4 2.6.9-55.0.9.EL (RHEL 4),
5 2.6.16.53-0.8 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 Severity : enhancement
17 Description: Build SLES10 patchless client fails
18 Details : The configure was broken by run ./configure with
19 --with-linux-obj=.... argument for patchless client. When the
20 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
21 can't search header adequately. Use absolute path such as
22 -I($LINUX)/include instead.
26 Description: interrupt oig_wait produce painc on resend.
27 Details : brw_redo_request can be used for resend requests from ptlrpcd and
28 private set, and this produce situation when rq_ptlrpcd_data not
29 copyed to new allocated request and triggered LBUG on assert
30 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
31 ptlrpcd set if request is changed and can be safety changed to use
36 Description: LASSERT_{REQ,REP}SWAB macros are buggy
37 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
38 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
39 these macros and replace them with their difinitions instead.
43 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
44 Details : Update which_patch & target file for SLES10 latest kernel.
47 Description: Remove client patches from SLES 10 kernel.
48 Details : This causes SLES 10 clients to behave as patchless clients
49 even on a Lustre-patched (server) kernel.
51 Severity : enhancement
53 Description: self-adjustable client's lru lists
54 Details : use adaptive algorithm for managing client cached locks lru
55 lists according to current server load, other client's work
56 pattern, memory activities, etc. Both, server and client
57 side namespaces provide number of proc tunables for controlling
60 Severity : enhancement
62 Description: light-weight GSS support
63 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
64 and reduce performance overhead.
66 Severity : enhancement
68 Description: Linux keyring support
69 Details : Support using service of Linux keyring for Lustre GSS internal
70 context refresh/cache mechanism.
74 Description: Fix errors in lfs documentation
75 Details : Fixes man pages
79 Description: don't use GFP_* in generic Lustre code.
80 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
84 Description: obdclass is limited by single OBD_ALLOC(idarray)
85 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
89 Description: warnings with build patchless client with vanila 2.6.19 and up
90 Details : change old ctl_table style and replace ctl_table/ctl_table_header
91 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
95 Frequency : Only for SLES
96 Description: sanity_quota fail test_1
97 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
98 file, which makes TSTID[2] inunique.
102 Frequency : Only for SLES9
103 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
104 Details : Update target/ChangeLog/which_patch .
108 Frequency : Only for test_10 in sanity-quota.sh
109 Description: a bug in quota test code
110 Details : checking a test flag in a wrong place
114 Frequency : Only for test_13 in sanity-quota.sh
115 Description: a bug in quota test script
116 Details : shouldn't check overfull allocated quota
120 Description: jbd statistics
121 Details : Port older jbd statistics patch for sles10
123 Severity : enhancement
125 Description: Add printing inode info into message about error in writepage.
130 Description: reply_lock_interpret crash due to race with it and lock cancel.
131 Details : Do not replay locks that are being cancelled. Do not reference
132 locks by their address during replay, just by their handle.
136 Frequency : When flocks are used.
137 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
138 Details : Flock locks might destroy just granted lock if it could be merged
139 with another existing flock, this is done in completion handler,
140 so teach ldlm_cli_enquque_fini that this is a valid case for
146 Description: Oops in read and write path when failing to allocate lock.
147 Details : Check if lock allocation failed and return error back.
151 Description: lstripe command fails for valid OST index
152 Details : The stripe offset is compared to lov->desc.ld_tgt_count
153 instead of lov->desc.ld_active_tgt_count.
157 Description: sanity.sh failed test 103
158 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
163 Description: df doesn't work properly if diskfs blocksize != 4K
164 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
168 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
169 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
173 Description: obd_health_check_timeout too short
174 Details : set obd_health_check_timeout as 1.5x of obd_timeout
178 Description: llapi_file_create() does not allow some changes
179 Details : add llapi_file_open() that allows specifying the mode and
180 open flags, and also returns an open file handle.
184 Description: merge and cleanup kernel patches.
185 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
189 Description: Add journal checksum support.(Kernel part)
190 Details : The journal checksum feature adds two new flags i.e
191 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
192 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
193 indicates that the commit block contains the checksum for
194 the blocks described by the descriptor blocks. Now commit
195 record can be sent to disk without waiting for descriptor
196 blocks to be written to disk. This behavior is controlled
197 using JBD2_FEATURE_ASYNC_COMMIT flag.
201 Description: OSS needs mutliple precreate threads
202 Details : Add ability to start more than one create thread per OSS.
206 Description: Sanity.sh test_65k failed: setstripe should have succeeded
207 Details : execute lfs setstripe on client
211 Description: mds_obd_create error creating tmp object
212 Details : When the user sets quota on root, llog will be affected and can't
213 create files and write files.
216 Frequency : Always on ia64 patchless client, and possibly others.
218 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
219 Details : This allows the patchless client to be loaded on architectures
224 Description: RedHat Update kernel for RHEL5
225 Details : Add the kernel config file for RHEL5.
229 Description: RedHat Update kernel for RHEL5
230 Details : Modify the kernel config file more closer RHEL5.
234 Description: Build failure against Centos5 (RHEL5)
235 Details : Define PAGE_SIZE when it isn't present.
239 Description: test_7 on sanity-quota.sh can't be used on separate machines
240 Details : add facet to handle it
244 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
245 Details : with NFS, the anon dentry's parent was set to itself in
246 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
247 corresponding dentry other than use rec->ur_name.
249 Severity : enhancement
251 Description: lfs setstripe enhancement
252 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
256 Description: testing performance impact of enabling checksumming
257 Details : enable checksum by default, allow --disable-checksum
258 configure option and "-o nochecksum" mount option
262 Description: lustre support for RHEL5
263 Details : Add support for RHEL5.
267 Description: Client eviction due to failover config
268 Details : after a connection loss, the lustre client should attempt to
269 reconnect to the last active server first before trying the
270 other potential connections.
274 Description: when mds and osts use different quota unit(32bit and 64bit),
275 quota will be released repeatly.
276 Details : void sending multiple quota reqs to mds, which will keep the
277 status between the reqs.
