
125 Severity : enhancement
127 Description: merge and cleanup kernel patches.
128 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
130 Severity : enhancement
132 Description: RedHat Update kernel for RHEL5
133 Details : Kernel config file for RHEL5.
135 Severity : enhancement
137 Description: OSS needs mutliple precreate threads
138 Details : Add ability to start more than one create thread per OSS.
141 Frequency : only with quota on the root user
143 Description: mds_obd_create error creating tmp object
144 Details : When the user sets quota on root, llog will be affected and can't
145 create files and write files.
148 Frequency : Always on ia64 patchless client, and possibly others.
150 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
151 Details : This allows the patchless client to be loaded on architectures
157 Description: disorder of journal start and llog_add cause deadlock.
158 Details : in llog_origin_connect, journal start should happen before
159 llog_add keep the same order as other functions to avoid
162 Severity : enhancement
164 Description: RedHat Update kernel for RHEL5
165 Details : Modify the kernel config file more closer RHEL5.
167 Severity : enhancement
169 Description: Build failure against Centos5 (RHEL5)
170 Details : Define PAGE_SIZE when it isn't present.
173 Frequency : occasionally when using NFS
175 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
176 Details : with NFS, the anon dentry's parent was set to itself in
177 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
178 corresponding dentry other than use rec->ur_name.
180 Severity : enhancement
182 Description: enable data checksumming by default
183 Details : enable checksum by default, allow --disable-checksum
184 configure option and "-o nochecksum" mount option. Checksums
185 can also be disabled at runtime via $LPROC/llite/*/checksum_pages.
188 Frequency : Occasionally with failover
190 Description: Client eviction due to failover config
191 Details : after a connection loss, the lustre client should attempt to
192 reconnect to the last active server first before trying the
193 other potential connections.
195 Severity : enhancement
197 Description: client-side metadata stat-ahead during readdir(directory readahead)
198 Details : perform client-side metadata stat-ahead when the client detects
199 readdir and sequential stat of dir entries therein
204 Description: when mds and osts use different quota unit(32bit and 64bit),
205 quota will be released repeatly.
206 Details : void sending multiple quota reqs to mds, which will keep the
207 status between the reqs.
210 Frequency : only with liblustre clients on XT3
212 Description: evictions taking too long
213 Details : allow llrd to evict clients directly on OSTs
217 Description: osts not allocated evenly to files
218 Details : change the condition to increase offset_idx
220 Severity : enhancement
222 Description: self-adjustable client's lru lists
223 Details : use adaptive algorithm for managing client cached locks lru
224 lists according to current server load, other client's work
225 pattern, memory activities, etc. Both, server and client
226 side namespaces provide number of proc tunables for controlling
231 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
232 Details : rewrite inherited code (uuid parsing code from ext2 utils and
233 readdir code from ext3) from scratch preserving functionality.
237 Description: Only those disconnect error should be returned by rq_status.
238 Details : In open/enqueue processs, Some errors, which will cause client
239 disconnected, should be returned by rq_status, while other
240 errors should still be returned by intent, then mdc or llite will
243 Severity : enhancement
245 Description: Tune the kernel for good SCSI performance.
246 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
247 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
251 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
252 Bugzilla : 13547, 13627
253 Description: Data corruption for OSTs that are formatted larger than 2TB
255 Details : When generating the bio request for lustre file writes the
256 sector number would overflow a temporary variable before being
257 used for the IO. The data reads correctly from Lustre (which
258 will overflow in a similar manner) but other file data or
259 filesystem metadata may be corrupted in some cases.
263 Description: "lfs find -obd UUID" prints directories
264 Details : "lfs find -obd UUID" will return all directory names instead
265 of just file names. It is incorrect because the directories
266 do not reside on the OSTs.
270 Description: MDS hang after unclean shutdown of lots of clients
271 Details : Never resend AST requests.
275 Frequency : Always, for kernels after 2.6.16
276 Description: Fix warning idr_remove called for id=.. which is not allocated.
277 Details : Last kernels save old s_dev before kill super and not allow
278 to restore from callback - restore it before call kill_anon_super.
280 --------------------------------------------------------------------------------
282 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
284 * Support for kernels:
285 2.6.5-7.286 (SLES 9),
286 2.6.9-55.0.2.EL (RHEL 4),
287 2.6.16.46-0.14 (SLES 10),
288 2.6.18-8.1.8.el5 (RHEL 5),
289 2.6.18.8 vanilla (kernel.org)
290 * Client support for unpatched kernels:
291 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
292 2.6.16 - 2.6.22 vanilla (kernel.org)
293 * Due to recently discovered recovery problems, we do not recommend
294 using patchless RHEL 4 clients with this or any earlier release.
295 * Recommended e2fsprogs version: 1.39.cfs8
296 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
301 Description: block reactivating mgc import until all deactivates complete
302 Details : Fix race when failing back MDT/MGS to itself (testing)
304 Severity : enhancement
306 Description: lfs setstripe enhancement
307 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
310 Frequency : mds/oss recovery
312 Description: llog ctxt is refrenced after it has been freed.
313 Details : llog ctxt refcount was added to avoide the race between ctxt free
314 and llog recovery process. Each llog user must hold ctxt refcount
315 before it access the llog. And the llog ctxt can only be freed
316 when its refcount is zero.
318 Severity : enhancement
320 Description: randomly memory allocation failure util
321 Details : Make lustre randomly failed allocating memory for testing purpose.
324 Frequency : only for SLES10
326 Description: Update kernel patch for SLES10 SP1
327 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
329 Severity : enhancement
331 Description: omit set fsid for export NFS
332 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
333 as FS_REQUIRES_DEV for avoid problems with generate fsid.
336 Frequency : after network failures
338 Description: Add sync option to mount_lustre.c
339 Details : Client loses data written to lustre after a network interruption.
341 Severity : enhancement
343 Description: Error message improvement.
344 Details : Merging of two LCONSOLE_ERROR_MSG into one.
346 Severity : enhancement
348 Description: don't use GFP_* in generic Lustre code.
349 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
351 Severity : enhancement
353 Description: obdclass is limited by single OBD_ALLOC(idarray)
354 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
356 Severity : enhancement
358 Description: updated patchess for new RHEL4 kernel
359 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
361 Severity : enhancement
363 Description: warnings with build patchless client with vanila 2.6.19 and up
364 Details : change old ctl_table style and replace ctl_table/ctl_table_header
365 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
367 Severity : enhancement
369 Description: O_DIRECT bypasses client statistics.
370 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
371 read_ahead_stats were updated. Copied stats section from
372 osc_send_oap_rpc() into async_internal().
375 Frequency : only for Cray XT3
377 Description: peer credits not enough on many OST per OSS systems.
378 Details : Use new lnet way to add credits as we need those for pings and ASTs
381 Frequency : only with liblustre
383 Description: Liblustre is not releasing flock locks on file close.
384 Details : Release flock locks on file close.
387 Frequency : only for RHEL4
389 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
390 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
391 Update target file and kernel config.
395 Description: lustre support for RHEL5
396 Details : Add support for RHEL5.
401 Description: ASSERTION(export != NULL) failed in target_handle_connect
402 Details : Assetion hit is result of rare race between disconnect and connet
403 to same nid. target_handle_connect found old connect cockie and
404 tried to reconnect, but can't find export for this cockie.
409 Description: umount blocks forever on error
410 Details : In result of wrong using obd_no_recov and obd_force flags client
411 can hand if cancel or some other requests is lost.
414 Frequency : Only for SLES
416 Description: sanity_quota fail test_1
417 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
418 file, which makes TSTID[2] inunique.
420 Severity : enhancement
422 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
423 Details : Update target/ChangeLog/which_patch .
425 Severity : enhancement
427 Description: jbd statistics
428 Details : Port older jbd statistics patch for sles10
433 Description: lvbo_init failed for resource with missing objects.
434 Details : Fix returning error if we do stat for file with missing/corrupted
435 objects and i_size set to all sum of size of all avaible objects.
436 if we to truncate/write to missing object - it is recreated.
441 Description: Oops in read and write path when failing to allocate lock.
442 Details : Check if lock allocation failed and return error back.
445 Frequency : When flocks are used.
447 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
448 Details : Flock locks might destroy just granted lock if it could be merged
449 with another existing flock, this is done in completion handler,
450 so teach ldlm_cli_enquque_fini that this is a valid case for
456 Description: reply_lock_interpret crash due to race with it and lock cancel.
457 Details : Do not replay locks that are being cancelled. Do not reference
458 locks by their address during replay, just by their handle.
461 Frequency : only with deactivated OSTs
463 Description: lstripe command fails for valid OST index
464 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
465 instead of lov->desc.ld_active_tgt_count.
467 Severity : enhancement
469 Description: Build failure against Centos5 (RHEL5)
470 Details : Use getpagesize() instead of PAGE_SIZE.
472 --------------------------------------------------------------------------------
474 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
476 * Support for kernels:
477 2.6.5-7.283 (SLES 9),
478 2.6.9-55.EL (RHEL 4),
479 2.6.16.46-0.14 (SLES 10),
480 2.6.18.8 vanilla (kernel.org)
481 * Client support for unpatched kernels:
482 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
483 2.6.16 - 2.6.22 vanilla (kernel.org)
484 * Due to recently discovered recovery problems, we do not recommend
485 using patchless RHEL 4 clients with this or any earlier release.
486 * Recommended e2fsprogs version: 1.39.cfs8
487 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
488 * Starting with this release, the ldiskfs backing filesystem required
489 by Lustre is now in its own package, lustre-ldiskfs. This package
490 should be installed. It is versioned separately from Lustre and
491 may be released separately in future.
496 Description: server LBUG when shutting down
497 Details : Block umount forever until the mount refcount is zero rather
498 than giving up after an arbitrary timeout.
500 Severity : enhancement
502 Description: add optional extra BUILD_VERSION info
503 Details : add a new environment variable (namely LUSTRE_VERS) which allows
504 to override the lustre version.
507 Frequency : 2.6.18 servers only
509 Description: ll_kern_mount() doesn't release the module reference
510 Details : The ldiskfs module reference count never drops down to 0
511 because ll_kern_mount() doesn't release the module reference.
516 Description: server LBUG when using old ost_num_threads parameter
517 Details : Accept the old ost_num_threads parameter but warn that it
518 is deprecated, and fix an off-by-one error that caused an LBUG.
523 Description: Transient SCSI error results in persistent IO issue
524 Details : iobuf->dr_error is not reinitialized to 0 between two
528 Frequency : sometimes when underlying device returns I/O errors
530 Description: OSTs not going read-only during write failures
531 Details : OSTs are not remounted read-only when the journal commit threads
532 get I/O errors because fsfilt_ext3 calls journal_start/stop()
533 instead of the ext3 wrappers.
536 Frequency : only with 10000 clients or more
538 Description: poor connect scaling with increasing client count
539 Details : Don't run filter_grant_sanity_check for more than 100 exports
540 to improve scaling for large numbers of clients.
543 Frequency : SLES10 only
545 Description: sanity-quota.sh quotacheck failed: rc = -22
546 Details : Quotas cannot be enabled on SLES10.
549 Frequency : liblustre clients only
551 Description: getdirentries does not give error when run on compute nodes
552 Details : getdirentries does not fail when the size specified as an argument
553 is too small to contain at least one entry
555 Severity : enhancement
557 Description: Add LNET router traceability for debug purposes
558 Details : If a checksum failure occurs with a router as part of the
559 IO path, the NID of the last router that forwarded the bulk data
560 is printed so it can be identified.
565 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
566 Details : Due to a race condition, liblustre clients were occasionally
569 Severity : enhancement
571 Description: lfs setstripe use optional parameters instead of postional
574 Severity : enhancement
576 Description: Nanosecond timestamp support for ldiskfs
577 Details : The on-disk ldiskfs filesystem has added support for nanosecond
578 resolution timestamps. There is not yet support for this at
579 the Lustre filesystem level.
582 Frequency : during server recovery
584 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
585 Details : request with rq_no_resend flag not awake l_wait_event if they get a
589 Frequency : nfs export on patchless client
591 Description: connectathon hang when test nfs export over patchless client
592 Details : Disconnected dentry cannot be found with lookup, so we do not need
593 to unhash it or make it invalid
596 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
597 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
598 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
603 Description: open req refcounting wrong on reconnect
604 Details : If reconnect happened between getting open reply from server and
605 call to mdc_set_replay_data in ll_file_open, we will schedule
606 replay for unreferenced request that we are about to free.
607 Subsequent close will crash in variety of ways.
608 Check that request is still eligible for replay in
609 mdc_set_replay_data().
614 Description: disable writes to filesystem when reading health_check file
615 Details : the default for reading the health_check proc file has changed
616 to NOT do a journal transaction and write to disk, because this
617 can cause reads of the /proc file to hang and block HA state
618 checking on a healthy but otherwise heavily loaded system. It
619 is possible to return to the previous behaviour during configure
620 with --enable-health-write.
622 Severity : enhancement
624 Description: 64-bit inode version
625 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
626 to the inode. This will be required for version-based recovery.
631 Description: MDS fails to start if a duplicate client export is detected
632 Details : in some rare cases it was possible for a client to connect to
633 an MDS multiple times. Upon recovery the MDS would detect this
634 and fail during startup. Handle this more gracefully.
636 Severity : enhancement
638 Description: Add -o localflock option to simulate old noflock behaviour.
639 Details : This will achieve local-only flock/fcntl locks coherentness.
644 Description: log_commit_thread vs filter_destroy race leads to crash
645 Details : Take import reference before releasing llog record semaphore
650 Description: Wrong request locking in request set processing
651 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
652 imp_delayed_list, in this place should be used imp_lock.
655 Frequency : when reconnecting
657 Description: Grant leak when OSC reconnect to OST
658 Details : When osc reconnect ost, OST(filter) should check whether it
659 should grant more space to client by comparing fed_grant and
660 cl_avail_grant, and return the granted space to client instead
661 of "new granted" space, because client will call osc_init_grant
662 to update the client grant space info.
665 Frequency : when client reconnects to OST
667 Description: Grant leak when OSC does a resend and replays bulk write
668 Details : When osc reconnect to OST, OST(filter) should clear grant info of
669 bulk write request, because the grant info while be sync between
670 OSC and OST when reconnect, and we should ignore the grant info
671 these of resend/replay write req.
676 Description: Grant space more than avaiable space sometimes.
677 Details : When then OST is about to be full, if two bulk writing from
678 different clients came to OST. Accord the avaliable space of the
679 OST, the first req should be permitted, and the second one
680 should be denied by ENOSPC. But if the seconde arrived before
681 the first one is commited. The OST might wrongly permit second
682 writing, which will cause grant space > avaiable space.
685 Frequency : when client is evicted
687 Description: Grant might be wrongly erased when osc is evicted by OST
688 Details : when the import is evicted by server, it will fork another
689 thread ptlrpc_invalidate_import_thread to invalidate the
690 import, where the grant will be set to 0. While the original
691 thread will update the grant it got when connecting. So if
692 the former happened latter, the grant will be wrongly errased
693 because of this race.
698 Description: Checking Stale with correct fid
699 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
700 to check whether it is stale, because sometimes, we want the
701 enqueue happened anyway, and op_data.fid2 will not be initialized.
703 Severity : enhancement
705 Description: update patchless client
706 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
709 Frequency : only with 2.4 kernel
711 Description: random memory corruption
712 Details : size of struct ll_inode_info is to big for union inode.u and this
713 can be cause of random memory corruption.
718 Description: Memory leak in recovery
719 Details : Lov_mds_md was not free in an error handler in mds_create_object.
720 It should also check obd_fail before fsfilt_start, otherwise if
721 fsfilt_start return -EROFS,(failover mds during mds recovery).
722 then the req will return with repmsg->transno = 0 and rc = EROFS.
723 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
724 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
725 be freed no matter whether fsfilt_commit success or not.
728 Frequency : only with huge numbers of clients
730 Description: Prevents from taking the superblock lock in llap_from_page for
732 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
733 for llap_from_page call in ll_removepage() prevents from taking
734 the superblock lock for a soon died page.
739 Description: Not check open intent error before release open handle
740 Details : in some rare cases, the open intent error is not checked before
741 release open handle, which may cause
742 ASSERTION(open_req->rq_transno != 0), because it tries to release
743 the failed open handle.
748 Description: Set cat log bitmap only after create log success.
749 Details : in some rare cases, the cat log bitmap is set too early. and it
750 should be set only after create log success.
754 Description: Accessing a block bevice can re-enable I/O when Lustre is
755 tearing down a device.
756 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
761 Description: service threads may hog cpus when there are a lot of requests
762 Details : Insert cond_resched to give other threads a chance to use some CPU
767 Description: the cat log was not initialized in recovery
768 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
769 unlink log on mds will not be initialized until mds post
770 recovery. And also in mds post recovery, the unlink log will
771 initialization will be done asynchronausly, so there will be race
772 between add unlink log and unlink log initialization.
776 Description: brw_stats were being printed incorrectly
777 Details : brw_stats were being printed as log2 but all of them were not
778 recorded as log2. Also remove some code duplication arising from
779 filter_tally_{read,write}.
783 Frequency : rare, only in recovery.
784 Description: ASSERTION(req->rq_type != LI_POISON) failed
785 Details : imp_lock should be held while iterating over imp_sending_list for
786 prevent destroy request after get timeout in ptlrpc_queue_wait.