281 Description: osts not allocated evenly to files
282 Details : change the condition to increase offset_idx
285 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
287 Description: Data corruption for OSTs that are formatted larger than 2TB
289 Details : When generating the bio request for lustre file writes the
290 sector number would overflow a temporary variable before being
291 used for the IO. The data reads correctly from Lustre (which
292 will overflow in a similar manner) but other file data or
293 filesystem metadata may be corrupted in some cases.
297 Description: Tune the kernel for good SCSI performance.
298 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
299 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
304 Description: rewrite ext2-derived code in obdclass/uuid.c
305 Details : rewrite inherited code (uuid parsing code from ext2 utils)
306 from scratch preserving functionality.
310 Description: "lfs find -obd UUID" prints directories
311 Details : "lfs find -obd UUID" will return all directory names instead
312 of just file names. It is incorrect because the directories
313 do not reside on the OSTs.
317 Description: use i_size_read and i_size_write in 2.6 port
318 Details : replace inode->i_size access with i_size_read/write()
321 Frequency : when removing large files
323 Description: scheduling issue during removal of large Lustre files
324 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
325 It causes scheduling issues when removing large files (17TB in the
329 Frequency : only with liblustre clients on XT3
331 Description: evictions taking too long
332 Details : allow llrd to evict clients directly on OSTs
335 Frequency : only on ppc
337 Description: /proc/fs/lustre/devices broken on ppc
338 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
339 We should make sure the type of 'index' is loff_t and then cast
340 explicitly as needed below. Do not assign an explicitly cast
344 Frequency : only for rhel5
346 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
347 Details : Modify the target file & which_kernel.
349 Severity : enhancement
351 Description: omit set fsid for export NFS
352 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
353 as FS_REQUIRES_DEV for avoid problems with generate fsid.
357 Frequency : Always, for kernels after 2.6.16
358 Description: Fix warning idr_remove called for id=.. which is not
360 Details : Last kernels save old s_dev before kill super and not allow
361 to restore from callback - restore it before call kill_anon_super.
365 Description: buffer overruns could theoretically occur
366 Details : llapi_semantic_traverse() modifies the "path" argument by
367 appending values to the end of the origin string, and a
368 overrun may occur. Adding buffer overrun check in liblustreapi.
372 Description: Fix error on 'ls .' at the top of the Lustre mount.
373 Details : Don't revalidate dentry if it is a root dentry.
377 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
378 Details : Modify vm-tunables-rhel4.patch.
382 Description: Kernel config for 2.6.18-vanilla.
383 Details : Modify targets/2.6-vanilla.target.in.
384 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
385 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
386 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
387 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
391 Description: improve handling recoverable errors
392 Details : if request processig with error which can be recoverable on server
393 request should be resend, otherwise page released from cache and
398 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
399 Details : Modify target file & which_patch.
400 A flaw was found in the IA32 system call emulation provided
401 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
402 value could be stored in the %RAX register, which could trigger an
403 out-of-bounds system call table access. An untrusted local user
404 could exploit this flaw to run code in the kernel
405 (ie a root privilege escalation). (CVE-2007-4573).
409 Description: Remove 2.4 kernel checks lustre 1.8
410 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
411 quota_interface.c in quota subsystem.
412 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
414 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
415 lvfs_linux.c upcall_cache.c in lvfs subsystem.
419 Description: O_DIRECT bypasses client statistics.
420 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
421 read_ahead_stats were updated. copied the stats section from
422 osc_send_oap_rpc() into async_internal().
426 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
430 Description: change order of libsysio includes
431 Details : '#include sysio.h' should always come before '#include xtio.h'
435 Description: handle "serious error: objid * already exists" more gracefully
436 Details : If LAST_ID value on disk is smaller than the objects existing in
437 the O/0/d* directories, it indicates disk corruption and causes an
438 LBUG(). If the object is 0-length, then we should use the existing
439 object. This will help to avoid a full fsck in most cases.
441 Severity : enhancement
443 Description: adapt the lustre_config script to support the upgrade case
444 Details : Add "-u" option for lustre_config script to support upgrading 1.4
445 server targets to 1.6 in parallel.
449 Description: To avoid grant space > avaible space when the disk is almost
450 full. Without this patch you might see the error "grant XXXX >
451 available" or some LBUG about grant, when the disk is almost
453 Details : In filter_check_grant, for non_grant cache write, we should
454 check the left space by if (*left > ungranted + bytes), instead
455 of (*left > ungranted), because only we are sure the left space
456 is enough for another "bytes", then the ungrant space should be
457 increase. In client, we should update cl_avail_grant only there
458 is OBD_MD_FLGRANT in the reply.
462 Description: Update RHEL 4 kernel to fix local root privilege escalation.
463 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
464 described in CVE-2007-4573. This problem could allow untrusted
465 local users to gain root access.
469 Description: Remove 2.4 kernel checks lustre 1.8
470 Details : Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
471 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
473 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
474 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
478 Frequency : when using O_DIRECT and quotas
480 Description: Incorrect file ownership on O_DIRECT output files
481 Details : block usage reported by 'lfs quota' does not take into account
482 files that have been written with O_DIRECT.
484 --------------------------------------------------------------------------------
486 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
488 * Support for kernels:
489 2.6.5-7.283 (SLES 9),
490 2.6.9-55.EL (RHEL 4),
491 2.6.16.46-0.14 (SLES 10),
492 2.6.18.8 vanilla (kernel.org)
493 * Client support for unpatched kernels:
494 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
495 2.6.16 - 2.6.22 vanilla (kernel.org)
496 * Due to recently discovered recovery problems, we do not recommend
497 using patchless RHEL 4 clients with this or any earlier release.
498 * Recommended e2fsprogs version: 1.39.cfs8
499 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
500 * Starting with this release, the ldiskfs backing filesystem required
501 by Lustre is now in its own package, lustre-ldiskfs. This package
502 should be installed. It is versioned separately from Lustre and
503 may be released separately in future.
507 Description: block reactivating mgc import until all deactivates complete
508 Details : Fix race when failing back MDT/MGS to itself (testing)
511 Frequency : at statup only
513 Description: mds_lov_synchronize race leads to various problems
514 Details : simultaneous MDT->OST connections at startup can cause the
515 sync to abort, leaving the OSC in a bad state.