790 Description: replay-single.sh test 52 fails
791 Details : A lock's skiplist need to be cleanup when it being unlinked
792 from its resource list.
797 Description: Short directio read returns full requested size rather than
799 Details : Direct I/O operations should return actual amount of bytes
800 transferred rather than requested size.
802 Severity : enhancement
804 Description: metadata RPC reduction (e.g. for rm performance)
805 Details : decrease the amount of synchronous RPC between clients and servers
806 by canceling conflicing lock before the operation on the client
807 and packing thier handles into the main operation RPC to server.
809 Severity : enhancement
811 Description: add #ifdef HAVE_KERNEL_CONFIG_H
812 Details : kernels from 2.6.19 not need include linux/config.h, but add
813 include linux/autoconf.h in commpiler command line.
815 Severity : enhancement
817 Description: patchless client support for 2.6.22 kernel
818 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
819 removed. In this case we need drop using os depended interface to
820 kmem_cache and use cfs_mem_cache API.
825 Description: fix mal-formatted messages
826 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
830 Frequency : always in liblustre
831 Description: wrong IS_ERR implementation in liblustre.h
832 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
837 Description: Correct condition for output debug message.
838 Details : inode i_nlink equal zero is not enough for output message about
839 disk corruption, i_ctime and i_mode should be also checked.
843 Frequency : always in patchless client
844 Description: add configure check for truncate_complete_page
845 Details : improve checks for exported symbols. This allow run check without
846 sources, but with Module.symvers shipped with kernel distribution.
847 add check for truncate_complete_page used by patchless client.
851 Description: sanity.sh test_77h fails with "test_77h file compare failed"
852 Details : test_77h uses a file which was messed by other test case.
856 Description: Not Check whether lov_tgts is NULL in some lov functions
857 Details : Checking whether lov_tgts is NULL in some functions.
861 Description: replace obdo_alloc() with OBDO_ALLOC macro
862 Details : nothing special is done in obdo_alloc() function, and for
863 debugging purpose, it needs to be replaced with macros.
867 Description: bad return value and errno from fcntl call
868 Details : In liblustre API, errno should be a negative value if error
873 Description: ptlrpc_check_set() LBUG
874 Details : In case of positive reply from server and failed client bulk
875 callback after bulk transfer shouldn't LBUG, but process this
876 request as erroneous.
878 Severity : enhancement
880 Description: VFS operations stats tool.
881 Details : Tool which collects stats by tracking value written in pid,
882 ppid, gid and uses llstat to generate output to plot graph using
884 Updated lustre/utils/Makefile.am
885 Added lustre/utils/ltrack_stats.c
887 Severity : enhancement
889 Description: 2.6.18 server support (lustre 1.6.1)
890 Details : Support for 2.6.18 kernels on the server side.
895 Description: ASSERTION(imp->imp_conn_current) failed
896 Details : an assertion failure is hit if a client node boots and attempts to
897 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
900 Frequency : only for i686
902 Description: 1.4.11 RC1 build fails for RHEL 4, i686
903 Details : Fixed config variable for build.
908 Description: Updated patchess for new RHEL4 kernel
909 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
910 Updated patch jbd-stats-2.6.9.patch
911 Updated patch qsnet-rhel4-2.6.patch
912 Updated patch quota-deadlock-on-pagelock-core.patch
913 Updated patch vfs_intent-2.6-rhel4.patch
914 Updated patch vfs_races-2.6-rhel4.patch
915 Updated series file 2.6-rhel4-titech.series
916 Updated series file 2.6-rhel4.series
917 Updated kernel config files
922 Description: lquota slave complains LBUG when reconnecting with mds
924 Details : quota slave depends qctxt->lqc_import to send its quota request.
925 This pointer will be invalid if mds did failover or broke its
926 connect to osts, which leads to LBUG.
929 Frequency : when qunit size is too small(less than 20M)
931 Description: write is stopped by improper -EDQUOT
932 Details : If the master is busy and qunit size is small enough(let's say 1M),
933 the slave can not get quota from master on time, which will lead
934 slave to trigger a -EQUOTA to client.
939 Description: Deadlock during metadata tests
940 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
941 called with the per-dentry lock held.
944 Frequency : SLES9 only
946 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
947 Details : set CONFIG_NR_CPUS to 128 instead of 8.
949 Severity : enhancement
951 Description: remove fs_prep_san_write operation and related patches
952 Details : remove the ext3-san-jdike patches which are no longer useful.
957 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
958 Details : a disk corruption can cause the mballoc code to assert on a
959 double free or other extent corruptions. Handle these with
960 ext3_error() instead of assertions.
963 Frequency : only with mballoc3 code and deep extent trees
965 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
966 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
967 can cause the filesystem to be remounted read-only.
972 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
973 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
974 few external users of this ioctl number is lfsck's call to
975 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
976 Add the old ioctl number to the handler so both old and new
981 Description: parallel lock callbacks
982 Details : Instead of sending blocking and completion callbacks as separated
983 requests, adding them to a set and sending in parallel.
987 Description: Disable most debugging by default
988 Details : To improve performance, disable most logging (for debug purposes)
989 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
990 default, and HA includes fewer messages.
994 Frequency : only run on patchless client.
995 Description: use do_facet on sanity.sh for test handling recoverables errors
996 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
1000 Description: hash tables for lists of nids, connections and uuids
1001 Details : Hash tables noticeably help when a lot of clients connect to a
1002 server, to faster identify duplicate connections or reconnects,
1003 also to faster find export to evict in manual eviction case.
1007 Description: Sometimes, when the server evict a client, and the client will
1008 not be evicted as soon as possible.
1009 Details : In enqueue req, the error was returned by intent, instead of
1010 rq_status which make ptlrpc layer not detect this error, and
1011 does not evict the client. So enqueue error should be returned
1015 Frequency : only at startup
1017 Description: Delay client connections to MDT until fist MDT->OST connect
1018 Details : If a client tried to create a new file before the MDT had
1019 connected to any OSTs, the create would return EIO. Now
1020 the client will simply block until the MDT connects to the
1021 first OST and the create can succeed.
1024 Frequency : only for SLES9
1026 Description: Routinely utilize latest Quadrics drivers in CFS releases
1027 Details : Update patch qsnet-suse-2.6.patch.
1030 Frequency : only for sles10
1032 Description: Update patches for SLES 10 SP1 kernel.
1033 Details : Update the patch vfs_intent-2.6-sles10.patch.
1038 Description: Routinely utilize latest Quadrics drivers in CFS releases
1039 Details : Update patch qsnet-rhel4-2.6.patch.
1042 Frequency : at statup only
1044 Description: mds_lov_synchronize race leads to various problems
1045 Details : simultaneous MDT->OST connections at startup can cause the
1046 sync to abort, leaving the OSC in a bad state.
1051 Description: Using wrong pointer in osc_brw_prep_request
1052 Details : Access to array[-1] can produce panic if kernel compiled with
1053 CONFIG_PAGE_ALLOC enabled
1055 Severity : enhancement
1057 Description: Async OSC create to avoid the blocking unnecessarily.
1058 Details : If a OST has no remain object, system will block on the creating
1059 when need to create a new object on this OST. Now, ways use
1060 pre-created objects when available, instead of blocking on an
1061 empty osc while others are not empty. If we must block, we block
1062 for the shortest possible period of time.
1066 Frequency : only in recovery
1067 Description: Mark OST as early accessible if his start SYNC.
1068 Details : osc_precreate return flag early accessible if oscc marked as
1069 OSCC_FLAG_SYNC_IN_PROGRESS.
1074 Description: Sometimes precreate code can triger create object on wrong ost
1075 Details : Wrong protected or not not restored variables aftre precreate loop
1076 can produce creation object on wrong ost.
1079 Frequency : oss recovery
1081 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1083 Details : llog_commit_thread_count should be synced between llog_commit
1084 start and cleanup, so new llog_commit thread should not be started
1085 when llog_commit threads being stopped to avoid accessing some
1088 Severity : enhancement
1090 Description: Add printing inode info into message about error in writepage.
1092 --------------------------------------------------------------------------------
1094 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1099 Frequency : on some architectures
1101 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1102 Details : Uninitialized flags sometimes cause configuration commands to
1106 Frequency : patchless clients only
1108 Description: missing __iget() symbol export
1109 Details : The __iget() symbol export is missing. To avoid the need for
1110 this on patchless clients the deathrow inode reaper is turned
1111 off, and we depend on the VM to clean up old inodes. This
1112 dependency was during via the fix for bug 12181.
1117 Description: sanity.sh fail: test_52b
1118 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1119 incorrect inode's flags to client.
1121 --------------------------------------------------------------------------------
1123 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1125 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1126 INTEROPERATE with older versions automatically. In many cases a
1127 special upgrade step is needed. Please read the
1128 user documentation before upgrading any part of a 1.4.x system.
1129 * WARNING: Lustre configuration and startup changes are required with
1130 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1132 * Support for kernels:
1133 2.4.21-47.0.1.EL (RHEL 3),
1134 2.6.5-7.283 (SLES 9),
1135 2.6.9-42.0.10.EL (RHEL 4),
1136 2.6.12.6 vanilla (kernel.org),
1137 2.6.16.27-0.9 (SLES10)
1138 * Client support for unpatched kernels:
1139 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1140 2.6.16 - 2.6.19 vanilla (kernel.org),
1141 2.6.9-42.0.8EL (RHEL 4)
1142 * Recommended e2fsprogs version: 1.39.cfs6
1143 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1146 Severity : enhancement
1148 Description: Async OSC create to avoid the blocking unnecessarily.
1149 Details : If a OST has no remain object, system will block on the creating
1150 when need to create a new object on this OST. Now, ways use
1151 pre-created objects when available, instead of blocking on an
1152 empty osc while others are not empty. If we must block, we block
1153 for the shortest possible period of time.
1155 Severity : enhancement
1157 Description: MountConf
1158 Details : Lustre configuration is now managed via mkfs and mount
1159 commands instead of lmc and lconf. New obd types (MGS, MGC)
1160 are added for dynamic configuration management. See
1161 https://mail.clusterfs.com/wikis/lustre/MountConf for
1164 Severity : enhancement
1166 Description: dynamic OST addition
1167 Details : OSTs can now be added to a live filesystem
1169 Severity : enhancement
1171 Description: startup order invariance
1172 Details : MDTs and OSTs can be started in any order. Clients only
1173 require the MDT to complete startup.
1175 Severity : enhancement
1177 Description: parallel, asynchronous orphan cleanup
1178 Details : orphan cleanup is now performed in separate threads for each
1179 OST, allowing parallel non-blocking operation.
1181 Severity : enhancement
1183 Description: optimized stripe assignment
1184 Details : stripe assignments are now made based on ost space available,
1185 ost previous usage, and OSS previous usage, in order to try
1186 to optimize storage space and networking resources.
1188 Severity : enhancement
1190 Description: Permanently set tunables
1191 Details : All writable /proc/fs/lustre tunables can now be permanently
1192 set on a per-server basis, at mkfs time or on a live
1195 Severity : enhancement
1197 Description: Lustre message v2
1198 Details : Add lustre message format v2.
1200 Severity : enhancement
1202 Description: client OST exclusion list
1203 Details : Clients can be started with a list of OSTs that should be
1204 declared "inactive" for known non-responsive OSTs.
1208 Description: ENOENT returned for valid filehandle during dbench.
1209 Details : Check if a directory has children when invalidating dentries
1210 associated with an inode during lock cancellation. This fixes
1211 an incorrect ENOENT sometimes seen for valid filehandles during
1212 testing with dbench.
1215 Frequency : SFS test only (otherwise harmless)
1217 Description: SPEC SFS validation failure on NFS v2 over lustre.
1218 Details : Changes the blocksize for regular files to be 2x RPC size,
1219 and not depend on stripe size.
1221 Severity : enhancement
1223 Description: fine-grained SMP locking inside DLM
1224 Details : Improve DLM performance on SMP systems by removing the single
1225 per-namespace lock and replace it with per-resource locks.
1227 Severity : enhancement
1229 Description: don't hold multiple extent locks at one time
1230 Details : To avoid client eviction during large writes, locks are not
1231 held on multiple stripes at one time or for very large writes.
1232 Otherwise, clients can block waiting for a lock on a failed OST
1233 while holding locks on other OSTs and be evicted.
1235 Severity : enhancement
1237 Description: Multiple MD RPCs in flight.
1238 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1239 To avoid overly-overloading MDT, introduce a limit on number of
1240 MDT RPCs in flight for a single client and add /proc controls
1241 to adjust this limit.
1243 Severity : enhancement
1245 Description: client read/write statistics
1246 Details : Add client read/write call usage stats for performance
1247 analysis of user processes.
1248 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1249 file access. extents_stats shows chunk size distribution.
1250 extents_stats_per_process show chunk size distribution per
1253 Severity : enhancement
1255 Description: per-client statistics on server
1256 Details : Add ldlm and operations statistics for each client in
1257 /proc/fs/lustre/mds|obdfilter/*/exports/
1259 Severity : enhancement
1261 Description: improved MDT statistics
1262 Details : Add detailed MDT operations statistics in
1263 /proc/fs/lustre/mds/*/stats
1265 Severity : enhancement
1267 Description: VFS operations stats
1268 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1269 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1274 Description: Fix client-side osc byte counters
1275 Details : The osc read/write byte counters in
1276 /proc/fs/lustre/osc/*/stats are now working
1279 Frequency : always as root on SLES
1281 Description: Failure of copying files with lustre special EAs.
1282 Details : Client side always return success for setxattr call for lustre
1283 special xattr (currently only "trusted.lov").
1288 Description: Refcount LNET uuids
1289 Details : The global LNET uuid list grew linearly with every startup;
1290 refcount repeated list entries instead of always adding to
1293 Severity : enhancement
1295 Description: Dynamic service threads
1296 Details : Within a small range, start extra service threads
1297 automatically when the request queue builds up.
1300 Frequency : mixed-endian client/server environments
1302 Description: mixed-endian crashes
1303 Details : The new msg_v2 system had some failures in mixed-endian
1306 Severity : enhancement
1308 Description: Easy OST removal
1309 Details : OSTs can be permanently deactivated with e.g. 'lctl
1310 conf_param lustre-OST0001.osc.active=0'
1312 Severity : enhancement
1314 Description: MGS proc entries
1315 Details : Added basic proc entries for the MGS showing what filesystems
1318 Severity : enhancement
1320 Description: provide MGS failover
1321 Details : Added config lock reacquisition after MGS server failover.
1323 Severity : enhancement
1325 Description: add Linux 2.4 support
1326 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1330 Description: a large application tries to do I/O to the same resource and dies
1331 in the middle of it.
1332 Details : Check the req->rq_arrival time after the call to
1333 ost_brw_lock_get(), but before we do anything about
1334 processing it & sending the BULK transfer request. This
1335 should help move old stale pending locks off the queue as
1336 quickly as obd_timeout.
1339 Frequency : when an incorrect nid is specified during startup
1341 Description: ptlrpc connect to non-existant node causes kernel crash
1342 Details : LNET can't be re-entered from an event callback, which
1343 happened when we expire a message after the export has been
1344 cleaned up. Instead, hand the zombie cleanup off to another
1347 Severity : enhancement
1349 Description: plain/inodebits lock performance improvement
1350 Details : Grouping plain/inodebits in granted list by their request modes
1351 and bits policy, thus improving the performance of search through
1355 Frequency : only if OST filesystem is corrupted
1357 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1358 Details : for a short time RPCs with bulk IO are in the replay list,
1359 but replay of bulk IOs is unimplemented. If the OST filesystem
1360 is corrupted due to disk cache incoherency and then replay is
1361 started it is possible to trip an assertion. Avoid putting
1362 committed RPCs into the replay list at all to avoid this issue.
1365 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1367 Description: System hang on startup
1368 Details : This bug allowed the liblustre (e.g. catamount) client to
1369 return to the app before handling all startup RPCs. This
1370 could leave the node unresponsive to lustre network traffic
1371 and manifested as a server ptllnd timeout.
1373 Severity : enhancement
1375 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1376 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1377 boolean to control whether to print peer debug info when a
1378 client's RPC times out.
1381 Frequency : only for kernels with patches from Lustre below 1.4.3
1383 Description: Remove old rdonly API
1384 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1387 Frequency : only for devices with external journals
1389 Description: Set external device read-only also
1390 Details : During a commanded failover stop, we set the disk device
1391 read-only while the server shuts down. We now also set any
1392 external journal device read-only at the same time.
1395 Frequency : when upgrading from 1.4 while trying to change parameters
1397 Description: The wrong (new) MDC name was used when setting parameters for
1398 upgraded MDT's. Also allows changing of OSC (and MDC)
1399 parameters if --writeconf is specified at tunefs upgrade time.
1402 Frequency : when setting specific ost indicies
1404 Description: QOS code breaks on skipped indicies
1405 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1406 created with --index need not be sequential.
1408 Severity : enhancement
1410 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1411 Details : The uninit_groups feature works in conjunction with the kernel
1412 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1413 up the pass1 processing of e2fsck. This is a read-only feature
1414 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1415 mount filesystems that have had this feature enabled.
1417 Severity : enhancement
1419 Description: Improve multi-block allocation algorithm to avoid fragmentation
1420 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1421 allocation locality and avoid filesystem fragmentation.