517 Severity : enhancement
519 Description: add optional extra BUILD_VERSION info
520 Details : add a new environment variable (namely LUSTRE_VERS) which allows
521 to override the lustre version.
524 Frequency : 2.6.18 servers only
526 Description: ll_kern_mount() doesn't release the module reference
527 Details : The ldiskfs module reference count never drops down to 0
528 because ll_kern_mount() doesn't release the module reference.
533 Description: server LBUG when using old ost_num_threads parameter
534 Details : Accept the old ost_num_threads parameter but warn that it
535 is deprecated, and fix an off-by-one error that caused an LBUG.
540 Description: Transient SCSI error results in persistent IO issue
541 Details : iobuf->dr_error is not reinitialized to 0 between two
545 Frequency : sometimes when underlying device returns I/O errors
547 Description: OSTs not going read-only during write failures
548 Details : OSTs are not remounted read-only when the journal commit threads
549 get I/O errors because fsfilt_ext3 calls journal_start/stop()
550 instead of the ext3 wrappers.
554 Description: poor connect scaling with increasing client count
555 Details : Don't run filter_grant_sanity_check for more than 100 exports
556 to improve scaling for large numbers of clients.
559 Frequency : SLES10 only
561 Description: sanity-quota.sh quotacheck failed: rc = -22
562 Details : Quotas cannot be enabled on SLES10.
565 Frequency : liblustre clients only
567 Description: getdirentries does not give error when run on compute nodes
568 Details : getdirentries does not fail when the size specified as an argument
569 is too small to contain at least one entry
571 Severity : enhancement
573 Description: Add LNET router traceability for debug purposes
574 Details : If a checksum failure occurs with a router as part of the
575 IO path, the NID of the last router that forwarded the bulk data
576 is printed so it can be identified.
581 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
582 Details : Due to a race condition, liblustre clients were occasionally
585 Severity : enhancement
587 Description: lfs setstripe use optional parameters instead of postional
590 Severity : enhancement
592 Description: Nanosecond timestamp support for ldiskfs
593 Details : The on-disk ldiskfs filesystem has added support for nanosecond
594 resolution timestamps. There is not yet support for this at
595 the Lustre filesystem level.
598 Frequency : during server recovery
600 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
601 Details : request with rq_no_resend flag not awake l_wait_event if they get a
605 Frequency : nfs export on patchless client
607 Description: connectathon hang when test nfs export over patchless client
608 Details : Disconnected dentry cannot be found with lookup, so we do not need
609 to unhash it or make it invalid
612 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
613 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
614 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
619 Description: open req refcounting wrong on reconnect
620 Details : If reconnect happened between getting open reply from server and
621 call to mdc_set_replay_data in ll_file_open, we will schedule
622 replay for unreferenced request that we are about to free.
623 Subsequent close will crash in variety of ways.
624 Check that request is still eligible for replay in
625 mdc_set_replay_data().
630 Description: disable writes to filesystem when reading health_check file
631 Details : the default for reading the health_check proc file has changed
632 to NOT do a journal transaction and write to disk, because this
633 can cause reads of the /proc file to hang and block HA state
634 checking on a healthy but otherwise heavily loaded system. It
635 is possible to return to the previous behaviour during configure
636 with --enable-health-write.
638 Severity : enhancement
640 Description: 64-bit inode version
641 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
642 to the inode. This will be required for version-based recovery.
647 Description: MDS fails to start if a duplicate client export is detected
648 Details : in some rare cases it was possible for a client to connect to
649 an MDS multiple times. Upon recovery the MDS would detect this
650 and fail during startup. Handle this more gracefully.
652 Severity : enhancement
654 Description: Add -o localflock option to simulate old noflock
656 Details : This will achieve local-only flock/fcntl locks
662 Description: log_commit_thread vs filter_destroy race leads to crash
663 Details : Take import reference before releasing llog record semaphore
668 Description: Wrong request locking in request set processing
669 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
670 imp_delayed_list, in this place should be used imp_lock.
673 Frequency : when reconnection
675 Description: Grant Leak when osc reconnect to OST
676 Details : When osc reconnect ost, OST(filter) should check whether it
677 should grant more space to client by comparing fed_grant and
678 cl_avail_grant, and return the granted space to client instead
679 of "new granted" space, because client will call osc_init_grant
680 to update the client grant space info.
683 Frequency : when client reconnect to OST
685 Description: Grant Leak when osc do resend and replay bulk write
686 Details : When osc reconnect to OST, OST(filter)should clear grant info of
687 bulk write request, because the grant info while be sync between
688 OSC and OST when reconnect, and we should ignore the grant info
689 these of resend/replay write req.
694 Description: Grant space more than avaiable left space sometimes.
695 Details : When then OST is about to be full, if two bulk writing from
696 different clients came to OST. Accord the avaliable space of the
697 OST, the first req should be permitted, and the second one
698 should be denied by ENOSPC. But if the seconde arrived before
699 the first one is commited. The OST might wrongly permit second
700 writing, which will cause grant space > avaiable space.
703 Frequency : when client is evicted
705 Description: Grant might be wrongly erased when osc is evicted by OST
706 Details : when the import is evicted by server, it will fork another
707 thread ptlrpc_invalidate_import_thread to invalidate the
708 import, where the grant will be set to 0. While the original
709 thread will update the grant it got when connecting. So if
710 the former happened latter, the grant will be wrongly errased
711 because of this race.
716 Description: Checking Stale with correct fid
717 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
718 to check whether it is stale, because sometimes, we want the
719 enqueue happened anyway, and op_data.fid2 will not be initialized.
721 Severity : enhancement
723 Description: update patchless client
724 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
727 Frequency : only with 2.4 kernel
729 Description: random memory corruption
730 Details : size of struct ll_inode_info is to big for union inode.u and this
731 can be cause of random memory corruption.
736 Description: Memory leak in recovery
737 Details : Lov_mds_md was not free in an error handler in mds_create_object.
738 It should also check obd_fail before fsfilt_start, otherwise if
739 fsfilt_start return -EROFS,(failover mds during mds recovery).
740 then the req will return with repmsg->transno = 0 and rc = EROFS.