1423 ------------------------------------------------------------------------------
1425 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1427 * Support for kernels:
1428 2.4.21-47.0.1.EL (RHEL 3)
1429 2.6.5-7.283 (SLES 9)
1430 2.6.9-42.0.10.EL (RHEL 4)
1431 2.6.12.6 vanilla (kernel.org)
1432 2.6.16.27-0.9 (SLES 10)
1433 * Recommended e2fsprogs version: 1.39.cfs5
1435 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1439 Frequency : occasional, depends on client load and configuration
1440 Bugzilla : 12181, 12203
1441 Description: data loss for recently-modified files
1443 Details : In some cases it is possible that recently written or created
1444 files may not be written to disk in a timely manner (this should
1445 normally be within 30s unless client IO load is very high).
1446 The problem appears as zero-length files or files that are a
1447 multiple of 1MB in size after a client crash or client eviction
1448 that are missing data at the end of the file.
1450 This problem is more likely to be hit on clients where files are
1451 repeatedly created and unlinked in the same directory, clients
1452 have a large amount of RAM, have many CPUs, the filesystem has
1453 many OSTs, the clients are rebooted frequently, and/or the files
1454 are not accessed by other nodes after being written.
1456 The presence of the problem can be detected by looking at
1457 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1458 smaller than the second (nr_unused) then dirty files will not
1459 be flushed automatically to disk. "sync; sleep 10" should be
1460 run several times on the node before unmounting it to update
1461 Lustre (this is also safe to run on nodes without this problem).
1463 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1464 that can cause this same problem, so customers using that kernel
1465 also need to update the kernel in addition to Lustre. In order
1466 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1468 It is normal that files written just before a client crash (less
1469 than 30s) may not yet have been flushed to disk, even for local
1473 Frequency : frequent on thin XT3 nodes
1475 Description: UUID collision on thin XT3 Linux nodes
1476 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1477 randomly, since we relied on an insufficiently-seeded PRNG.
1482 Description: OSS hangs after "All ost request buffers busy"
1483 Details : A deadlock between quota and journal operations caused OSS
1484 hangs after printing "All ost request buffers busy."
1487 Frequency : always on liblustre builds
1489 Description: Cleanup compiler warnings on liblustre
1492 Frequency : always on liblustre builds on XT3
1494 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1498 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1499 > 32000 subdirectories
1500 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1501 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1502 subdirectory count crosses 32000. This will aid e2fsck to
1503 correctly handle more than 32000 subdirectories.
1506 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1508 Description: System hang on startup
1509 Details : This bug allowed the liblustre (e.g. catamount) client to
1510 return to the app before handling all startup RPCs. This
1511 could leave the node unresponsive to lustre network traffic
1512 and manifested as a server ptllnd timeout.
1514 Severity : enhancement
1516 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1517 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1518 boolean to control whether to print peer debug info when a
1519 client's RPC times out.
1524 Description: make O_SYNC working on 2.6 kernels
1525 Details : 2.6 kernels use different method for mark pages for write,
1526 so need add a code to lustre for O_SYNC work.
1531 Description: Failure to close file and release space on NFS
1532 Details : Put inode details into lock acquired in ll_intent_file_open.
1533 Use mdc_intent_lock in ll_intent_open to properly
1534 detect all kind of errors unhandled by mdc_enqueue.
1539 Description: proc file read during shutdown sometimes raced obd removal,
1541 Details : Add lock to prevent obd access after proc file removal.
1544 Frequency : Only for files larger than 4GB on 32-bit clients.
1546 Description: improperly doing page alignment of locks
1547 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1548 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1553 Description: RPCs being resent when they shouldn't be
1554 Details : Some RPCs that should not be resent are being resent. This
1555 can cause inconsistencies in the RPC state machine. Do not
1556 resend such requests.
1559 Frequency : rare, only with NFS export
1561 Description: Crash on NFS re-export node
1562 Details : under very unusual load conditions an assertion is hit in
1563 ll_intent_file_open()
1566 Frequency : only if OST filesystem is corrupted
1568 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1569 Details : for a short time RPCs with bulk IO are in the replay list,
1570 but replay of bulk IOs is unimplemented. If the OST filesystem
1571 is corrupted due to disk cache incoherency and then replay is
1572 started it is possible to trip an assertion. Avoid putting
1573 committed RPCs into the replay list at all to avoid this issue.
1578 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1579 Details : Large single O_DIRECT read and write calls can fail to allocate
1580 a sufficiently large buffer to process the request. In case of
1581 allocation failure the allocation is retried with a smaller
1582 buffer and broken into smaller requests.
1584 Severity : enhancement
1586 Description: Add -o localflock option to simulate old noflock behaviour.
1587 Details : This will achieve local-only flock/fcntl locks coherentness.
1592 Description: versioning check is incomplete
1593 Details : Checking the version difference of client vs. server, report
1594 error if the gap is too big.
1599 Description: add support PG_writeback bit
1600 Details : add support for PG_writeback bit for Lustre, for more carefull
1601 work with page cache in 2.6 kernel. This also fix some deadlocks
1602 and remove hack for work O_SYNC with 2.6 kernel.
1604 Severity : enhancement
1606 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1607 Details : The uninit_groups feature works in conjunction with the kernel
1608 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1609 up the pass1 processing of e2fsck. This is a read-only feature
1610 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1611 mount filesystems that have had this feature enabled.
1613 Severity : enhancement
1615 Description: Improve multi-block allocation algorithm to avoid fragmentation
1616 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1617 allocation locality and avoid filesystem fragmentation.
1619 ------------------------------------------------------------------------------
1621 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1623 * Support for kernels:
1624 2.6.9-42.0.3.EL (RHEL 4)
1625 2.6.5-7.276 (SLES 9)
1626 2.4.21-47.0.1.EL (RHEL 3)
1627 2.6.12.6 vanilla (kernel.org)
1628 2.6.16.21-0.8 (SLES10)
1629 * Recommended e2fsprogs version: 1.39.cfs2-0
1631 * The backwards-compatible /proc/sys/portals symlink has been removed
1632 in this release. Before upgrading, please ensure that you change
1633 any configuration scripts or /etc/sysctl.conf files that access
1634 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1635 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1636 in advance of the upgrade on any system running Lustre 1.4.6 or
1637 newer, since /proc/sys/lnet was added in that version.
1638 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1642 Frequency : only when quota is used
1644 Description: avoid scanning export list for quota master
1645 Details : Change the algorithms to avoid scanning export list in order
1646 to improve the efficiency.
1649 Frequency : MDS failover only, very rarely
1651 Description: "went back in time" messages on mds failover
1652 Details : The greatest transno may be lost when the current operation
1653 finishes with an error (transno==0) and the client's last_rcvd
1654 record is over-written. Save the greatest transno in the
1655 mds_last_transno for this case.
1658 Frequency : always for specific kernels and striping counts
1660 Description: client may get "Matching packet too big" without ACL support
1661 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1662 when trying to access files in certain configurations. The
1663 clients should in fact be denied when mounting because they do
1664 not understand ACLs.
1667 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1669 Description: many clients connecting with IO in progress causes connect timeouts
1670 Details : Avoid synchronous journal commits to avoid delays caused by many
1671 clients connecting/disconnecting when bulk IO is in progress.
1672 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1673 OST_IO_PORTAL to avoid delays behind potentially many pending
1677 Frequency : occasionally with multiple writers to a single file
1679 Description: shared writes to file may result in wrong size reported by stat()
1680 Details : Allow growing of kms when extent lock is cancelled
1683 Frequency : always with random mmap IO to multi-striped file
1685 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1686 Details : Only if the hole is at the end of OST object so that kms is too
1687 small. Fix is to increase kms accordingly in ll_nopage.
1690 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1692 Description: writes to a missing object would leak memory on the OST
1693 Details : If there is an inconsistency between the MDS and OST filesystems,
1694 such that the MDS references an object that doesn't exist, writes
1695 to that object will leak memory due to incorrect cleanup in the
1696 error handling path, eventually running out of memory on the OST.
1701 Description: Creating too long symlink causes lustre errors
1702 Details : Check symlink and name lengths before sending requests to MDS.
1705 Frequency : only if flock is enabled (not on by default)
1707 Description: posix locks not released on fd closure on 2.6.9+
1708 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1709 kernels, this caused such locks not to be released on fd close and
1710 then assertions on fs unmount about still used locks.
1713 Frequency : MDS failover only, very rarely
1715 Description: clients may get ASSERTION(granted_lock != NULL)
1716 Details : When request was taking a long time, and a client was resending
1717 a getattr by name lock request. The were multiple lock requests
1718 with the same client lock handle and
1719 mds_getattr_name->fixup_handle_for_resent_request found one of the
1720 lock handles but later failed with ASSERTION(granted_lock != NULL).
1725 Description: handle->h_buffer_credits > 0, assertion failure
1726 Details : h_buffer_credits is zero after truncate, causing assertion
1727 failure. This patch extends the transaction or creates a new
1731 Frequency : NFS re-export or patchless client
1732 Bugzilla : 11179, 10796
1733 Description: Crash on NFS re-export node (__d_move)
1734 Details : We do not want to hash the dentry if we don't have a lock.
1735 But if this dentry is later used in d_move, we'd hit uninitialised
1736 list head d_hash, so we just do this to init d_hash field but
1737 leave dentry unhashed.
1740 Frequency : NFS re-export or patchless client
1742 Description: NFS exports has problem with symbolic link
1743 Details : lustre client didn't properly install dentry when re-exported
1744 to NFS or running patchless client.
1747 Frequency : NFS re-export or patchless client
1749 Description: Various nfs/patchless fixes.
1750 Details : fixes reuse disconected alias for lookup process - this fixes
1751 warning "find_exported_dentry: npd != pd",
1752 fix permission error with open files at nfs.
1753 fix apply umask when do revalidate.
1756 Frequency : occasional
1758 Description: Crash on NFS re-export node
1759 Details : calling clear_page() on the wrong pointer triggered oops in
1760 generic_mapping_read().
1763 Frequency : rarely, using O_DIRECT IO
1765 Description: unaligned directio crashes client with LASSERT
1766 Details : check for unaligned buffers before trying any requests.
1769 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1771 Description: stale data returned from RAID cache
1772 Details : If only a small amount of IO is done to the RAID device before
1773 reading it again it is possible to get stale data from the RAID
1774 cache instead of reading it from disk.
1777 Frequency : always for sles10 kernel
1779 Description: sles10 support
1780 Details : ll_follow_link: compile fixes and using of nd_set_link
1781 under newer kernels.
1784 Frequency : depends on arch, kernel and compiler version, always on sles10
1787 Description: recursive or deep enough symlinks cause stack overflow
1788 Details : getting rid of large stack-allocated variable in
1792 Frequency : depends on hardware
1794 Description: lustre write performance loss in the SLES10 kernel
1795 Details : the performance loss is caused by using of write barriers in the
1796 ext3 code. The SLES10 kernel turns barrier support on by
1797 default. The fix is to undo that change for ldiskfs.
1799 ------------------------------------------------------------------------------
1801 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1803 * Support for kernels:
1804 2.6.9-42.0.3EL (RHEL 4)
1805 2.6.5-7.276 (SLES 9)
1806 2.4.21-47.0.1.EL (RHEL 3)
1807 2.6.12.6 vanilla (kernel.org)
1811 Frequency : quota enabled and large files being deleted
1813 Description: releasing more than 4GB of quota at once hangs OST
1814 Details : If a user deletes more than 4GB of files on a single OST it
1815 will cause the OST to spin in an infinite loop. Release
1816 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1821 Description: statfs data retrieved from /proc may be stale or zero
1822 Details : When reading per-device statfs data from /proc, in the
1823 {kbytes,files}_{total,free,avail} files, it may appear
1824 as zero or be out of date.
1827 Frequency : systems with MD RAID1 external journal devices
1829 Description: lconf's call to blkid is confused by RAID1 journal devices
1830 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1831 of returning all block devices that match the journal UUID.
1834 Frequency : always, for aggregate stripe size over 4GB
1836 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1837 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1838 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1841 Frequency : always if "lfs find" used on a local file/directory
1843 Description: "lfs find" segfaults if used on a local file/directory
1844 Details : The case where a directory component was not specified wasn't
1845 handled correctly. Handle this properly.
1848 Frequency : always on ppc64
1850 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1851 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1852 perform endianness conversion.
1855 Frequency : rarely (truncate to non-zero file size after write under load)
1856 Bugzilla : 10730, 10687
1857 Description: Files padded with zeros to next 4K multiple
1858 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1859 it is possible that files that are truncated to a non-zero size
1860 immediately after being written are filled with zero bytes beyond
1861 the truncated size. No file data is lost.
1863 Severity : enhancement
1865 Description: Allow recovery/failover for liblustre clients.
1866 Details : liblustre clients were unaware of failover configurations until
1869 Severity : enhancement
1871 Description: user file locks should fail when not mounting with flock option
1872 Details : Set up an error-returning stub in ll_file_operations.lock field
1873 to prevent incorrect behaviour when client is mounted without
1874 flock option. Also, set up properly f_op->flock field for
1878 Frequency : always on ia64
1880 Description: "lfs df" loops on printing out MDS statfs information
1881 Details : The obd_ioctl_data was not initialized and in some systems
1882 this caused a failure during the ioctl that did not return
1883 an error. Initialize the struct and return an error on failure.
1886 Frequency : SLES 9 only
1888 Description: Error of copying files with lustre special EAs as root
1889 Details : Client side always return success for setxattr call for lustre
1890 special xattr (currently only "trusted.lov").
1893 Frequency : rarely on clusters with both ia64+i386 clients
1895 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1896 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1897 clients are doing concurrent writes to the same file it is
1898 possible that the ia64 clients may LASSERT because the OST
1899 extent locks are not PAGE_SIZE aligned. Ensure that grown
1900 locks are always aligned on the request boundary.
1903 Frequency : specific use, occasional
1905 Description: Overwriting in use executable truncates on-disk binary image
1906 Details : If one node attempts to overwrite an executable in use by
1907 another node, we now correctly return ETXTBSY instead of
1908 truncating the file.
1910 Severity : enhancement
1912 Description: Async OSC create to avoid the blocking unnecessarily.
1913 Details : If a OST has no remain object, system will block on the creating
1914 when need to create a new object on this OST. Now, ways use
1915 pre-created objects when available, instead of blocking on an
1916 empty osc while others are not empty. If we must block, we block
1917 for the shortest possible period of time.
1922 Description: chmod on Lustre root is propagated to other clients
1923 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1924 invalid by the follow_mount time.
1927 Frequency : liblustre clients only
1929 Description: Race in 'instant cancel' lock handling could lead to such locks
1930 never to be granted in case of SMP MDS
1931 Details : Do not destroy not yet granted but cbpending locks in
1935 Frequency : replay/resend of open
1937 Description: non null lock assetion failure in mds_intent_policy
1938 Details : Trying to replay/resend lockless open requests resulted in
1939 mds_open() returning 0 with no lock. Now it sets a flag if
1940 a lock is going to be returned.
1942 Severity : enhancement
1944 Description: Checksum enhancements
1945 Details : New checksum enhancements allow for resending RPCs that failed
1948 Severity : enhancement
1950 Description: Tunables on number of dirty pages in cacche
1951 Details : Allow to set limit on number of dirty pages cached.
1956 Description: client crash on unmount - lock still has references
1957 Details : In some error handling cases it was possible to leak a lock
1958 reference on a client while accessing a file. This was not
1959 harmful to the client during operation, but would cause the
1960 client to crash when the filesystem is unmounted.
1963 Frequency : specific case, rare
1965 Description: ETXTBSY on mds though file not in use by client
1966 Details : ETXTBSY is no longer incorrectly returned when attempting to
1967 chmod or chown a directory that the user previously tried to
1968 execute or a currently-executing binary.
1971 Frequency : extremely rare except on liblustre-based clients
1973 Description: Lustre space not freed when files are deleted
1974 Details : Clean up open-unlinked files after client eviction. Previously
1975 the unlink was skipped and the files remained as orphans.
1980 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1981 Details : In some cases it was possible to send a blocking callback to
1982 a client doing a glimpse, even though that client didn't get
1983 a lock granted. When the glimpse lock is cancelled on the OST
1984 the freed lock is left on the waiting list and corrupted the list.
1987 Frequency : all core dumps
1989 Description: Broke core dumps to lustre
1990 Details : Negative dentry may be unhashed if parent does not have UPDATE
1991 lock, but some callers, e.g. do_coredump, expect dentry to be
1992 hashed after successful create, hash it in ll_create_it.
1994 ------------------------------------------------------------------------------
1996 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1998 * Support for kernels:
1999 2.6.9-42.0.2.EL (RHEL 4)
2000 2.6.5-7.276 (SLES 9)
2001 2.4.21-40.EL (RHEL 3)
2002 2.6.12.6 vanilla (kernel.org)
2006 Frequency : always on RHEL 3
2008 Description: Number of open files grows over time
2009 Details : The number of open files grows over time, whether or not
2010 Lustre is started. This was due to a filp leak introduced
2011 by one of our kernel patches.
2013 ------------------------------------------------------------------------------
2015 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
2017 * Support for kernels:
2018 2.6.9-42.EL (RHEL 4)
2019 2.6.5-7.267 (SLES 9)
2020 2.4.21-40.EL (RHEL 3)
2021 2.6.12.6 vanilla (kernel.org)
2026 Bugzilla : 5719, 9635, 9792, 9684
2027 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2028 Details : If a server is under heavy load and cannot reply to new
2029 connection requests before the client resends the (re)connect,
2030 the connection handling code can behave badly if two service
2031 threads are concurrently handing separate (re)connections from
2032 the same client. Add better locking to the connection handling
2033 code, and ensure that only a single connection will be processed
2034 for a given client UUID, even if the lock is dropped.