741 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
742 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
743 be freed no matter whether fsfilt_commit success or not.
746 Frequency : only with huge count clients
748 Description: Prevents from taking the superblock lock in llap_from_page for
750 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
751 for llap_from_page call in ll_removepage prevents from taking the
752 superblock lock for a soon died page.
757 Description: Not check open intent error before release open handle
758 Details : in some rare cases, the open intent error is not checked before
759 release open handle, which may cause
760 ASSERTION(open_req->rq_transno != 0), because it tries to release
761 the failed open handle.
766 Description: Set cat log bitmap only after create log success.
767 Details : in some rare cases, the cat log bitmap is set too early. and it
768 should be set only after create log success.
772 Description: Accessing a block bevice can re-enable I/O when Lustre is
773 tearing down a device.
774 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
779 Description: service threads may hog cpus when there are a lot of requests
781 Details : Insert cond_resched to give other threads a chance to use some of
787 Description: the cat log was not initialized in recovery
788 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
789 unlink log on mds will not be initialized until mds post
790 recovery. And also in mds post recovery, the unlink log will
791 initialization will be done asynchronausly, so there will be race
792 between add unlink log and unlink log initialization.
796 Description: brw_stats were being printed incorrectly
797 Details : brw_stats were being printed as log2 but all of them were not
798 recorded as log2. Also remove some code duplication arising from
799 filter_tally_{read,write}.
803 Frequency : rare, only in recovery.
804 Description: ASSERTION(req->rq_type != LI_POISON) failed
805 Details : imp_lock should be held while iterating over imp_sending_list for
806 prevent destroy request after get timeout in ptlrpc_queue_wait.
810 Description: replay-single.sh test 52 fails
811 Details : A lock's skiplist need to be cleanup when it being unlinked
812 from its resource list.
816 Description: Short directio read returns full requested size rather than
818 Details : Direct I/O operations should return actual amount of bytes
819 transferred rather than requested size.
821 Severity : enhancement
823 Description: metadata RPC reduction (e.g. for rm performance)
824 Details : decrease the amount of synchronous RPC between clients and servers
825 by canceling conflicing lock before the operation on the client side
826 and packing thier handles into the main operation RPC to server.
828 Severity : enhancement
830 Description: Async OSC create to avoid the blocking unnecessarily.
831 Details : If a OST has no remain object, system will block on the creating
832 when need to create a new object on this OST. Now, ways use
833 pre-created objects when available, instead of blocking on an
834 empty osc while others are not empty. If we must block, we block
835 for the shortest possible period of time.
837 --------------------------------------------------------------------------------
839 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
844 Frequency : on some architectures
846 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
847 Details : Uninitialized flags sometimes cause configuration commands to
851 Frequency : patchless clients only
853 Description: missing __iget() symbol export
854 Details : The __iget() symbol export is missing. To avoid the need for
855 this on patchless clients the deathrow inode reaper is turned
856 off, and we depend on the VM to clean up old inodes. This
857 dependency was during via the fix for bug 12181.
859 --------------------------------------------------------------------------------
861 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
863 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
864 INTEROPERATE with older versions automatically. In many cases a
865 special upgrade step is needed. Please read the
866 user documentation before upgrading any part of a live system.
867 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
868 version will not interoperate with 1.6 betas before beta5 (1.5.95).
869 * WARNING: Lustre configuration and startup changes are required with
870 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
876 Severity : enhancement
878 Description: MountConf
879 Details : Lustre configuration is now managed via mkfs and mount
880 commands instead of lmc and lconf. New obd types (MGS, MGC)
881 are added for dynamic configuration management. See
882 https://mail.clusterfs.com/wikis/lustre/MountConf for
885 Severity : enhancement
887 Description: dynamic OST addition
888 Details : OSTs can now be added to a live filesystem
890 Severity : enhancement
892 Description: startup order invariance
893 Details : MDTs and OSTs can be started in any order. Clients only
894 require the MDT to complete startup.
896 Severity : enhancement
898 Description: parallel, asynchronous orphan cleanup
899 Details : orphan cleanup is now performed in separate threads for each
900 OST, allowing parallel non-blocking operation.
902 Severity : enhancement
904 Description: optimized stripe assignment
905 Details : stripe assignments are now made based on ost space available,
906 ost previous usage, and OSS previous usage, in order to try
907 to optimize storage space and networking resources.
909 Severity : enhancement
911 Description: Permanently set tunables
912 Details : All writable /proc/fs/lustre tunables can now be permanently
913 set on a per-server basis, at mkfs time or on a live
916 Severity : enhancement
918 Description: Lustre message v2
919 Details : Add lustre message format v2.
921 Severity : enhancement
923 Description: client OST exclusion list
924 Details : Clients can be started with a list of OSTs that should be
925 declared "inactive" for known non-responsive OSTs.
929 Description: SPEC SFS validation failure on NFS v2 over lustre.
930 Details : Changes the blocksize for regular files to be 2x RPC size,
931 and not depend on stripe size.
933 Severity : enhancement
935 Description: Multiple MD RPCs in flight.
936 Details : Further unserialise some read-only MDS RPCs - learn about intents.
937 To avoid overly-overloading MDS, introduce a limit on number of
938 MDS RPCs in flight for a single client and add /proc controls
939 to adjust this limit.
941 Severity : enhancement
943 Description: client read/write statistics
944 Details : Add client read/write call usage stats for performance
945 analysis of user processes.
946 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
947 file access. extents_stats shows chunk size distribution.
948 extents_stats_per_process show chunk size distribution per
951 Severity : enhancement
953 Description: mds statistics
954 Details : Add detailed mds operations statistics in
955 /proc/fs/lustre/mds/*/stats.
959 Description: Failure of copying files with lustre special EAs.
960 Details : Client side always return success for setxattr call for lustre
961 special xattr (currently only "trusted.lov").
966 Description: Refcount LNET uuids
967 Details : The global LNET uuid list grew linearly with every startup;
968 refcount repeated list entries instead of always adding to
971 Severity : enhancement
973 Description: Dynamic service threads
974 Details : Within a small range, start extra service threads
975 automatically when the request queue builds up.