2036 Severity : enhancement
2038 Description: add TCP zero-copy support to kernel
2039 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2040 sends if the hardware supports scatter-gather and checksumming.
2041 This allows socklnd to do client-write and server-read more
2042 efficiently and reduce CPU utilization from skbuf copying.
2045 Frequency : only if NFS exporting from client
2047 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2048 Details : If setting attributes on a file created under NFS that had
2049 never been opened it would be possible to oops the client
2050 if the file had no objects.
2054 Bugzilla : 9326, 10402, 10897
2055 Description: client crash in ptlrpcd_wake() thread when sending async RPC
2056 Details : It is possible that ptlrpcd_wake() dereferences a freed async
2057 RPC. In rare cases the ptlrpcd thread alread processed the RPC
2058 before ptlrpcd_wake() was called and the request was freed.
2061 Frequency : always for liblustre
2063 Description: liblustre client does MDS+OSTs setattr RPC for each write
2064 Details : When doing a write from a liblustre client, the client
2065 incorrectly issued an RPC to the MDS and each OST the file was
2066 striped over in order to update the timestamps. When writing
2067 with small chunks and many clients this could overwhelm the MDS
2068 with RPCs. In all cases it would slow down the write because
2069 these RPCs are unnecessary.
2071 Severity : enhancement
2073 Description: allow number of MDS service threads to be changed at module load
2074 Details : It is now possible to change the number of MDS service threads
2075 running. Adding "options mds mds_num_threads={N}" to the MDS's
2076 /etc/modprobe.conf will set the number of threads for the next
2077 time Lustre is restarted (assuming the "mds" module is also
2078 reloaded at that time). The default number of threads will
2079 stay the same, 32 for most systems.
2084 Description: OST crash if filesystem is unformatted or corrupt
2085 Details : If an OST is started on a device that has never been formatted
2086 or if the filesystem is corrupt and cannot even mount then the
2087 error handling cleanup routines would dereference a NULL pointer.
2092 Description: NULL pointer deref in llap_from_page.
2093 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2094 incorrect) mapping if the page was truncated/reclaimed while it was
2095 searched for. Check for this condition and skip such pages when
2096 doing readahead. Introduce extra check to llap_from_page() to
2097 verify page->mapping->host is non-NULL (so page is not anonymous).
2100 Frequency : Sometimes when using sys_sendfile
2102 Description: "page not covered by a lock" warnings from ll_readpage
2103 Details : sendfile called ll_readpage without right page locks present.
2104 Now we introduced ll_file_sendfile that does necessary locking
2105 around call to generic_file_sendfile() much like we do in
2109 Frequency : with certain MDS communication failures at client mount time
2111 Description: NULL pointer deref after failed client mount
2112 Details : a client connection request may delayed by the network layer
2113 and not be sent until after the PTLRPC layer has timed out the
2114 request. If the client fails the mount immediately it will try
2115 to clean up before the network times out the request. Add a
2116 reference from the request import to the obd device and delay
2117 the cleanup until the network drops the request.
2120 Frequency : occasionally during client (re)connect
2122 Description: assertion failure during client (re)connect
2123 Details : processing a client connection request may be delayed by the
2124 client or server longer than the client connect timeout. This
2125 causes the client to resend the connection request. If the
2126 original connection request is replied in this interval, the
2127 client may trip an assertion failure in ptlrpc_connect_interpret()
2128 which thought it would be the only running connect process.
2131 Frequency : only with obd_echo servers and clients that are rebooted
2133 Description: kernel BUG accessing uninitialized data structure
2134 Details : When running an obd_echo server it did not start the ping_evictor
2135 thread, and when a client was evicted an uninitialized data
2136 structure was accessed. Start the ping_evictor in the RPC
2137 service startup instead of the OBD startup.
2139 Severity : enhancement
2140 Bugzilla : 10193 (patchless)
2141 Description: Remove dependency on various unexported kernel interfaces.
2142 Details : No longer need reparent_to_init, exit_mm, exit_files,
2143 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2146 Frequency : rare (only users of deprecated and unsupported LDAP config)
2148 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2149 Details : LDAP apparently contains 'lustreName' attributes instead of
2150 'name'. A simple remapping of the name is sufficient.
2153 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2155 Description: waiting_locks_callback trips kernel BUG if client is evicted
2156 Details : Running with the dump_on_timeout debug flag turned on makes
2157 it possible that the waiting_locks_callback() can try to dump
2158 the Lustre kernel debug logs from an interrupt handler. Defer
2159 this log dumping to the expired_lock_main() thread.
2161 Severity : enhancement
2163 Description: Support NFS exporting on 2.6 kernels.
2164 Details : Implement non-rawops metadata methods for NFS server to use without
2165 changing NFS server code.
2168 Frequency : very rare (synthetic metadata workload only)
2170 Description: two racing renames might cause an MDS thread to deadlock
2171 Details : Running the "racer" program may cause one MDS thread to rename
2172 a file from being the source of a rename to being the target of
2173 a rename at exactly the same time that another thread is doing
2174 so, and the second thread has already enqueued these locks after
2175 doing a lookup of the target and is trying to relock them in
2176 order. Ensure that we don't try to re-lock the same resource.
2179 Frequency : only very large systems with liblustre clients
2181 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2182 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2183 all OSTs in parallel. This allows the eviction of stale liblustre
2184 clients to proceed much faster than if they were done in series,
2185 and also offers similar improvements for other set_info RPCs.
2190 Description: excessive CPU usage during initial read phase on client
2191 Details : During the initial read phase on a client, it would agressively
2192 retry readahead on the file, consuming too much CPU and impacting
2193 performance (since 1.4.5.8). Improve the readahead algorithm
2194 to avoid this, and also improve some other common cases (read
2195 of small files in particular, where "small" is files smaller than
2196 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2201 Description: MDS crash when receiving packet with unknown intent.
2202 Details : Do not LBUG in unknown intent case, just return -EFAULT
2204 Severity : enhancement
2205 Bugzilla : 9293, 9385
2206 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2207 Details : Do not serialize getattr (non-intent version) and statfs.
2210 Frequency : occasional, when OST network is overloaded/intermittent
2212 Description: client evicted by OST after bulk IO timeout
2213 Details : If a client sends a bulk IO request (read or write) the OST
2214 may evict the client if it is unresposive to its data GET/PUT
2215 request. This is incorrect if the network is overloaded (takes
2216 too long to transfer the RPC data) or dropped the OST GET/PUT
2217 request. There is no need to evict the client at all, since
2218 the pinger and/or lock callbacks will handle this, and the
2219 client can restart the bulk request.
2222 Frequency : Always when mmapping file with no objects
2224 Description: client crashes when mmapping file with no objects
2225 Details : Check that we actually have objects in a file before doing any
2226 operations on objects in ll_vm_open, ll_vm_close and
2232 Description: Request leak when working with deleted CWD
2233 Details : Introduce advanced request refcount tracking for requests
2234 referenced from lustre intent.
2236 Severity : Enhancement
2238 Description: Cache open file handles on client.
2239 Details : MDS now will return special lock along with openhandle, if
2240 requested and client is allowed to hold openhandle, even if unused,
2241 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2242 closing files for every read/write openration.
2244 Severity : Enhancement
2246 Description: Cache open negative dentries on client when possible.
2247 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2248 negative dentries on lock revocation.
2253 Description: Remounting a client read-only wasn't possible with a zconf mount
2254 Details : It wasn't possible to remount a client read-only with llmount.
2256 Severity : enhancement
2257 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2258 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2259 MPICH is included to add Lustre-specific ADIO interfaces.
2260 This is based closely on the UFS ADIO layer and only differs
2261 in file creation, in order to allow the OST striping to be set.
2262 This is user-contributed code and not supported by CFS.
2267 Description: extended inode attributes (immutable, append-only) work improperly
2268 when 2.4 and 2.6 kernels are used on client/server or vice versa
2269 Details : Introduce kernel-independent values for these flags.
2271 Severity : enhancement
2274 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2275 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2276 granularity. Now, Fractional MB granularity is be supported,
2277 this is very useful for low memory system.
2279 Severity : enhancement
2281 Description: Getattr by fid
2282 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2283 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2284 have locks and avoids getattr rpc requests for it on every stat.
2287 Frequency : Always, for filesystems larger than 2TB
2289 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2290 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2291 filesystems larger than 2TB to be mounted with mballoc enabled.
2294 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2296 Description: filesystem corruption for non-standard kernels and very large OSTs
2297 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2298 filesystems larger than 2TB is mounted then the kernel will
2299 silently corrupt the start of the filesystem. CONFIG_LBD is
2300 enabled for all CFS-supported kernels, but the possibility of
2301 this happening with a modified kernel config exists.
2303 Severity : enhancement
2305 Description: add client O_DIRECT support for 2.6 kernels
2306 Details : It is now possible to do O_DIRECT reads and writes to files
2307 in the Lustre client mountpoint on 2.6 kernel clients.
2309 Severity : enhancement
2311 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2312 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2313 parallel, not waiting for response from every OST before sending
2314 a rpc to the next OST.
2319 Description: setattr vs write race when updating file timestamps
2320 Details : Client processes that update a file timestamp into the past
2321 right after writing to the file (e.g. tar) it is possible that
2322 the updated file modification time can be reset to the current
2323 time due to a race between processing the setattr and write RPC.
2325 Severity : enhancement
2327 Description: Bring 'lfs find' closer in line with regular Linux find.
2328 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2329 -print0 options and obtains all the needed info through the lustre
2332 Severity : enhancement
2334 Description: support up to 1024 configured devices on one node
2335 Details : change obd_dev array from statically allocated to dynamically
2336 allocated structs as they are first used to reduce memory usage
2341 Description: Flush dirty partially truncated pages during truncate
2342 Details : Immediatelly flush partially truncated pages in filter_setattr,
2343 this way we completely avoid having any pages in page cache on OST
2344 and can retire ugly workarounds during writes to flush such pages.
2349 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2350 Details : protect inode from truncation within vfs_unlink() context
2351 just take a reference before calling vfs_unlink() and release it
2352 when parent's i_sem is free.
2355 Frequency : always, if extents are used on OSTs
2357 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2358 Details : OSTs using the extents format would not zero the high 16 bits of
2359 the index physical block number. This is not a problem for any
2360 OST filesystems smaller than 16TB, and no kernels support ext3
2361 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2362 new/modified files) and can be fixed for existing filesystems
2363 with e2fsprogs-1.39-cfs1.
2368 Description: import connection selection may be incorrect if timer wraps
2369 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2370 import connections to be ignored if the 32-bit jiffies counter
2371 wraps. Use a 64-bit jiffies counter.
2374 Frequency : during server recovery
2376 Description: crash after server is denying duplicate export
2377 Details : If clients are resending connect requests to the server, the
2378 server refuses to allow a client to connect multiple times.
2379 Fixed a bug in the handling of this case.
2382 Frequency : very large clusters immediately after boot
2384 Description: LNET request buffers exhausted under heavy short-term load
2385 Details : If a large number of client requests are generated on a service
2386 that has previously never seen so many requests it is possible
2387 that the request buffer growth cannot keep up with the spike in
2388 demand. Instead of dropping incoming requests, they are held in
2389 the LND until the RPC service can accept more requests.
2392 Frequency : Sometimes during replay
2394 Description: Assertion failure in ll_local_open after replay.
2395 Details : If replay happened on an open request reply before we were able
2396 to set replay handler, reply will become not swabbed tripping the
2397 assertion in ll_local_open. Now we set the handler right after
2398 recognising of open request
2401 Frequency : very rare
2403 Description: kernel reports "badness in vsnprintf"
2404 Details : Reading from the "recovery_status" /proc file in small chunks
2405 may cause a negative length in lprocfs_obd_rd_recovery_status()
2406 call to vsnprintf() (which is otherwise harmless). Exit early
2407 if there is no more space in the output buffer.
2409 Severity : enhancement
2411 Description: clear OBD RPC statistics by writing to them
2412 Details : It is now possible to clear the OBD RPC statistics by writing
2413 to the "stats" file.
2418 Description: Client mtime is not the same on different clients after utimes
2419 Details : In some cases, the client was using the utimes() syscall on
2420 a file cached on another node. The clients now validate the
2421 ctime from the MDS + OSTs to determine which one is right.
2426 Description: Inability to activate failout mode
2427 Details : lconf script incorrectly assumed that in python string's numeric
2428 value is used in comparisons.
2431 Frequency : always with multiple stripes per file
2433 Description: Inefficient object allocation for mutli-stripe files
2434 Details : When selecting which OSTs to stripe files over, for files with
2435 a stripe count that divides evenly into the number of OSTs,
2436 the MDS is always picking the same starting OST for each file.
2437 Return the OST selection heuristic to the original design.
2442 Description: mount failures may take full timeout to return an error
2443 Details : Under some heavy load conditions it is possible that a
2444 failed mount can wait for the full obd_timeout interval,
2445 possibly several minutes, before reporting an error.
2446 Instead return an error as soon as the status is known.
2448 ------------------------------------------------------------------------------
2450 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
2452 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2453 INTEROPERATE with older versions automatically. Please read the
2454 user documentation before upgrading any part of a live system.
2455 * WARNING: Lustre networking configuration changes are required with
2456 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2459 * Support for kernels:
2460 2.6.9-22.0.2.EL (RHEL 4)
2461 2.6.5-7.244 (SLES 9)
2462 2.6.12.6 vanilla (kernel.org)
2465 Severity : enhancement
2466 Bugzilla : 7981/8208
2467 Description: Introduced Lustre Networking (LNET)
2468 Details : LNET is new networking infrastructure for Lustre, it includes
2469 a reorganized network configuration mode (see the user
2470 documentation for full details) as well as support for routing
2471 between different network fabrics. Lustre Networking Devices
2472 (LNDs) for the supported network fabrics have also been
2473 created for this new infrastructure.
2475 Severity : enhancement
2476 Description: Introduced Access control lists
2477 Details : clients can set ACLs on files and directories in order to have
2478 more fine-grained permissions than the standard Unix UGO+RWX.
2479 The MDS must be started with the "-o acl" mount option.
2481 Severity : enhancement
2482 Description: Introduced filesystem quotas
2483 Details : Administrators may now establish per-user quotas on the
2486 Severity : enhancement
2488 Description: Configuration change for the XT3
2489 The PTLLND is now used to run Lustre over Portals on the XT3
2490 The configure option(s) --with-cray-portals are no longer used.
2491 Rather --with-portals=<path-to-portals-includes> is used to
2492 enable building on the XT3. In addition to enable XT3 specific
2493 features the option --enable-cray-xt3 must be used.
2498 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2499 Details : A race between allocating a new llcd and re-getting the llcd_lock
2500 allowed another thread to grab newly-allocated llcd.
2502 Severity : enhancement
2504 Description: 2.6 OST async journal commit and locking fix to improve performance
2505 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2506 2.6 kernels are now async as they already were in 2.4 kernels so
2507 that they can commit concurrently with the network bulk transfer.
2508 For block-allocated files the filter allocation semaphore is held
2509 to avoid filesystem fragmentation during allocation. BKL lock
2510 removed for 2.6 xattr operations where it is no longer needed.
2515 Description: lconf incorrectly determined whether two IP networks could talk
2516 Details : In some more complicated routing and multiple-network
2517 configurations, lconf will avoid trying to make a network
2518 connection to a disjoint part of the IP space. It was doing the
2519 math incorrectly for one set of cases.
2524 Description: Fix for potential infinite loop processing records in an llog.
2525 Details : If an llog record is corrupted/zeroed, it is possible to loop
2526 forever in llog_process(). Validate the llog record length
2527 and skip the remainder of the block on error.
2530 Frequency : occasional (liblustre only)
2532 Description: liblustre could not open files whose last component is a symlink
2533 Details : sysio_path_walk() would incorrectly pass the open intent to
2534 intermediate path components.
2537 Frequency : rare (liblustre only with non-standard tuning)
2538 Bugzilla : 7201 (7350)
2539 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2540 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2541 after it has been released, only do this for referenced locks
2543 Severity : enhancement
2545 Description: specifying an (invalid) directory default stripe_size of -1
2546 would reset the directory default striping
2547 Details : stripe_size -1 was used internally to signal directory stripe
2548 removal, now use "all default" to signal dir stripe removal
2549 as a directory striping of "all default" is not useful
2552 Frequency : common for large clusters running liblustre clients
2554 Description: doing an ls when liblustre clients are running is slow
2555 Details : sending a glimpse AST to a liblustre client waits for every AST
2556 to time out, as liblustre clients will not respond. Since they
2557 cannot cache data we refresh the OST lock LVB from disk instead.
2559 Severity : enhancement
2561 Description: doing an ls at the same time as file IO can be slow
2562 Details : enqueue and other "small" requests can be blocked behind many
2563 large IO requests. Create a new OST IO portal for non-IO
2564 requests so they can be processed faster.
2567 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2569 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2570 created file with mode 000
2571 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2572 allow an ftruncate() as a non-root user to a file with mode 000.
2573 Lustre now respects this flag to disable mode checks when
2574 truncating a file owned by the user
2577 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2580 Description: Revoking locks from clients that went dead or catatonic might take
2582 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2583 cancellation of such locks instant on servers without waiting for
2584 any reply from clients. Clients drops these locks when cancel
2585 notification from server is received without replying.