978 Frequency : mixed-endian client/server environments
980 Description: mixed-endian crashes
981 Details : The new msg_v2 system had some failures in mixed-endian
984 Severity : enhancement
986 Description: Easy OST removal
987 Details : OSTs can be permanently deactivated with e.g. 'lctl
988 conf_param lustre-OST0001.osc.active=0'
990 Severity : enhancement
992 Description: MGS proc entries
993 Details : Added basic proc entries for the MGS showing what filesystems
996 Severity : enhancement
998 Description: provide MGS failover
999 Details : Added config lock reacquisition after MGS server failover.
1001 Severity : enhancement
1003 Description: add Linux 2.4 support
1004 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1008 Description: a large application tries to do I/O to the same resource and dies
1009 in the middle of it.
1010 Details : Check the req->rq_arrival time after the call to
1011 ost_brw_lock_get(), but before we do anything about
1012 processing it & sending the BULK transfer request. This
1013 should help move old stale pending locks off the queue as
1014 quickly as obd_timeout.
1017 Frequency : when an incorrect nid is specified during startup
1019 Description: ptlrpc connect to non-existant node causes kernel crash
1020 Details : LNET can't be re-entered from an event callback, which
1021 happened when we expire a message after the export has been
1022 cleaned up. Instead, hand the zombie cleanup off to another
1025 Severity : enhancement
1027 Description: plain/inodebits lock performance improvement
1028 Details : Grouping plain/inodebits in granted list by their request modes
1029 and bits policy, thus improving the performance of search through
1033 Frequency : only if OST filesystem is corrupted
1035 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1036 Details : for a short time RPCs with bulk IO are in the replay list,
1037 but replay of bulk IOs is unimplemented. If the OST filesystem
1038 is corrupted due to disk cache incoherency and then replay is
1039 started it is possible to trip an assertion. Avoid putting
1040 committed RPCs into the replay list at all to avoid this issue.
1043 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1045 Description: System hang on startup
1046 Details : This bug allowed the liblustre (e.g. catamount) client to
1047 return to the app before handling all startup RPCs. This
1048 could leave the node unresponsive to lustre network traffic
1049 and manifested as a server ptllnd timeout.
1051 Severity : enhancement
1053 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1054 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1055 boolean to control whether to print peer debug info when a
1056 client's RPC times out.
1059 Frequency : only for kernels with patches from Lustre below 1.4.3
1061 Description: Remove old rdonly API
1062 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1065 Frequency : only for devices with external journals
1067 Description: Set external device read-only also
1068 Details : During a commanded failover stop, we set the disk device
1069 read-only while the server shuts down. We now also set any
1070 external journal device read-only at the same time.
1073 Frequency : when upgrading from 1.4 while trying to change parameters
1075 Description: The wrong (new) MDC name was used when setting parameters for
1076 upgraded MDT's. Also allows changing of OSC (and MDC)
1077 parameters if --writeconf is specified at tunefs upgrade time.
1080 Frequency : when setting specific ost indicies
1082 Description: QOS code breaks on skipped indicies
1083 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1084 created with --index need not be sequential.
1086 Severity : enhancement
1088 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1089 Details : The uninit_groups feature works in conjunction with the kernel
1090 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1091 up the pass1 processing of e2fsck. This is a read-only feature
1092 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1093 mount filesystems that have had this feature enabled.
1095 Severity : enhancement
1097 Description: Improve multi-block allocation algorithm to avoid fragmentation
1098 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1099 allocation locality and avoid filesystem fragmentation.
1101 ------------------------------------------------------------------------------
1103 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1105 * Support for kernels:
1106 2.4.21-47.0.1.EL (RHEL 3)
1107 2.6.5-7.283 (SLES 9)
1108 2.6.9-42.0.10.EL (RHEL 4)
1109 2.6.12.6 vanilla (kernel.org)
1110 2.6.16.27-0.9 (SLES 10)
1111 * Recommended e2fsprogs version: 1.39.cfs5
1113 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1117 Frequency : occasional, depends on client load and configuration
1118 Bugzilla : 12181, 12203
1119 Description: data loss for recently-modified files
1121 Details : In some cases it is possible that recently written or created
1122 files may not be written to disk in a timely manner (this should
1123 normally be within 30s unless client IO load is very high).
1124 The problem appears as zero-length files or files that are a
1125 multiple of 1MB in size after a client crash or client eviction
1126 that are missing data at the end of the file.
1128 This problem is more likely to be hit on clients where files are
1129 repeatedly created and unlinked in the same directory, clients
1130 have a large amount of RAM, have many CPUs, the filesystem has
1131 many OSTs, the clients are rebooted frequently, and/or the files
1132 are not accessed by other nodes after being written.
1134 The presence of the problem can be detected by looking at
1135 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1136 smaller than the second (nr_unused) then dirty files will not
1137 be flushed automatically to disk. "sync; sleep 10" should be
1138 run several times on the node before unmounting it to update
1139 Lustre (this is also safe to run on nodes without this problem).
1141 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1142 that can cause this same problem, so customers using that kernel
1143 also need to update the kernel in addition to Lustre. In order
1144 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1146 It is normal that files written just before a client crash (less
1147 than 30s) may not yet have been flushed to disk, even for local
1151 Frequency : frequent on thin XT3 nodes
1153 Description: UUID collision on thin XT3 Linux nodes
1154 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1155 randomly, since we relied on an insufficiently-seeded PRNG.
1160 Description: OSS hangs after "All ost request buffers busy"
1161 Details : A deadlock between quota and journal operations caused OSS
1162 hangs after printing "All ost request buffers busy."
1165 Frequency : always on liblustre builds
1167 Description: Cleanup compiler warnings on liblustre
1170 Frequency : always on liblustre builds on XT3
1172 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1176 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1177 > 32000 subdirectories
1178 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1179 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1180 subdirectory count crosses 32000. This will aid e2fsck to
1181 correctly handle more than 32000 subdirectories.
1184 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1186 Description: System hang on startup
1187 Details : This bug allowed the liblustre (e.g. catamount) client to
1188 return to the app before handling all startup RPCs. This
1189 could leave the node unresponsive to lustre network traffic
1190 and manifested as a server ptllnd timeout.
1192 Severity : enhancement
1194 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1195 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1196 boolean to control whether to print peer debug info when a
1197 client's RPC times out.
1202 Description: make O_SYNC working on 2.6 kernels
1203 Details : 2.6 kernels use different method for mark pages for write,
1204 so need add a code to lustre for O_SYNC work.
1209 Description: Failure to close file and release space on NFS
1210 Details : Put inode details into lock acquired in ll_intent_file_open.
1211 Use mdc_intent_lock in ll_intent_open to properly
1212 detect all kind of errors unhandled by mdc_enqueue.
1217 Description: proc file read during shutdown sometimes raced obd removal,
1219 Details : Add lock to prevent obd access after proc file removal.
1222 Frequency : Only for files larger than 4GB on 32-bit clients.
1224 Description: improperly doing page alignment of locks
1225 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1226 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1231 Description: RPCs being resent when they shouldn't be
1232 Details : Some RPCs that should not be resent are being resent. This
1233 can cause inconsistencies in the RPC state machine. Do not
1234 resend such requests.
1237 Frequency : rare, only with NFS export
1239 Description: Crash on NFS re-export node
1240 Details : under very unusual load conditions an assertion is hit in
1241 ll_intent_file_open()
1244 Frequency : only if OST filesystem is corrupted
1246 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1247 Details : for a short time RPCs with bulk IO are in the replay list,
1248 but replay of bulk IOs is unimplemented. If the OST filesystem
1249 is corrupted due to disk cache incoherency and then replay is
1250 started it is possible to trip an assertion. Avoid putting
1251 committed RPCs into the replay list at all to avoid this issue.
1256 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1257 Details : Large single O_DIRECT read and write calls can fail to allocate
1258 a sufficiently large buffer to process the request. In case of
1259 allocation failure the allocation is retried with a smaller
1260 buffer and broken into smaller requests.
1262 Severity : enhancement
1264 Description: Add -o localflock option to simulate old noflock behaviour.
1265 Details : This will achieve local-only flock/fcntl locks coherentness.
1270 Description: versioning check is incomplete
1271 Details : Checking the version difference of client vs. server, report
1272 error if the gap is too big.
1277 Description: add support PG_writeback bit
1278 Details : add support for PG_writeback bit for Lustre, for more carefull
1279 work with page cache in 2.6 kernel. This also fix some deadlocks
1280 and remove hack for work O_SYNC with 2.6 kernel.
1282 Severity : enhancement
1284 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1285 Details : The uninit_groups feature works in conjunction with the kernel
1286 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1287 up the pass1 processing of e2fsck. This is a read-only feature
1288 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1289 mount filesystems that have had this feature enabled.
1291 Severity : enhancement
1293 Description: Improve multi-block allocation algorithm to avoid fragmentation
1294 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1295 allocation locality and avoid filesystem fragmentation.
1297 ------------------------------------------------------------------------------
1299 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1301 * Support for kernels:
1302 2.6.9-42.0.3.EL (RHEL 4)
1303 2.6.5-7.276 (SLES 9)
1304 2.4.21-47.0.1.EL (RHEL 3)
1305 2.6.12.6 vanilla (kernel.org)
1306 2.6.16.21-0.8 (SLES10)
1307 * Recommended e2fsprogs version: 1.39.cfs2-0
1309 * The backwards-compatible /proc/sys/portals symlink has been removed
1310 in this release. Before upgrading, please ensure that you change
1311 any configuration scripts or /etc/sysctl.conf files that access
1312 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1313 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1314 in advance of the upgrade on any system running Lustre 1.4.6 or
1315 newer, since /proc/sys/lnet was added in that version.
1316 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1320 Frequency : only when quota is used
1322 Description: avoid scanning export list for quota master
1323 Details : Change the algorithms to avoid scanning export list in order
1324 to improve the efficiency.
1327 Frequency : MDS failover only, very rarely
1329 Description: "went back in time" messages on mds failover
1330 Details : The greatest transno may be lost when the current operation
1331 finishes with an error (transno==0) and the client's last_rcvd
1332 record is over-written. Save the greatest transno in the
1333 mds_last_transno for this case.
1336 Frequency : always for specific kernels and striping counts
1338 Description: client may get "Matching packet too big" without ACL support
1339 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1340 when trying to access files in certain configurations. The
1341 clients should in fact be denied when mounting because they do
1342 not understand ACLs.
1345 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1347 Description: many clients connecting with IO in progress causes connect timeouts
1348 Details : Avoid synchronous journal commits to avoid delays caused by many
1349 clients connecting/disconnecting when bulk IO is in progress.
1350 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1351 OST_IO_PORTAL to avoid delays behind potentially many pending
1355 Frequency : occasionally with multiple writers to a single file
1357 Description: shared writes to file may result in wrong size reported by stat()
1358 Details : Allow growing of kms when extent lock is cancelled
1361 Frequency : always with random mmap IO to multi-striped file
1363 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1364 Details : Only if the hole is at the end of OST object so that kms is too
1365 small. Fix is to increase kms accordingly in ll_nopage.
1368 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1370 Description: writes to a missing object would leak memory on the OST
1371 Details : If there is an inconsistency between the MDS and OST filesystems,
1372 such that the MDS references an object that doesn't exist, writes
1373 to that object will leak memory due to incorrect cleanup in the
1374 error handling path, eventually running out of memory on the OST.
1379 Description: Creating too long symlink causes lustre errors
1380 Details : Check symlink and name lengths before sending requests to MDS.
1383 Frequency : only if flock is enabled (not on by default)
1385 Description: posix locks not released on fd closure on 2.6.9+
1386 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1387 kernels, this caused such locks not to be released on fd close and
1388 then assertions on fs unmount about still used locks.
1391 Frequency : MDS failover only, very rarely
1393 Description: clients may get ASSERTION(granted_lock != NULL)
1394 Details : When request was taking a long time, and a client was resending
1395 a getattr by name lock request. The were multiple lock requests
1396 with the same client lock handle and
1397 mds_getattr_name->fixup_handle_for_resent_request found one of the
1398 lock handles but later failed with ASSERTION(granted_lock != NULL).