2588 Frequency : liblustre-only, when liblustre client dies or becomes busy
2590 Description: Doing ls on Linux clients can take a long time with active
2592 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2593 granting such locks to it in the first place if possible. Locks
2594 are taken by proxy on the OST during the read or write and
2595 dropped immediately afterward. Add connect flags handling, do
2596 not grant locks to liblustre clients for glimpse ASTs.
2598 Severity : enhancement
2600 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2601 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2602 sequential reads and behaves badly with applications doing random
2603 reads. Improve it by reading ahead at least read region, and
2604 avoiding excessive large RPC for small reads.
2606 Severity : enhancement
2608 Description: Creating more than 1000 files for a single job may cause a load
2609 imbalance on the OSTs if there are also a large number of OSTs.
2610 Details : qos_prep_create() uses an OST index reseed value that is an
2611 even multiple of the number of available OSTs so that if the
2612 reseed happens in the middle of the object allocation it will
2613 still utilize the OSTs as uniformly as possible.
2618 Description: OST or MDS may oops in ping_evictor_main()
2619 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2620 but doesn't restart at beginning of obd_exports_timed list
2623 Severity : enhancement
2625 Description: improve by-nid export eviction on the MDS and OST
2626 Details : allow multiple exports with the same NID to be evicted at one
2627 time without re-searching the exports list.
2630 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2632 Description: MDS may oops in groups_free()
2633 Details : in rare race conditions a newly allocated group_info struct is
2634 freed again, and this can be NULL. The 2.4 compatibility code
2635 for groups_free() checked for a NULL pointer, but 2.6 did not.
2638 Frequency : common for liblustre clients doing little filesystem IO
2639 Bugzilla : 9352, 7313
2640 Description: server may evict liblustre clients accessing contended locks
2641 Details : if a client is granted a lock or receives a completion AST
2642 with a blocking AST already set it would not reply to the AST
2643 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2646 Frequency : lfs setstripe, only systems with more than 160 OSTs
2648 Description: unable to set striping with a starting offset beyond OST 160
2649 Details : llapi_create_file() incorrectly limited the starting stripe
2650 index to the maximum single-file stripe count.
2653 Frequency : LDAP users only
2655 Description: lconf did not handle in-kernel recovery with LDAP properly
2656 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2658 Severity : enhancement
2660 Description: bind OST threads to NUMA nodes to improve performance
2661 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2662 node and do their allocations there to localize memory access
2664 Severity : enhancement
2666 Description: llmount can determine client NID directly from Myrinet (GM)
2667 Details : the client NID code from gmnalnid was moved directly into
2668 llmount, removing the need to use this or specifying the
2669 client NID explicitly when mounting GM clients with zeroconf
2672 Frequency : if client is started with down MDS
2674 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2675 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2676 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2677 ptlrpc_import_delayed_req() because we didn't check if the
2678 request was interrupted, and we also didn't break out of the
2679 event loop if there was no timeout
2684 Description: data loss during non-page-aligned writes to a single file from
2685 both multiple nodes and multiple threads on one node at same time
2686 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2687 inconsistency led to false short-reads, that were cached and later
2688 used by ->prepare_write() to fill in partially written page,
2689 leading to data loss.
2692 Frequency : always, if lconf --abort_recovery used
2694 Description: lconf --abort_recovery fails with 'Operation not supported'
2695 Details : lconf was attempting to abort recovery on the MDT device and not
2698 Severity : enhancement
2700 Description: remove cleanup logs
2701 Details : replace lconf-generated cleanup logs with lustre internal
2702 cleanup routines. Eliminates the need for client-cleanup and
2705 Severity : enhancement
2707 Description: add support for EAs (user and system) on lustre filesystems
2708 Details : it is now possible to store extended attributes in the Lustre
2709 client filesystem, and with the user_xattr mount option it
2710 is possible to allow users to store EAs on their files also
2712 Severity : enhancement
2714 Description: Add possibility (config option) to show minimal available OST free
2716 Details : When compiled with --enable-mindf configure option, statfs(2)
2717 (and so, df) will return least minimal free space available from
2718 all OSTs as amount of free space on FS, instead of summary of
2719 free spaces of all OSTs.
2721 Severity : enhancement
2723 Description: do not expand extent locks acquired on OST-side
2724 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2725 by server: they are not cached anyway.
2728 Frequency : when mmap is used/binaries executed from Lustre
2730 Description: Unmmap pages before throwing them away from read cache.
2731 Details : llap_shrink cache now attempts to unmap pages before discarding
2732 them (if unmapping failed - do not discard). SLES9 kernel has
2733 extra checks that trigger if this unmapping is not done first.
2738 Description: lconf didn't resolve symlinks before checking to see whether a
2739 given mountpoint was already in use
2742 Frequency : when migrating failover services
2743 Bugzilla : 6395, 9514
2744 Description: When migrating a subset of services from a node (e.g. failback
2745 from a failover service node) the remaining services would
2746 time out and evict clients.
2747 Details : lconf --force (implied by --failover) sets the global obd_timeout
2748 to 5 seconds in order to quickly disconnect, but this caused
2749 other RPCs to time out too quickly. Do not change the global
2750 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2752 Severity : enhancement
2753 Frequency : if MDS is started with down OST
2754 Bugzilla : 9439,5706
2755 Description: Allow startup/shutdown of an MDS without depending on the
2756 availability of the OSTs.
2757 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2758 Add appropriate locking and lov-osc refcounts for safe
2759 cleaning. Add osc abort_inflight calls in case the
2760 synchronize never started.
2763 Frequency : occasional (Cray XT3 only)
2765 Description: root not authorized to access files in CRAY_PORTALS environment
2766 Details : The client process capabilities were not honoured on the MDS in
2767 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2768 been accessed by an authorized user then root was able to access
2769 the file on the local client also. The root user capabilities
2770 are now allowed on the MDS, as this environment has secure UID.
2773 Frequency : occasional
2775 Description: ldiskfs "too long searching" message happens too often
2776 Details : A debugging message (otherwise harmless) prints too often on
2777 the OST console. This has been reduced to only happen when
2778 there are fragmentation problems on the filesystem.
2783 Description: Division by zero in statfs when all OSCs are inactive
2784 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2785 lov_statfs divides by value returned by lov_get_stripecnt().
2789 Bugzilla : 9489, 3273
2790 Description: First write from each client to each OST was only 4kB in size,
2791 to initialize client writeback cache, which caused sub-optimal
2792 RPCs and poor layout on disk for the first writen file.
2793 Details : Clients now request an initial cache grant at (re)connect time
2794 and so that they can start streaming writes to the cache right
2795 away and always do full-sized RPCs if there is enough data.
2796 If the OST is rebooted the client also re-establishes its grant
2797 so that client cached writes will be honoured under the grant.
2802 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2803 Details : Now I/O RPCs go to different portal number and (presumably) fast
2804 lock requests (and glimses) and other RPCs get their own service
2805 threads pool that should be able to service those RPCs
2808 Severity : enhancement
2810 Description: Ability to exchange lustre version between client and servers and
2811 issue warnings at client side if client is too old. Also for
2812 liblustre clients there is ability to refuse connection of too old
2814 Details : New 'version' field is added to connect data structure that is
2815 filled with version info. That info is later checked by server and
2819 Frequency : rare, liblustre only.
2820 Bugzilla : 9296, 9581
2821 Description: Two simultaneous writes from liblustre at offset within same page
2822 might proceed at the same time overwriting eachother with stale
2824 Details : I/O lock withing llu_file_prwv was released too early, before data
2825 actually was hitting the wire. Extended lock-holding time until
2826 server acknowledges receiving data.
2829 Frequency : extremely rare. Never observed in practice.
2831 Description: avoid generating lustre_handle cookie of 0.
2832 Details : class_handle_hash() generates handle cookies by incrementing
2833 global counter, and can hit 0 occasionaly (this is unlikely, but
2834 not impossible, because initial value of cookie counter is
2835 selected randonly). Value of 0 is used as a sentinel meaning
2836 "unassigned handle" --- avoid it. Also coalesce two critical
2837 sections in this function into one.
2839 Severity : enhancement
2841 Description: allow liblustre clients to delegate truncate locking to OST
2842 Details : To avoid overhead of locking, liblustre client instructs OST to
2843 take extent lock in ost_punch() on client's behalf. New connection
2844 flag is added to handle backward compatibility.
2846 Severity : enhancement
2847 Bugzilla : 4928, 7341, 9758
2848 Description: allow number of OST service threads to be specified
2849 Details : a module parameter allows the number of OST service threads
2850 to be specified via "options ost ost_num_threads={N}" in the
2851 OSS's /etc/modules.conf or /etc/modprobe.conf.
2855 Bugzilla : 6146, 9635, 9895
2856 Description: servers crash with bad pointer in target_handle_connect()
2857 Details : In rare cases when a client is reconnecting it was possible that
2858 the connection request was the last reference for that export.
2859 We would temporarily drop the export reference and get a new
2860 one, but this may have been the last reference and the export
2861 was just destroyed. Get new reference before dropping old one.
2863 Severity : enhancement
2864 Frequency : if client is started with failover MDS
2866 Description: Allow multiple MDS hostnames in the mount command
2867 Details : Try to read the configuration from all specified MDS
2868 hostnames during a client mount in case the "primary"
2871 Severity : enhancement
2873 Description: Stop sending data to evicted clients as soon as possible.
2874 Details : Check if the client we are about to send or are sending data to
2875 was evicted already. (Check is done every second of waiting,
2876 for which l_wait_event interface was extended to allow checking
2877 of exit condition at specified intervals).
2880 Frequency : rare, normally only when NFS exporting is done from client
2882 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2884 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2887 Frequency : always, if srand() is called before liblustre initialization
2889 Description: Liblustre uses system PRNG disturbing its usage by user application
2890 Details : Introduce internal to lustre fast and high-quality PRNG for
2891 lustre usage and make liblustre and some other places in generic
2892 lustre code to use it.
2894 Severity : enhancement
2895 Bugzilla : 9477, 9557, 9870
2896 Description: Verify that the MDS configuration logs are updated when xml is
2897 Details : Check if the .xml configuration logs are newer than the config
2898 logs stored on the MDS and report an error if this is the case.
2899 Request --write-conf, or allow starting with --old_conf.
2901 Severity : enhancement
2903 Description: Handle symlinks in the path when checking if Lustre is mounted.
2904 Details : Resolve intermediate symlinks when checking if a client has
2905 mounted a filesystem to avoid duplicate client mounts.
2910 Description: lconf can hit an error exception but still return success.
2911 Details : The lconf command catches the Command error exception at the top
2912 level script context and will exit with the associated exit
2913 status, but doesn't ensure that this exit status is non-zero.
2918 Description: failure of ptlrpc thread startup can cause oops
2919 Details : Starting a ptlrpc service thread can fail if there are a large
2920 number of threads or the server memory is very fragmented.
2921 Handle this without oopsing.
2924 Frequency : always, only if liblustre and non-default acceptor port was used
2926 Description: liblustre cannot connect to servers with non-default acceptor port
2927 Details : tcpnal_set_default_params() was not called and was therefore
2928 ignoring the environment varaible TCPNAL_PORT, as well as other
2929 TCPNAL_ environment variables
2934 Description: two objects could be created on the same OST for a single file
2935 Details : If an OST is down, in some cases it was possible to create two
2936 objects on a single OST for a single file. No problems other
2937 than potential performance impact and spurious error messages.
2941 Bugzilla : 5681, 9562
2942 Description: Client may oops in ll_unhash_aliases
2943 Details : Client dcache may become inconsistent in race condition.
2944 In some cases "getcwd" can fail if the current directory is
2950 Description: Inode refcounting problems in NFS export code
2951 Details : link_raw functions used to call d_instantiate without obtaining
2952 extra inode reference first.
2956 Bugzilla : 9942, 9903
2957 Description: Referencing freed requests leading to crash, memleaks with NFS.
2958 Details : We used to require that call to ll_revalidate_it was always
2959 followed by ll_lookup_it. Also with revalidate_special() it is
2960 possible to call ll_revalidate_it() twice for the same dentry
2961 even if first occurence returned success. This fix changes semantic
2962 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2963 reference on a request referred from the intent.
2964 ll_intent_release() then releases such a request.
2967 Frequency : rare, normally benchmark loads only
2969 Description: unlinked inodes were kept in memory on the client
2970 Details : If a client is repeatedly creating and unlinking files it
2971 can accumulate a lot of stale inodes in the inode slab cache.
2972 If there is no other client load running this can cause the
2973 client node to run out of memory. Instead flush old inodes
2974 from client cache that have the same inode number as a new inode.
2977 Frequency : SLES9 2.6.5 kernel and long filenames only
2978 Bugzilla : 9969, 10379
2979 Description: utime reports stale NFS file handle
2980 Details : SLES9 uses out-of-dentry names in some cases, which confused
2981 the lustre dentry revalidation. Change it to always use the
2985 Frequency : rare, unless heavy write-truncate concurrency is continuous
2986 Bugzilla : 4180, 6984, 7171, 9963, 9331
2987 Description: OST becomes very slow and/or deadlocked during object unlink
2988 Details : filter_destroy() was holding onto the parent directory lock
2989 while truncating+unlinking objects. For very large objects this
2990 may block other threads for a long time and slow overall OST
2991 responsiveness. It may also be possible to get a lock ordering
2992 deadlock in this case, or run out of journal credits because of
2993 the combined truncate+unlink. Solution is to do object truncate
2994 first in one transaction without parent lock, and then do the
2995 final unlink in a new transaction with the parent lock. This
2996 reduces the lock hold time dramatically.
2999 Frequency : rare, 2.4 kernels only
3001 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3002 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3003 before unmounting their filesystem in order to avoid deadlock.
3004 The kernel_locked() function in 2.4 kernels only checks whether
3005 the kernel lock is held, not whether it is this process that is
3006 holding it as 2.6 kernels do.
3011 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3012 Details : The client ptlrpc code may be trying to reconnect to a down
3013 server before a previous connection attempt has timed out.
3014 Increase the reconnect interval to be longer than the connection
3015 timeout interval to avoid sending duplicate connections to
3019 Frequency : echo_client brw_test command
3021 Description: fix echo_client to work with OST preallocated code
3022 Details : OST preallocation code (5137) didn't take echo_client IO path
3023 into account: echo_client calls filter methods outside of any
3024 OST thread and, hence, there is no per-thread preallocated
3025 pages and buffers to use. Solution: hijack pga pages for IO. As
3026 a byproduct, this avoids unnecessary data copying.
3030 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3031 Description: Client can oops in mdc_commit_close() after open replay
3032 Details : It was possible for the MDS to return an open request with no
3033 transaction number in mds_finish_transno() if the client was
3034 evicted, but without actually returning an error. Clients
3035 would later try to replay that open and may trip an assertion
3036 Simplify the client close codepath, and always return an error
3037 from the MDS in case the open is not successful.
3040 Frequency : rare, 2.6 OSTs only
3042 Description: OST may deadlock under high load on fragmented files
3043 Details : If there was a heavy load and highly-fragmented OST filesystems
3044 it was possible to have all the OST threads deadlock waiting on
3045 allocation of biovecs, because the biovecs were not released
3046 until the entire RPC IO was completed. Instead, release biovecs
3047 as soon as they are complete to ensure forward IO progress.
3049 Severity : enhancement
3051 Description: Support for specifying external journal device at mount
3052 Details : If an OST or MDS device is formatted with an external journal
3053 device, this device major/minor is stored in the ext3 superblock
3054 and may not be valid for failover. Allow detecting and
3055 specifying the external journal at mount time.
3060 Description: Mounting an MDS with pending unlinked files may cause oops
3061 Details : target_finish_recovery() calls mds_postrecov() which returned
3062 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3063 considers this an error and immediately begins cleaning up the
3064 lov, just after starting the mds_lov process
3066 Severity : enhancement
3068 Description: Implement 'lfs df' to report actual free space on per-OST basis
3069 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3070 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3071 usage of objects; '-h' to report in human readable format.
3073 ------------------------------------------------------------------------------
3075 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3082 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3083 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3084 at startup during orphan unlinking. Instead, skip buddy update
3085 before setup, as it will be regenerated after recovery is complete.
3090 Description: If an OST is inactive, its locks might reference stale inodes.
3091 Details : lov_change_cbdata() must iterate over all namespaces, even if
3092 they are inactive to clear inode references from the lock.
3094 Severity : enhancement
3095 Frequency : occasional, if non-standard max_dirty_mb used
3097 Description: Client will block write RPCs if not enough grant
3098 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3099 then the client will block writes while waiting for another RPC
3100 to complete instead of consuming its dirty limit. With change
3101 we get improved performance when max_dirty_mb is small.
3103 Severity : enhancement
3104 Bugzilla : 3389, 6253
3105 Description: Add support for supplementary groups on the MDS.
3106 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3107 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3108 lookups for user supplementary groups into a cache.
3112 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3113 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3114 for liblustre O_EXCL is enforced by MDS.
3118 Description: readdir never returns NULL in liblustre.
3119 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3120 dentry in struct dirent.
3124 Frequency : liblustre only, depends on application IO pattern
3125 Description: liblustre clients evicted if not contacting servers
3126 Details : Don't put liblustre clients into the ping_evictor list, so
3127 they will not be evicted by the pinger ever.