1403 Description: handle->h_buffer_credits > 0, assertion failure
1404 Details : h_buffer_credits is zero after truncate, causing assertion
1405 failure. This patch extends the transaction or creates a new
1409 Frequency : NFS re-export or patchless client
1410 Bugzilla : 11179, 10796
1411 Description: Crash on NFS re-export node (__d_move)
1412 Details : We do not want to hash the dentry if we don't have a lock.
1413 But if this dentry is later used in d_move, we'd hit uninitialised
1414 list head d_hash, so we just do this to init d_hash field but
1415 leave dentry unhashed.
1418 Frequency : NFS re-export or patchless client
1420 Description: NFS exports has problem with symbolic link
1421 Details : lustre client didn't properly install dentry when re-exported
1422 to NFS or running patchless client.
1425 Frequency : NFS re-export or patchless client
1427 Description: Various nfs/patchless fixes.
1428 Details : fixes reuse disconected alias for lookup process - this fixes
1429 warning "find_exported_dentry: npd != pd",
1430 fix permission error with open files at nfs.
1431 fix apply umask when do revalidate.
1434 Frequency : occasional
1436 Description: Crash on NFS re-export node
1437 Details : calling clear_page() on the wrong pointer triggered oops in
1438 generic_mapping_read().
1441 Frequency : rarely, using O_DIRECT IO
1443 Description: unaligned directio crashes client with LASSERT
1444 Details : check for unaligned buffers before trying any requests.
1447 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1449 Description: stale data returned from RAID cache
1450 Details : If only a small amount of IO is done to the RAID device before
1451 reading it again it is possible to get stale data from the RAID
1452 cache instead of reading it from disk.
1455 Frequency : always for sles10 kernel
1457 Description: sles10 support
1458 Details : ll_follow_link: compile fixes and using of nd_set_link
1459 under newer kernels.
1462 Frequency : depends on arch, kernel and compiler version, always on sles10
1465 Description: recursive or deep enough symlinks cause stack overflow
1466 Details : getting rid of large stack-allocated variable in
1470 Frequency : depends on hardware
1472 Description: lustre write performance loss in the SLES10 kernel
1473 Details : the performance loss is caused by using of write barriers in the
1474 ext3 code. The SLES10 kernel turns barrier support on by
1475 default. The fix is to undo that change for ldiskfs.
1477 ------------------------------------------------------------------------------
1479 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1481 * Support for kernels:
1482 2.6.9-42.0.3EL (RHEL 4)
1483 2.6.5-7.276 (SLES 9)
1484 2.4.21-47.0.1.EL (RHEL 3)
1485 2.6.12.6 vanilla (kernel.org)
1489 Frequency : quota enabled and large files being deleted
1491 Description: releasing more than 4GB of quota at once hangs OST
1492 Details : If a user deletes more than 4GB of files on a single OST it
1493 will cause the OST to spin in an infinite loop. Release
1494 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1499 Description: statfs data retrieved from /proc may be stale or zero
1500 Details : When reading per-device statfs data from /proc, in the
1501 {kbytes,files}_{total,free,avail} files, it may appear
1502 as zero or be out of date.
1505 Frequency : systems with MD RAID1 external journal devices
1507 Description: lconf's call to blkid is confused by RAID1 journal devices
1508 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1509 of returning all block devices that match the journal UUID.
1512 Frequency : always, for aggregate stripe size over 4GB
1514 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1515 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1516 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1519 Frequency : always if "lfs find" used on a local file/directory
1521 Description: "lfs find" segfaults if used on a local file/directory
1522 Details : The case where a directory component was not specified wasn't
1523 handled correctly. Handle this properly.
1526 Frequency : always on ppc64
1528 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1529 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1530 perform endianness conversion.
1533 Frequency : rarely (truncate to non-zero file size after write under load)
1534 Bugzilla : 10730, 10687
1535 Description: Files padded with zeros to next 4K multiple
1536 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1537 it is possible that files that are truncated to a non-zero size
1538 immediately after being written are filled with zero bytes beyond
1539 the truncated size. No file data is lost.
1541 Severity : enhancement
1543 Description: Allow recovery/failover for liblustre clients.
1544 Details : liblustre clients were unaware of failover configurations until
1547 Severity : enhancement
1549 Description: user file locks should fail when not mounting with flock option
1550 Details : Set up an error-returning stub in ll_file_operations.lock field
1551 to prevent incorrect behaviour when client is mounted without
1552 flock option. Also, set up properly f_op->flock field for
1556 Frequency : always on ia64
1558 Description: "lfs df" loops on printing out MDS statfs information
1559 Details : The obd_ioctl_data was not initialized and in some systems
1560 this caused a failure during the ioctl that did not return
1561 an error. Initialize the struct and return an error on failure.
1564 Frequency : SLES 9 only
1566 Description: Error of copying files with lustre special EAs as root
1567 Details : Client side always return success for setxattr call for lustre
1568 special xattr (currently only "trusted.lov").
1571 Frequency : rarely on clusters with both ia64+i386 clients
1573 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1574 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1575 clients are doing concurrent writes to the same file it is
1576 possible that the ia64 clients may LASSERT because the OST
1577 extent locks are not PAGE_SIZE aligned. Ensure that grown
1578 locks are always aligned on the request boundary.
1581 Frequency : specific use, occasional
1583 Description: Overwriting in use executable truncates on-disk binary image
1584 Details : If one node attempts to overwrite an executable in use by
1585 another node, we now correctly return ETXTBSY instead of
1586 truncating the file.
1588 Severity : enhancement
1590 Description: Async OSC create to avoid the blocking unnecessarily.
1591 Details : If a OST has no remain object, system will block on the creating
1592 when need to create a new object on this OST. Now, ways use
1593 pre-created objects when available, instead of blocking on an
1594 empty osc while others are not empty. If we must block, we block
1595 for the shortest possible period of time.
1600 Description: chmod on Lustre root is propagated to other clients
1601 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1602 invalid by the follow_mount time.
1605 Frequency : liblustre clients only
1607 Description: Race in 'instant cancel' lock handling could lead to such locks
1608 never to be granted in case of SMP MDS
1609 Details : Do not destroy not yet granted but cbpending locks in
1613 Frequency : replay/resend of open
1615 Description: non null lock assetion failure in mds_intent_policy
1616 Details : Trying to replay/resend lockless open requests resulted in
1617 mds_open() returning 0 with no lock. Now it sets a flag if
1618 a lock is going to be returned.