3129 Severity : enhancement
3131 Description: Add ability to evict clients by NID from MDS.
3132 Details : By echoing "nid:$NID" string into
3133 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3134 $NID would be instantly evicted from this MDS and from all active
3135 OSTs connected to it.
3139 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3140 Details : lookup_it_finish() used to query file size from OSTs that was not
3145 Description: service threads change working directory to that of init
3146 Details : Starting lustre service threads may pin the working directory
3147 of the parent thread, making that filesystem busy. Threads
3148 now change to the working directory of init to avoid this.
3152 Frequency : during shutdown only
3153 Description: shutdown with a failed MDS or OST can cause unmount to hang
3154 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3159 Frequency : 2.6 only
3160 Description: chmod/chown may include an extra supplementary group
3161 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3162 supplementary group and if none is specified this is used.
3165 Bugzilla : 5479 (6816)
3167 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3168 Details : If lookup is in progress on a file that is unlinked we might try
3169 to revalidate the inode and fail in revalidate after lookup is
3170 complete and ll_file_open() enqueues the open again but
3171 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3174 Frequency : always, if lconf --abort_recovery used
3176 Description: lconf --abort_recovery fails with 'Operation not supported'
3177 Details : lconf was attempting to abort recovery on the MDT device and not
3180 ------------------------------------------------------------------------------
3182 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3187 Frequency : rare (only unsupported configurations with a node running as an
3189 Bugzilla : 6514, 5137
3190 Description: Mounting a Lustre file system on a node running as an OST could
3192 Details : OSTs now preallocates memory needed to write out data at
3193 startup, instead of when needed, to avoid having to
3194 allocate memory in possibly low memory situations.
3195 Specifically, if the file system is mounted on on OST,
3196 memory pressure could force it to try to write out data,
3197 which it needed to allocate memory to do. Due to the low
3198 memory, it would be unable to do so and the node would
3199 become unresponsive.
3201 Severity : enhancement
3203 Description: Addition of lconf --service command line option
3204 Details : lconf now accepts a '--service <arg>' option, which is
3205 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3207 Severity : enhancement
3209 Description: Failover mode is now the default for OSTs.
3210 Details : By default, OSTs will now run in failover mode. To return to
3211 the old behaviour, add '--failout' to the lmc line for OSTs.
3213 Severity : enhancement
3215 Description: Health checks are now provided for MDS and OSTs
3216 Details : Additional detailed health check information on MSD and OSTs
3217 is now provided through the procfs health_check value.
3220 Frequency : occasional, depends on IO load
3222 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3223 after numerous create/delete cycles
3224 Details : The ext3 inode allocation policy would not allocate new inodes
3225 very well on the OSTs because there are no new directories
3226 being created. Instead we look for groups with free space if
3227 the parent directories are nearly full.
3232 Description: Network or server problems during mount may cause partially
3233 mounted clients instead of returning an error.
3234 Details : The config llog parsing code may overwrite the error return
3235 code during mount error handling, returning success instead
3241 Description: MDS can fail to allocate large reply buffers
3242 Details : After long uptimes the MDS can fail to allocate large reply
3243 buffers (e.g. zconf client mount config records) due to memory
3244 fragmentation or consumption by the buffer cache. Preallocate
3245 some large reply buffers so that these replies can be sent even
3246 under memory pressure.
3250 Frequency : rare (liblustre)
3251 Description: fsx running with liblustre complained that using truncate() to
3252 extend the file doesn't work. This patch corrects that issue.
3253 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3254 ATTR_SIZE and lsm use in llu_setattr_raw.
3258 Frequency : rare, only 2.6 kernels
3259 Description: Unusual file access patterns on the MDS may result in inode
3260 data being lost in very rare circumstances.
3261 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3262 code in the 2.6 kernel caused the inode and/or EA data not to
3263 be read from disk, causing single-file corruption.
3267 Frequency : rare, only 2.6 filesystems using extents
3268 Description: Heavy concurrent write and delete load may cause data corruption.
3269 Details : It was possible under high-load situations to have an extent
3270 metadata block in the block device cache from a just-unlinked
3271 file overwrite a newly-allocated data block. We now unmap any
3272 metadata buffers that alias just-allocated data blocks.
3276 Frequency : filesystems with default stripe_count larger than 77
3277 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3278 Details : lconf specifies an inode size of 4096 bytes when the default
3279 stripe_count is larger than 77. This conflicts with the default
3280 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3281 this case to avoid pinning too much memory for large EAs.
3283 ------------------------------------------------------------------------------
3285 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3290 Frequency : rare (extremely heavy IO load with hundreds of clients)
3292 Description: Client is evicted, gets IO error writing to file
3293 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3294 caused clients to be evicted for AST timeouts. The fixes in
3295 bug 5192 mean we no longer need to have such short AST timeouts
3296 so ldlm_timeout has been increased.
3299 Frequency : occasional during --force or --failover shutdown under load
3300 Bugzilla : 5949, 4834
3301 Description: Server oops/LBUG if stopped with --force or --failover under load
3302 Details : a collection of import/export refcount and cleanup ordering
3303 issues fixed for safer force cleanup
3306 Frequency : only filesystems larger than 120 OSTs
3307 Bugzilla : 5990, 6223
3308 Description: lfs getstripe would oops on a very large filesystem
3309 Details : lov_getconfig used kfree on vmalloc'd memory
3312 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3313 Bugzilla : 6242, 6243
3314 Description: reading from files that had been truncated to a non-zero size
3315 but never opened returned no data
3316 Details : ll_file_read() reads zeros from no-object files to EOF
3321 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3322 Details : There's little harm in aborting MDS/OSS recovery and letting it
3323 try again, so I removed the LASSERT and return an error instead.
3325 Severity : enhancement
3327 Description: New debugging infrastructure for tracking down data corruption
3328 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3329 (b) cover more of the client-side code path, and (c) try to narrow
3330 down where problems occurred
3334 Bugzilla : 3819, 4364, 4397, 6313
3335 Description: Racing close and eviction MDS could cause assertion in mds_close
3336 Details : It was possible to get multiple mfd references during close and
3337 client eviction, leading to one thread referencing a freed mfd.
3339 Severity: : enhancement
3340 Bugzilla : 3262, 6359
3341 Description: Attempts to reconnect to servers are now more aggressive.
3342 Details : This builds on the enhanced upcall-less recovery that was added
3343 in 1.4.2. When trying to reconnect to servers, clients will
3344 now try each server in the failover group every 10 seconds. By
3345 default, clients would previously try one server every 25 seconds.
3350 Description: After recovery, certain operations trigger a failed
3351 assertion on a client.
3352 Details : Failing over an mds, using lconf -d --failover, while a
3353 client was doing a readdir() call would cause the client to
3354 LBUG after recovery completed and the readdir() was resent.
3356 Severity : enhancement
3358 Description: Default groups are now added by lconf
3359 Details : You can now run lconf --group <servicename> without having to
3360 manually add groups with lmc.
3363 Frequency : occasional
3365 Description: Nodes with an elan id of 0 trigger a failed assertion
3368 Frequency : always when accessing e.g. tty/console device nodes
3370 Description: tty and some other devices nodes cannot be used on lustre
3371 Details : file's private_data field is used by device data and lustre
3372 values in there got lost. New field was added to struct file to
3373 store fs-specific private data.
3376 Frequency : when exporting Lustre via NFS
3378 Description: NFSD failed occasionally when looking up a path component
3379 Details : NFSD is looking up ".." which was broken in ext3 directories
3380 that had grown large enough to become hashed.
3383 Frequency : Clusters with multiple interfaces not on the same subnet
3385 Description: Nodes will repeatedly try to reconnect to an interface which it
3386 cannot reach and report an error to the log.
3387 Details : Extra peer list entries will be created by lconf with some peers
3388 unreachable. lconf now validates the peer before adding it.
3391 Frequency : Only if a default stripe is set on the filesystem root.
3393 Description: Setting a default stripe on the filesystem root prevented the
3394 filesystem from being remounted.
3395 Details : The client was sending extra request flags in the root getattr
3396 request and did not allocate a reply buffer for the dir EA.
3399 Frequency : occasional, higher if lots of files are accessed by one client
3400 Bugzilla : 6159, 6097
3401 Description: Client trips assertion regarding lsm mismatch/magic
3402 Details : While revalidating inodes the VFS looks up inodes with ifind()
3403 and in rare cases can find an inode that is being freed.
3404 The ll_test_inode() code will free the lsm during ifind()
3405 when it finds an existing inode and then the VFS later attaches
3406 this free lsm to a new inode.
3410 Bugzilla : 6422, 7030
3411 Description: MDS deadlock between mkdir and client eviction
3412 Details : Creating a new file via mkdir or mknod (starting a transaction
3413 and getting the ns lock) can deadlock with client eviction
3414 (gets ns lock and trying to finish a synchronous transaction).
3417 Frequency : occasional
3418 Description: While starting a server, the fsfilt_ext3 module could not be
3420 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3421 kernels. Previously, lconf would still use the ext3 name
3422 when trying to load modules. Now, it will correctly use
3423 ext3 on 2.4 and ldiskfs on 2.6.
3425 Severity : enhancement
3426 Description: The default stripe count has been changed to 1
3427 Details : The interpretation of the default stripe count (0, to lfs
3428 or lmc) has been changed to mean striping across a single
3429 OST, rather than all available. For general usage we have
3430 found a stripe count of 1 or 2 works best.
3432 Severity : enhancement
3433 Description: Add support for compiling against Cray portals.
3434 Details : Conditional compiling for some areas that are different
3438 Frequency : occasional
3439 Bugzilla : 6409, 6834
3440 Description: Creating files with an explicit stripe count may lead to
3441 a failed assertion on the MDS
3442 Details : If some OSTs are full or unavailable, creating files may
3443 trigger a failed assertion on the MDS. Now, Lustre will
3444 try to use other servers or return an error to the
3448 Frequency : occasional
3450 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3451 caused lock timeout during readahead (since 1.4.2).
3452 Details : Processes doing readahead might match a lock that hasn't been
3453 granted yet if there are overlapping and conflicting lock
3454 requests. The readahead process waits on ungranted lock
3455 (original lock is CBPENDING), while OST waits for that process
3456 to cancel CBPENDING read lock and eventually evicts client.
3458 Severity : enhancement
3460 Description: Initial enabling of flock support for clients
3461 Details : Implements fcntl advisory locking and file status functions.
3462 This feature is provided as an optional mount flag (default
3463 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3464 locking are implemented yet, and those that are are not guaranteed
3465 to be completely correct in production environments.
3466 mount -t lustre -o [flock|noflock] ...
3469 Frequency : occasional
3471 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3472 assertion in the ext3 JBD (journaling) layer.
3473 Details : The b_committed_data struct is protected by the big kernel lock
3474 in 2.4 kernels, serializing journal_commit_transaction() and
3475 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3476 access to this struct. In 2.6 kernels there is finer grained
3477 locking to improve SMP performance of the JBD layer.
3481 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3483 -----------------------------------------------------------------------------
3485 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3487 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3488 versions of Lustre. Please update all servers and clients to
3489 version 1.4.2 or later at the same time. You must also run
3490 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3491 to update the configuration logs.
3493 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3494 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3495 - swab llog records of type '0' so we get proper header size/idx (5861)
3496 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3497 - fix rename of one directory over another leaking an inode (5953)
3498 - avoid SetPageDirty on 2.6 (5981)
3499 - don't re-add just-being-destroyed locks to the waiting list (5653)
3500 - when creating new directories, inherit the parent's custom
3501 striping settings if present parent (3048)
3502 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3503 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3504 - don't hold client locks on temporary worklist from l_lru (5666)
3505 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3506 - automatically evict dead clients (5921)
3507 - Update file size properly in create+truncate+fstat case (6196)
3508 - Do not unhash mountpoint dentries, do not allow removal of
3510 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3511 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3512 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3513 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3514 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3515 - Correctly handle reads of files with no objects (6243)
3516 - lctl recover will also mark a device active if deactivate used (5933)
3518 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3519 - allow --write-conf on an MDS with different nettype than client (5619)
3520 - don't write config llogs to MDS for mounts not from that MDS (5617)
3521 - lconf should create multiple TCP connections from a client (5201)
3522 - init scripts are now turned off by default; run chkconfig --on
3523 lustre and chkconfig --on lustrefs to use them
3524 - upcalls are no longer needed for clients to recover to failover
3526 - add --abort-recovery option to lconf to abort recovery on device
3528 - add support for an arbitrary number of OSTs (3026)
3529 - Quota support protocol changes.
3530 - forward compatibility changes to wire structs (6007)
3531 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3532 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3533 - improved "lustre status" script
3534 - initialize blocksize for non-regular files (6062)
3535 - added --disable-server and --disable-client configure options (5782)
3536 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3537 - Vanilla 2.4.29 support
3538 - increase maximum number of obd devices to 520 (6242)
3539 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3540 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3541 SLES 9 2.6.5 kernels
3542 - limit stripes per file to 160 (the maximum EA size) (6093)
3544 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3547 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3548 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3549 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3550 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3551 - small ext3 extents cleanups and fixes (5733)
3552 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3553 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3554 - avoid needless client->OST connect, fix handle mismatch (5317)
3555 - fix DLM error path that led to out-of-sync client, long delays (5779)
3556 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3557 - fix several locking issues related to i_size (5492,5624,5654,5672)
3558 - don't move pending lock onto export if it is already evicted (5683)
3559 - fix kernel oops when creating .foo in unlinked directory (5548)
3560 - fix deadlock in obdfilter statistics vs. object create (5811)
3561 - use time_{before,after} to avoid timer jiffies wrap (5882)
3562 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3563 - Do not leak request if server was not able to process it (5154)
3564 - If mds_open unable to find parent dir, make that negative lookup(5154)
3565 - don't create new directories with extent-mapping (5909, 5936)
3567 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3568 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3569 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3570 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3571 - support for RHEL4 kernel on i686 (5773)
3572 - provide error messages when incompatible logs are encountered (5898)
3574 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3575 * version 1.4.0.10 (1.4.1 release candidate 1)
3577 - don't keep a lock reference when lock is not granted (4238)
3578 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3579 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3580 - drop import inflight refcount on signal_completed_replay error (5255)
3581 - unlock page after async write error during send (3677)
3582 - handle missing objects in filter_preprw_read properly (5265)
3583 - no transno return for symlink open, don't save no-trasno open (3440)
3584 - don't try to complete elan receive that already failed (4012)
3585 - free RPC server reply state on error (5406)
3586 - clean up thread from ptlrpc_start_thread() on error (5160)
3587 - readahead could read extra page into cache that wasn't ejected (5388)
3588 - prevent races in class_attach/setup/cleanup/detach (5260)
3589 - don't dereference de->d_inode after l_dput of de (5458)
3590 - use "int" for stripe value returned from lock_to_stripe (5544)
3591 - mballoc allocation and error-checking fixes in 2.6 (5504)
3592 - block device patches to fix I/O request sizes in 2.6 (5482)
3593 - look up hostnames for IB nals (5602)
3594 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3595 - don't start multiple acceptors for the same port (5277)
3596 - fix incorrect LASSERT in mds_getattr_name (5635)
3597 - export a proc file for general "ping" checking (5628)
3598 - fix "lfs check" to not block when the MDS is down (5628)
3600 - service request history (4965)
3601 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3602 - create an "evict_client" /proc entry on OSTs, like the MDS has
3603 - fix mount usage message, return errors per mount(8) (5168)
3604 - change grep [] to grep "[]" in tests so they work in more UMLs
3605 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3606 - remove ancient LOV_MAGIC_V0 EA support (5047)
3607 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3608 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3609 - allow readahead allocations to fail when low on memory (5383)
3610 - mmap locking landed again, after considerable improvement (2828)
3611 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3613 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3616 - send OST transaction number in read/write reply to free req (4966)
3617 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3618 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3619 - the watchdog thread now runs as interruptible (5246)
3620 - flock/lockf fixes (but it's still disabled, pending 5135)
3621 - don't use EXT3 constants in llite code (5094)
3622 - memory shortage at startup could cause assertion (5176)
3624 - reorganization of lov code
3625 - single portals codebase
3627 - add extents/mballoc support (5025)
3628 - direct I/O reads in the obdfilter (4048)
3629 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3631 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3634 - send OST transaction number in read/write reply to free req (4966)
3635 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3636 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3637 - the watchdog thread now runs as interruptible (5246)
3638 - handle missing objects in filter_preprw_read properly (5265)
3639 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3640 - drop import inflight refcount on signal_completed_replay error (5255)
3641 - unlock page after async write error during send (3677)
3642 - return original error code on reconstructed replies (3761)
3643 - no transno return for symlink open, don't save no-trasno open (3440)
3645 - add pid to ldlm debugging output (4922)
3646 - bump the watchdog timeouts -- we can't handle 30sec yet
3647 - extra debugging for orphan dentry/inode bug (5259)
3649 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3652 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3653 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3654 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3655 - fix NULL dereference in /proc/sys/portals/routes (4827)
3656 - allow failed mdc_close() operations to be interrupted (4561)
3657 - stop precreate on OST before MDS would time out on it (4778)
3658 - don't send partial-page writes before EOF from client (4410)
3659 - discard client grant for sub-page writes on large-page clients (4520)
3660 - don't free dentries not owned by NFS code, check generation (4806)
3661 - fix lsm leak if mds_create_objects() fails (4801)
3662 - limit debug_daemon file size, always print CERROR messages (4789)
3663 - use transno after validating reply (3892)
3664 - process timed out requests if import state changes (3754)
3665 - update mtime on OST during writes, return in glimpse (4829)
3666 - add mkfsoptions to LDAP (4679)
3667 - use ->max_readahead method instead of zapping global ra (5039)
3668 - don't interrupt __l_wait_event() during strace
3670 - add software watchdogs to catch hung threads quickly (4941)
3671 - make lustrefs init script start after nfs is mounted
3672 - fix CWARN/ERROR duplication (4930)
3673 - return async write errors to application if possible (2248)
3674 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3675 - print NAL number in %x format (4645)
3676 - update barely-supported suse-2.4.21-171 series (4842)
3677 - support for sles 9 %post scripts
3678 - support for building 2.6 kernel-source packages
3679 - support for sles km_* packages
3681 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3684 - ignore -ENOENT errors in osc_destroy (3639)
3685 - notify osc create thread that OSC is being cleaned up (4600)
3686 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3687 - reconstruct ost_handle() like mds_handle() (4657)
3688 - create a new thread to do import eviction to avoid deadlock (3969)
3689 - let lconf resolve symlinked-to devices (4629)
3690 - don't unlink "objects" from directory with default EA (4554)
3691 - hold socknal file ref over connect in case target is down (4394)
3692 - allow more than 32000 subdirectories in a single directory (3244)
3693 - fix blocks count for O_DIRECT writes (3751)
3694 - OST returns ENOSPC from object create when no space left (4539)
3695 - don't send truncate RPC if file size isn't changing (4410)
3696 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3697 - bind to privileged port in socknal and tcpnal (3689)
3699 - rate limit CERROR/CWARN console message to avoid overload (4519)
3700 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3701 - basic mmap support (3918)
3702 - kernel patch series update from b1_4 (4711)
3704 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3707 - avoid crash during MDS cleanup with OST shut down (2775)
3708 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3709 - don't use bad inodes on the MDS (3744)
3710 - dynamic object preallocation to improve recovery speed (4236)
3711 - don't hold spinlock over lock dumping or change debug flags (4401)
3712 - don't zero obd_dev when it is force cleaned (3651)
3713 - print grants to console if they go negative (4431)
3714 - "lctl deactivate" will stop automatic recovery attempts (3406)
3715 - look for existing locks in ldlm_handle_enqueue() (3764)
3716 - don't resolve lock handle twice in recovery avoiding race (4401)
3717 - revalidate should check working dir is a directory (4134)
3719 - don't always mark "slow" obdfilter messages as errors (4418)
3721 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3724 - don't close LustreDB during write_conf until it is done (3860)
3725 - fix typo in lconf for_each_profile (3821)
3726 - allow dumping logs from multiple threads at one time (3820)
3727 - don't allow multiple threads in OSC recovery (3812)
3728 - fix debug_size parameters (3864)
3729 - fix mds_postrecov to initialize import for llog ctxt (3121)
3730 - replace config semaphore with spinlock (3306)
3731 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3732 - don't allow enqueue to complete on a destroyed export (3822)
3733 - down write_lock before checking llog header bitmap (3825)
3734 - recover from lock replay timeout (3764)
3735 - up llog sem before sending rpc (3652)
3736 - reduce ns lock hold times when setting kms (3267)
3737 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3738 - fix NULL deref and obd_dev leak on setup error (3312)
3739 - replace some LBUG about llog ops with error handling (3841)
3740 - don't match INVALID dentries from d_lookup and spin (3784)
3741 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3742 - fix invalid assertion in ptlrpc_set_wait (3880)
3744 - add libwrap support for the TCP acceptor (3996)
3745 - add /proc/sys/portals/routes for non-root route listing (3994)
3746 - allow setting MDS UUID in .xml (2580)
3747 - print the stack of a process that LBUGs (4228)
3749 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3752 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3753 - make sure to unset replay flag from failed open requests (3440)
3754 - if default stripe count is 0, use OST count for inode size (3636)
3755 - update parent mtime/ctime on client for create/unlink (2611)
3756 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3757 - free recovery state on server during a forced cleanup (3571)
3758 - unregister_reply for resent reqs (3063)
3759 - loop back devices mounting and status check on 2.6 (3563)
3760 - fix resource-creation race that can provoke i_size == 0 (3513)
3761 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3762 - more debugging for page-accounting assertion (3746)
3763 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3764 - avoid deadlock after precreation failure (3758)
3765 - fix race and lock order deadlock in orphan handling (3450, 3750)
3766 - add validity checks when grabbing inodes from l_ast_data (3599)
3768 - add /proc/.../recovery_status to obdfilter (3428)
3769 - lightweight CDEBUG infrastructure, debug daemon (3668)
3770 - change default OSC RPC parameters to be better on small clusters
3771 - turn off OST read cache for files smaller than 32MB
3772 - install man pages and include them in rpms (3100)
3773 - add new init script for (un)mounting lustre filesystems (2593)
3774 - run chkconfig in %post for init scripts (3701)
3775 - drop scimac NAL (unmaintained)
3777 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3780 - clean kiobufs before and after use (3485)
3781 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3782 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3783 - change init script to fail more clearly if not run as root (1528)
3784 - allow clients to reconnect during replay (1742)
3785 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3786 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3787 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3788 - fix cond_resched (3554)
3789 - only evict unfinished clients after recovery (3515)
3790 - allow bulk resend, prevent data loss (3570)
3791 - dynamic ptlrpc request buffer allocation (2102)
3792 - don't allow unlinking open directory if it isn't empty (2904)
3793 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3794 - remove extraneous obd dereference causing LASSERT failure (3334)
3795 - don't use get_cycles() when creating temp. files on the mds (3156)
3796 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3797 - handle EEXIST for set-stripe, set proper directory name (3336)
3799 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3800 - fix ksocknal_fmb_callback() error messages (2918)
3802 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3805 - don't copy lvb into (possibly NULL) reply on error (2983)
3806 - don't deref dentry after dput, don't free lvb on error (2922)
3807 - use the kms to determine writeback rpc length (2947)
3808 - increment oti_logcookies when osc is inactive (2948)
3809 - update client's i_blocks count via lvb messages (2543)
3810 - handle intent open/close of special files properly (1557)
3811 - mount MDS with errors=remount-ro, like obdfilter (2009)
3812 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3813 - don't use cancelling-locks' kms values (2947)
3814 - use highest lock extent for kms, not last one (2925)
3815 - don't dereference ERR_PTR() dentry in error handling path (3107)
3816 - fix thread race in portals_debug_dumplog() (3122)
3817 - create lprocfs device entries at setup instead of at attach (1519)
3818 - common AST error handler, don't evict client on completion race (3145)
3819 - zero nameidata in detach_mnt in 2.6 (3118)
3820 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3821 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3822 - initialize RPC timeout timer earlier for 2.6 (3219)
3823 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3824 - print nal/nid for unknown nid (3258)
3825 - additional checks for oscc recovery before doing precreate (3284)
3826 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3827 - don't crash in mdc_close for bad permissions on open (3285)
3828 - zero i_rdev for non-device files (3147)
3829 - clear page->private before handing to FS, better assertion (3119)
3830 - tune the read pipeline (3236)
3831 - fix incorrect decref of invalidated dentry (2350)
3832 - provide read-ahead stats and refine rpc in flight stats (3328)
3833 - don't hold journal transaction open across create RPC (3313)
3834 - update atime on MDS at close time (3265)
3835 - close LDAP connection when recovering to avoid server load (3315)
3836 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3837 - don't leak open file on MDS after open resend (3325)
3838 - serialize filter_precreate and filter_destroy_precreated (3329)
3839 - loop device shouldn't call sync_dev() for nul device (3092)
3840 - clear page cache after eviction (2766)
3841 - resynchronize MDS->OST in background (2824)
3842 - refuse to mount the same filesystem twice on same mountpoint (3394)
3843 - allow llmount to create routes for mounting behind routers (3320)
3844 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3845 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3846 - fix rare problem with rename on htree directories (3417)
3848 - allow default OST striping configuration per directory (1414)
3849 - fix compilation for qswnal for 2.6 kernels (3125)
3850 - increase maximum number of MDS request buffers for large systems
3851 - change liblustreapi to be useful for external progs like lfsck (3098)
3852 - increase local configuration timeout for slow disks (3353)
3853 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3855 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3858 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3859 - don't overwrite extent policy data in reply if lock was blocked (2901)
3860 - drop filter export grants atomically with removal from device (2663)
3861 - del obd_self_export from work_list in class_disconnect_exports (2908)
3862 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3863 - swab reply message in mdc_close, other PPC fixes (2464)
3864 - fix destroying of named logs (2325)
3865 - overwrite old logs when running lconf --write_conf (2264)
3866 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3867 - fix race in target_handle_connect (2898)
3868 - mds_reint_create() should take same inode create lock (2926)
3869 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3870 - don't close files for self_export to avoid uninitialized obd (2936)
3871 - allow MDS with the same name as client node (2939)
3872 - hold dentry reference for closed log files for unlink (2325)
3873 - reserve space for all logs during transactions (2059)
3874 - don't evict page beyond end of stripe extent (2925)
3875 - don't oops on a deleted current working directory (2399)
3876 - handle hard links to targets without a parent properly (2517)
3877 - don't dereference NULL lock when racing during eviction (2867)
3878 - don't grow lock extents when lots of conflicting locks (2919)
3880 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3883 - account for cache space usage on clients to avoid data loss (974)
3884 - lfsck support in lustre kernel code (2349)
3885 - reduce journal credits needed for BRW writes (2370)
3886 - orphan handling to avoid losing space on client/server crashes
3887 - ptlrpcd can be blocked, stopping ALL progress (2477)
3888 - use lock value blocks to assist in proper KMS, faster stat (1021)
3889 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3890 - recovery for initial connections (2355)
3891 - fixes for mds_cleanup_orphans (1934)
3892 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3893 - block all file creations until orphan recovery completes (1901)
3894 - client remove rq_connection from request struct (2423)
3895 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3896 - recovery timer race (2670)
3897 - mdc_close recovey bug (2532)
3898 - ptlrpc cleanup bug (2710)
3899 - mds timeout on local locks (2588)
3900 - namespace lock held during RPCs (2431)
3901 - handle interrupted sync write properly (2503)
3902 - don't try to handle a message that hasn't been replied to (2699)
3903 - client assert failure during cleanup after abort recovery (2701)
3904 - leak mdc device after failed mount (2712)
3905 - ptlrpc_check_set allows timedout requests to complete (2714)
3906 - wait for inflight reqs when ptlrpcd finishes (2710)
3907 - make sure unregistered services are removed from the srv_list
3908 - reset bulk XID's when resending them (caught by 1138 test)
3909 - unregister_bulk after timeout
3910 - fix lconf error (2694)
3911 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3912 - readahead locks pages, leaves pending causing memory pressure (2673)
3913 - increase OST request buffers to 4096 on large machines (2729)
3914 - fix up permission of existing directories in simple_mkdir (2661)
3915 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3916 - don't assign transno to errored transactions (2742)
3917 - don't delete objects on OST if given a bogus objid from MDS (2751)
3918 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3919 - if rq_no_resend, then timeout request after recovery (2432)
3920 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3921 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3922 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3923 - MDS recovery shouldn't delete objects that it has given out (2730)
3924 - if enqueue arrives after completion, don't clobber LVB (2819)
3925 - don't unlock pages twice when trigger_group_io returns error (2814)
3926 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3927 - don't write pages to disk if there was an error (1450)
3928 - don't ping imports that have recovery disabled (2676)
3929 - take buffered bytes into account when balancing socknal conn (2817)
3930 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3931 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3932 - remove little-endian swabbing of llog records (1987)
3933 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3934 - retry reposting request buffers if they fail (1191)
3935 - grow extent at grant time to avoid granting a revoked lock (2809)
3936 - lock revoke doesn't evict page if covered by a second lock (2765)
3937 - disable VM readahead to avoid reading outside lock extents (2805)
3939 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3940 - updated kernel patches for hp-2.4.20 kernel (2681)
3942 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3945 - fix truncated write corruption (2366)
3946 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3948 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3949 - ensure len(uuid) < 37 in lmc (1171)
3950 - fix ia64 OOPS in llog_test (2255)
3951 - zero end of page at obdfilter for partial page writes (2648)
3952 - don't leave stale dentries around after renames (bug 2428)
3953 - fix timeouts when evicting a client with a single lock held (2642)
3954 - set deadline for the initial HELLO message to drain (2634)
3955 - print out dotted-quad IP addresses in the socknal (2302)
3957 - additional debugging for MDS client eviction problem (2443)
3958 - fix mkfsoptions support for osts (2603, 2604)
3960 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3963 - add series for the vanilla 2.6.0 kernel
3964 - add series for the vanilla 2.4.24 kernel
3965 - add series for a cray x86/64 UL kernel drop
3966 - fix xattr patches for the vanilla 2.4.19 series
3968 - generate true UUIDs in lmc (1171)
3969 - have portals stack dumping break in UML (2466)
3970 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3971 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3972 - wake callback waiting threads on client eviction (2460)
3973 - Add --ptldebug and --subsystem to lmc (1719)
3974 - update assertion to allow safe interrupt allocation
3975 - set rq_no_resend for cancel requests (2432)
3976 - recalculate ptlrpcd timeout after resend (2494)
3977 - call vfs_rmdir when removing pending directories (2368)
3978 - fix renaming a file to itself (2429)
3979 - lmc creates a default one-stripe lov (2454)
3980 - expand procfs space to handle large clusters (2326)
3981 - increase UML stack to avoid overflow
3982 - update lconf's list of debug and subsystem masks
3983 - fix lfs find --obd (2510)
3984 - /proc tunable for disabling filter read caching (2591)
3985 - stop rpm packages from altering slapd.conf (2301)
3986 - disable nagle in the socknal under 0conf (2578)
3987 - choose mds inode size based on stripe count (2572)
3988 - fix kernel-source rpm problems (2516)
3990 - add --disable-doc to avoid pdf generation (2421)
3991 - update documentation, tests, type-os, comments
3992 - avoid format warnings on ia64
3993 - remove the TOE NAL
3994 - tiny code cleanups by removing unused fields
3996 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3999 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4000 - remove the most blatant lies from BUILDING file (2371)
4001 - change default debug level to reasonable production setting
4002 - reduce client side cache size to reduce cache flush time
4003 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4004 - make TCP zerocopy and pinger support enabled by default (2476)
4005 - sync writes completed after process exits caused crashes (2319)
4006 - maintain correct mount count on the MDS (2356)
4007 - backout 1557, because 2316 wasn't really fixed
4008 - better file I/O statistics gathering in /proc
4009 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4010 - another kernel patch to fix zero-copy TCP function export
4011 - don't take duplicate lock when processing re-sent getattr (2420)
4012 - lctl uses obd_self_export instead of creating new conn (2353)
4013 - MDS/OST recovery case which requires object creation asserted (2425)
4014 - move lfs from /usr/sbin to /usr/bin in packages
4015 - fix race between mds_client_add and mds_client_free (2417)
4016 - use kmalloc instead of slabs in portals (2430)
4017 - don't create duplicate records when a failover MDS is present (2442)
4018 - remove unnecessary mount age check (2332)
4019 - don't remove directory inodes from locks prematurely (2451)
4020 - don't break if MDS service name is the same as hostname (2103)
4021 - fix races in client write RPC generation when cache full (2482)
4023 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4026 - remove now-unused request->rq_obd (278)
4027 - if an allocation fails, print out how much memory we've used (1933)
4028 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4029 - add the "configurable stack size" patch to most series files (1256)
4030 - ability to write large log records, for 100+ OST configs (2306)
4031 - fix NULL deref when filter_prep fails (2314)
4032 - fix operator precedence error in filter_sync
4033 - dynamic allocation of socknal TX descriptors (2315)
4034 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4035 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4036 - ensure that CWARN messages in Portals always get to the syslog
4037 - __init/__exit are not for prototype decls (ldlm_init/exit)
4038 - x86-64 compile warning fixes
4039 - fix gateway LMC keyword conflict (2318)
4040 - fix MDS lock inversions in getattr/reint paths (1844)
4041 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4042 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4043 - DLM race condition prevented some lock evictions (2328)
4044 - ENOMEM detection and retry on socknal sends (2230)
4045 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4046 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4047 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4048 - better error messages when a client is rejected during recovery (1505)
4049 - avoid cancelling locks which were never granted, after failure (2330)
4050 - fix i_sem/journal inversion in mds_client_add (2333)
4051 - fix truncate/getattr lock cycle deadlock (2334)
4052 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4053 - fix two rare exit paths which could leak an l_lock() ref (2321)
4054 - fencepost error in MDS/OST orphan recovery (2226)
4055 - make log record alignment 8 bytes (1988)
4056 - lstripe now fails when requested offset > ost_count (2237)
4057 - ensure that all kernel series have a complete list.h (1607)
4058 - fix crashes in special-file operations (2316)
4059 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4061 - allow configurable automake binary, for testing new versions
4062 - small update to the lfs documentation
4064 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4066 * fix negative export reference count in fsfilt_sync (2312)
4068 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4069 * release candidate 0.9.1
4071 - orphans are moved into the PENDING directory for possible recovery
4072 - replayed opens now open by fid for orphan/rename safety (1042)
4073 - last close of an orphan inode generates a transno (683)
4074 - chdir() and mount() now pin the directory entry (1020)
4075 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4076 - discard very old requests without processing them (1502)
4077 - remove some common, well-understood CERRORs (1505)
4078 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4079 - clear "grant" flags in OST replies until OST grant code lands (1644)
4080 - fix read performance by not clobbering i_blksize on client (1598)
4081 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4082 - make LRU size a /proc tunable, clears locks when reduced (707)
4083 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4084 - server locks take references on exports now (1558)
4085 - build fixes for 2.4.20-rh trees (1663)
4086 - return an error from lov_create if all OSCs are inactive (1751)
4087 - fix import levels when a reconnect happens without a timeout (1597)
4088 - exit early from mds_open if we get a lookup error (1749)
4089 - partial page read at EOF wouldn't wait for disk before sending (1642)
4090 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4091 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4092 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4093 - write/truncate lock inversion (1639)
4094 - Don't auto-load obdclass, portals modules during cleanup (1495)
4095 - fix timestamps from jumping to "now" (1763)
4096 - extra journal assertions (1648)
4097 - add an extra multiunlink test (1771)
4098 - fix read_record/write_record API (1776)
4099 - fix leak of offset_extent, possible incorrect i_size later (1772)
4100 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4101 - Debugging for the kqswnal_get_idle_tx problems (1820)
4102 - Allow recovery to be attempted multiple times (1536)
4103 - Write out MDS last_rcvd file after it is first created (1600)
4104 - Fix tx_descriptor leak in failed transmit situations (1827)
4105 - ext3 journaling fixes for assertion failure after IO error (1871)
4106 - class_export_put() on freed export after completion AST error (1896)
4107 - Fix revalidate looping in VFS (1322)
4108 - Don't access a freed export during MDS_REINT timeout (1521)
4109 - Add open-unlink recovery support on the MDS (1673,1764)
4110 - Return an error if no MDS data was read from last_rcvd (1946)
4111 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4112 - Don't setup a disk device that doesn't match exported UUID (317)
4113 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4114 - avoid committing NULL handle in force close
4115 - local.sh is now a one-stripe LOV configuration
4116 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4117 - fix ext3 htree duplicate directory entry corruption (1516)
4118 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4119 - update to new LOV EA format (2097)
4120 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4122 2003-06-15 Phil Schwan <phil@clusterfs.com>
4125 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4126 - per-import/export recovery handling (958, 931, 959)
4127 - multiple last-rcvd slots, for serving multiple FSes (949)
4128 - connections are again shared between multiple imp/exports (963, 964)
4129 - "umount -f" would hang if any requests needed to be sent (393, 978)
4130 - avoid pinning large req buffer by copying for queued messages (989)
4131 - add "uuid" to "lctl device" command to help upcalls (991)
4132 - "open" RPCs with transnos would confuse recovery counters (1037)
4133 - do proper endian conversion of all wire messages (288, 340, 891)
4134 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4135 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4136 - fix ptlrpc_connection leak in target_handle_connect (1174)
4137 - fix import refcounting bug in OST and MDS cleanup (1134)
4138 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4139 - fix very unlikely obd_types race condition (501)
4140 - remove osc_open hack for echo_client (1187)
4141 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4142 - a failure in read_inode2 leads to deadlock (1139)
4143 - cancel ack-locks as soon as transaction is committed (1072)
4144 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4145 - make sure to commitrw after any preprw to avoid deadlock (1162)
4146 - failing to execute a file in a lustre FS would lock inode (1203)
4147 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4148 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4149 - obd_uuid2tgt would walk off the end of the list (1255)
4150 - on IA64 the osc would give portals incorrect bulk size (1258)
4151 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4152 - fix lock inversion caused by new llite matching code (1282)
4153 - limit the number of dirty pages on a client to 10MB (1286)
4154 - timed out locks were not being corrected cancelled (1289)
4155 - fix O_DIRECT above 4GB on IA-32 (1292)
4156 * major user-visible changes
4157 - fail out/fail over policy now controlled by the upcall (993)
4159 - add OBD_PING to check server availability and failure (954)
4160 - lustre messages are now sent in sending host order (288, 340, 891)
4161 - add eadatalen to MDS getattr reply (340)
4162 - OST read replies may contain second buffer, with per-page status (593)
4164 2003-03-11 Phil Schwan <phil@clusterfs.com>
4167 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4168 - failed open()s could cause deadlock; fixed (867, 869)
4169 - stop cancelling OST locks when files are closed (481)
4170 - overlapping XID spaces caused network corruption (851, 853)
4171 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4172 - setattr_raw added, to do single-RPC, server-side setattrs
4173 - lmc/lconf syntax change for OST UUIDs
4174 - fix crashy race condition between ptlrpc_free_req and osc_close
4175 - don't use request in mdc_enqueue if we hit a timeout (889)
4176 - don't set the inode i_size for regular files from the MDS (896)
4177 - handle out of order completion AST (842)
4178 - don't LBUG if a lock request times out after receiving AST (913)
4179 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4180 - if a bad lock AST arrives, send an error instead of dropping entirely
4181 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4182 - fix leak in bulk IO when only partially completed (899, 900, 926)
4183 - fix O_DIRECT for ia64 (55)
4184 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4185 - C-z after timeout could hang a process forever; fixed (977)
4187 - client-side I/O cache (678, 924, 929, 941, 970)
4189 - READPAGE and SETATTRs which don't take server-side locks get
4192 2003-02-11 Phil Schwan <phil@clusterfs.com>
4195 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4196 - Fix fsfilt_extN_readpage() to read a full page of directory
4197 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4198 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4199 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4200 - Most intent operations were removed, in favour of a new RPC mode
4201 that does a single RPC to the server and bypasses most of the VFS
4202 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4203 - Aggressively cancel local locks on DLM servers
4204 - mds_reint_unlink sends EA to the client if it's the last nlink.
4205 client uses that EA to unlink OST objects.
4206 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4207 - recursive symlinks were fixed (439)
4208 - fixed NULL deref in DEBUG_REQ
4209 - filter_update_lastobjid no longer calls sync, which annoyed extN
4210 - fixed multi-client small-writes to a single file problem (445)
4211 - fixed mtime updates during file writes (607)
4212 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4213 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4214 - fixed Linux OST size reporting problem (444, 656)
4215 - OST now updates object mtime with writes or setattr (607, 619)
4216 - client verifies file size before zeroing page past EOF (445)
4217 - OST now writes last allocated objid to disk with allocation (108)
4218 - LOV on echo now works (409)
4220 - mds_reint_unlink sends a new buffer, with the EA included. this
4221 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4222 set if a regular file was being unlinked, and it was the last link
4223 - use PtlGet from the target for bulk writes (315)
4224 - OST now updates object mtime with writes or setattr (607, 619)
4225 - LDLM now has a grant-time callback to revalidate locked items, if
4227 - Many MDS operations were reorganized to combat race conditions
4229 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4230 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4232 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4235 - Fully reactivate OST imports after reconnection (512, others)
4236 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4237 - More graceful error handling for truncating on dead OST (515)
4238 - Don't error out unless we're actually accessing dead stripes (474)
4239 - Fix garbage sizes when stripes are missing (410)
4240 - LRU counters were broken, causing constant lock purge (433, 432)
4241 - garbage on read from stripes with failed OSTs (441)
4242 - mark OSCs as active before reconnecting during recovery (438)
4243 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4244 - lfind did not preserve OST order in output (443)
4245 - symlinks cause hung clients, incorrect data (439)
4246 - stop dereferencing request after dropping refcount (457)
4247 - don't LASSERT(spin_is_locked) on non-SMP (455)
4248 - fixes for many rename() bugs
4249 - fstat didn't correctly synchronize attributes (399)
4250 - server must handle lock cancellation during blocking AST prep (487)
4251 - bulk descriptors were free()d too soon (511)
4252 - fix paths in lconf, which would load incorrect modules (451, 507)
4253 - fix confusing lconf 'host not found' error message (386)
4254 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4255 - fix race condition in mdc_blocking_ast() for inode access (526)
4256 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4257 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4258 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4259 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4260 - fix lstripe in several ways (kernel side) (527)
4261 - fix request leak in ldlm_cli_enqueue (262)
4262 - incorrect OSC was marked inactive after OST failure
4263 - call mds_fs_cleanup before unmounting filesystem (524)
4264 - fix races between taking ns_lock and ldlm_lock_change_resource
4265 - fix races updating LOV export open file list
4266 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4267 - fix recovery NULL deref in ldlm_cli_cancel_unused
4268 - fix some DLM races by using new hash table for lock handles (419)
4269 - permit the client to specify desired inodes, at replay
4270 - duplicate requests when we queue them for replay reintegration
4271 - fix last_rcvd offset calculation
4272 - sync after each recovered transaction, so we always make progress
4273 - never, not always, ERESTART requests without transnos
4274 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4276 - skip replay if the MDS says that the client is already connected
4277 - don't check for a recovery-enabled export to match lctl's UUID
4278 - don't INC_USE_COUNT for phantom exports
4279 - don't crash when cleaning up phantom exports (567)
4280 - don't double-finish or set replay data for errored mdc_open requests
4281 - abort requests when they time out, so we don't get old replies
4282 - send/receive replies for AST messages again
4283 - if the client says that it doesn't have the lock, cancel it on the
4285 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4286 mark it as destroyed, it will all work out when decref is called
4287 - fix module use counts (22, 581)
4289 - ASTs now expect a reply (server cancels lock on error reply)
4291 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4294 - fix many simultaneous client startup (392)
4295 - fix dentry->d_it clobbering
4296 - credentials weren't being shipped for readdir/getattr operations
4297 - remove invalid assertions triggered during some concurrent MD
4299 - proper Lustre versions added (336, 389)
4300 - fix memory leak for create error case (398)
4301 - fix LOV locking bug that would get cli/srv out of sync
4302 - fix echo client over LOV (409)
4303 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4304 - fix double-O_EXCL intent crash (424)
4305 - avoid sending multiple lock CANCELs (352)
4307 - MDS can do multi-client recovery (modulo bugs in new code)
4309 - many updates, edits, cleanups
4311 2002-11-18 Phil Schwan <phil@clusterfs.com>
4314 - fix null d_it dereference (346)
4315 - fix full OST/dbench hang (333)
4316 - fix permission problem with file removal (286)
4317 - fix removal of OSCs from LOV when they fail
4318 - fix NULL deref during bulk timeout (214)
4319 - fix problems related to multiple filesystems on one MDS (241)
4320 - fixed serious subtle metadata locking bugs
4321 - free locks on clients when inodes are removed due to memory
4323 - fix inode pointer in lock data (285)
4324 - partial support for multiple MDS on a single host (241)
4325 - data locks weren't cancelled at clear_inode time (290, 311)
4326 - intent locks could lead to unbounded lock growth (205)
4327 - added a maximum lock count, an LRU list, and a flusher
4328 - fix multiple rename (365)
4329 - properly abstracted the echo client
4330 - OSC locked 1 byte too many; fixed
4331 - rewrote brw callback code:
4332 - fixed recovery bugs related to LOVs (306)
4333 - fixed too-many-pages-in-one-write crash (191)
4334 - fixed (again) crash in sync_io_timeout (214)
4335 - probably fixed callback-related race (385)
4337 - Add capability to MDS protocol
4338 - LDLM cancellations and callbacks on different portals
4340 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4343 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4344 - defer open object destruction to close time (601981, 138)
4345 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4346 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4347 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4348 - don't call obd_setattr in ll_file_release for destroyed objects
4349 * protocol change to lustre_msg: move |version| and add |flags|
4350 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4351 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4352 * added replay of create, unlink, link and rename operations during
4353 MDS failover; recovery should be much more robust now
4354 * remove failed OSCs from LOVs (only lov_create uses this so far)
4355 * the lustre-HOWTO was brought (more) up to date (582544)
4357 2002-10-23 Phil Schwan <phil@clusterfs.com>
4360 - in-use dentries weren't being reused properly (617851)
4361 - prevent multiple LDLM setup (599178)
4362 - fix LOV size calculations for truncate (617853)
4363 - fix client handling of MDS intent errors (POSIX)
4364 - fix permission bug in lovstripe.c test (624321)
4365 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4366 - truncate past end of file could corrupt data
4367 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4368 - a race in recovery could return ETIMEDOUT to apps (623947)
4369 - building outside the source directory was fixed
4370 * the lustre-HOWTO was brought (more) up to date (582544)
4371 * major progress was made on recovery functionality
4373 2002-10-10 Phil Schwan <phil@clusterfs.com>
4376 - recovery deadlock fix
4377 - rm -rf causes LBUG fix (617817)
4378 - file open by multiple tasks fix (618962)
4379 - directory permissions bugs (602707 and 620007)
4380 - journal_stop fixed with locking (611313)
4381 - O_APPEND failures resolved (618273, perhaps 614459)
4382 - lconf PATH fix (619770)
4383 - IA64 build fix (621450)
4384 - RPC buffer sizes scale with amount of memory
4386 2002-10-01 Phil Schwan <phil@clusterfs.com>
4389 - locks would be cancelled without throwing away data pages,
4390 resulting in inconsistent data (605627)
4391 - inode attributes were not always being refreshed (605627, 612449)
4392 - lconf now continues to cleanup after lctl reports an error
4393 - MDS now enforces user permissions (602707)
4394 - lprocfs cleanup fixed, but not yet enabled (614157)
4395 - fixed infinite server hang, should a client not respond to an AST
4396 - avoid going into recovery if user calls readlink() with a buffer
4397 that's too small (613941)
4398 - AST RPCs no longer require replies (614867) -- this may be changed
4399 - don't crash server if client sends an IOV that's too big (611336)
4400 - fixed lock conversion deadlock (611892)
4401 - fixed the following of symlinks (614622)
4402 * recovery: the server can remove locks from a client that dies, other
4403 clients can make progress
4404 * more extN patch fixes
4405 * compile-time configurable ptlrpc buffer allocations
4407 - collaborative read cache document
4408 - Lustre Lite Performance CDR document-in-progress
4410 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4413 - fix typo in patch-2.4.18
4415 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4418 - clear ptlrpc request each time in handle_incoming_request()
4419 - unlink of files now destroys the object on the OST
4421 2002-09-19 Peter Braam <braam@clusterfs.com>
4423 * add hard link support
4424 * change obdfile creation method
4425 * kernel patch changed
4427 2002-09-19 Peter Braam <braam@clusterfs.com>
4430 - stack overflow bug in extN fixed
4432 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4434 * documentation updates
4435 - add man pages for config tools
4436 - update tests/README to describe testing with new config tools
4437 - finish metadata API descriptions
4438 * bug fixes and cleanups
4439 - statfs workaround for 16TB limit
4440 - LOV stripe allocation improved, can stripe on subset of OSTs
4441 - LOV file size/IO offset was wrong for files > 4GB in size
4442 - object EA data was being dropped, caused files to be unreadable
4443 - memory overflow with non-LOV OST caused memory corruption
4444 - fixed regression tests to work with new config tools, obdfilter
4445 - fixed bug when directory size became larger than 1 block
4446 - fixed bug (for single client case) when PWD was deleted
4447 - invalidate local directory pages when doing intent-based ops
4448 - avoid LDLM oops when lock callback contained bad data
4450 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4452 * documentation updates
4453 * bug fixes and cleanups
4454 - configuration tools
4457 - 64-bit compile warnings
4458 - 64-bit internal statfs data
4460 * test_brw on persistent OST devices
4462 * lprocfs (disabled)
4464 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4466 * documentation updates
4467 * bug fixes and cleanups
4468 * configuration tools
4470 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4473 * many small fixes to 0_5_4
4474 * io/network handling
4475 * thinkos in MDS operations
4477 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4480 * crucial basic fixes to 0.5.3
4481 * IOR, Iozone work over Elan
4484 2002-08-07 Phil Schwan <phil@clusterfs.com>
4485 * version 0_5_3, our first alpha
4486 * we use the new Portals iovs
4487 * documentation updates
4488 * bug fixes and cleanups
4489 * small changes in the DLM wire protocol
4491 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4492 * version 0_5_1 with some initial stability,
4493 * locking on MD and file I/O.
4494 * documentation updates
4495 * several bug fixes since 0.5.0
4496 * small changes in wire protocol
4498 2002-07-18 Phil Schwan <phil@clusterfs.com>
4500 * delivered as Lustre Light Alpha
4501 * fixed a crash after handling invalid MDS requests
4502 * fixed directory pages for architectures with non-4k pages sizes
4504 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4505 * release version v0_4_4
4506 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4508 * Fixes a number of interruption problems with OST operations.
4509 * Update documentation for portals header changes
4510 * Move all wire protocol structs/defines to lustre_idl.h
4511 * Fixes symlink length bug.
4512 * Add tcpdump to repository.
4514 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4515 * release version v0_4_3
4516 * Fixes statfs for inodes on extN.
4517 * Fixes bug in runtests which would delete /etc/hosts.
4518 * Use 64-bit object IDs wherever possible (not into VFS though)
4519 Remove ost_get_info, which is unused by lustre, and out of date.
4521 2002-07-03 Peter Braam <braam@clusterfs.com>
4522 * release version v0_4_2 Fixes a lookup error (type not passed)
4523 * move forward to head of Portals
4524 * move forward to latest Lustre kernel
4526 2002-06-25 Peter Braam <braam@clusterfs.com>
4527 * release version v0_4_1. Hopefully stable on single node use.