1620 Severity : enhancement
1622 Description: Checksum enhancements
1623 Details : New checksum enhancements allow for resending RPCs that failed
1626 Severity : enhancement
1628 Description: Tunables on number of dirty pages in cacche
1629 Details : Allow to set limit on number of dirty pages cached.
1634 Description: client crash on unmount - lock still has references
1635 Details : In some error handling cases it was possible to leak a lock
1636 reference on a client while accessing a file. This was not
1637 harmful to the client during operation, but would cause the
1638 client to crash when the filesystem is unmounted.
1641 Frequency : specific case, rare
1643 Description: ETXTBSY on mds though file not in use by client
1644 Details : ETXTBSY is no longer incorrectly returned when attempting to
1645 chmod or chown a directory that the user previously tried to
1646 execute or a currently-executing binary.
1649 Frequency : extremely rare except on liblustre-based clients
1651 Description: Lustre space not freed when files are deleted
1652 Details : Clean up open-unlinked files after client eviction. Previously
1653 the unlink was skipped and the files remained as orphans.
1658 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1659 Details : In some cases it was possible to send a blocking callback to
1660 a client doing a glimpse, even though that client didn't get
1661 a lock granted. When the glimpse lock is cancelled on the OST
1662 the freed lock is left on the waiting list and corrupted the list.
1665 Frequency : all core dumps
1667 Description: Broke core dumps to lustre
1668 Details : Negative dentry may be unhashed if parent does not have UPDATE
1669 lock, but some callers, e.g. do_coredump, expect dentry to be
1670 hashed after successful create, hash it in ll_create_it.
1672 ------------------------------------------------------------------------------
1674 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1676 * Support for kernels:
1677 2.6.9-42.0.2.EL (RHEL 4)
1678 2.6.5-7.276 (SLES 9)
1679 2.4.21-40.EL (RHEL 3)
1680 2.6.12.6 vanilla (kernel.org)
1684 Frequency : always on RHEL 3
1686 Description: Number of open files grows over time
1687 Details : The number of open files grows over time, whether or not
1688 Lustre is started. This was due to a filp leak introduced
1689 by one of our kernel patches.
1691 ------------------------------------------------------------------------------
1693 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1695 * Support for kernels:
1696 2.6.9-42.EL (RHEL 4)
1697 2.6.5-7.276 (SLES 9)
1698 2.4.21-40.EL (RHEL 3)
1699 2.6.12.6 vanilla (kernel.org)
1704 Bugzilla : 5719, 9635, 9792, 9684
1705 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1706 Details : If a server is under heavy load and cannot reply to new
1707 connection requests before the client resends the (re)connect,
1708 the connection handling code can behave badly if two service
1709 threads are concurrently handing separate (re)connections from
1710 the same client. Add better locking to the connection handling
1711 code, and ensure that only a single connection will be processed
1712 for a given client UUID, even if the lock is dropped.
1714 Severity : enhancement
1716 Description: add TCP zero-copy support to kernel
1717 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1718 sends if the hardware supports scatter-gather and checksumming.
1719 This allows socklnd to do client-write and server-read more
1720 efficiently and reduce CPU utilization from skbuf copying.
1723 Frequency : only if NFS exporting from client
1725 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1726 Details : If setting attributes on a file created under NFS that had
1727 never been opened it would be possible to oops the client
1728 if the file had no objects.
1731 Frequency : always for liblustre
1733 Description: liblustre client does MDS+OSTs setattr RPC for each write
1734 Details : When doing a write from a liblustre client, the client
1735 incorrectly issued an RPC to the MDS and each OST the file was
1736 striped over in order to update the timestamps. When writing
1737 with small chunks and many clients this could overwhelm the MDS
1738 with RPCs. In all cases it would slow down the write because
1739 these RPCs are unnecessary.
1741 Severity : enhancement
1743 Description: allow number of MDS service threads to be changed at module load
1744 Details : It is now possible to change the number of MDS service threads
1745 running. Adding "options mds mds_num_threads={N}" to the MDS's
1746 /etc/modprobe.conf will set the number of threads for the next
1747 time Lustre is restarted (assuming the "mds" module is also
1748 reloaded at that time). The default number of threads will
1749 stay the same, 32 for most systems.
1754 Description: OST crash if filesystem is unformatted or corrupt
1755 Details : If an OST is started on a device that has never been formatted
1756 or if the filesystem is corrupt and cannot even mount then the
1757 error handling cleanup routines would dereference a NULL pointer.
1762 Description: NULL pointer deref in llap_from_page.
1763 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1764 incorrect) mapping if the page was truncated/reclaimed while it was
1765 searched for. Check for this condition and skip such pages when
1766 doing readahead. Introduce extra check to llap_from_page() to
1767 verify page->mapping->host is non-NULL (so page is not anonymous).
1770 Frequency : Sometimes when using sys_sendfile
1772 Description: "page not covered by a lock" warnings from ll_readpage
1773 Details : sendfile called ll_readpage without right page locks present.
1774 Now we introduced ll_file_sendfile that does necessary locking
1775 around call to generic_file_sendfile() much like we do in
1779 Frequency : with certain MDS communication failures at client mount time
1781 Description: NULL pointer deref after failed client mount
1782 Details : a client connection request may delayed by the network layer
1783 and not be sent until after the PTLRPC layer has timed out the
1784 request. If the client fails the mount immediately it will try
1785 to clean up before the network times out the request. Add a
1786 reference from the request import to the obd device and delay
1787 the cleanup until the network drops the request.
1790 Frequency : occasionally during client (re)connect
1792 Description: assertion failure during client (re)connect
1793 Details : processing a client connection request may be delayed by the
1794 client or server longer than the client connect timeout. This
1795 causes the client to resend the connection request. If the
1796 original connection request is replied in this interval, the
1797 client may trip an assertion failure in ptlrpc_connect_interpret()
1798 which thought it would be the only running connect process.
1801 Frequency : only with obd_echo servers and clients that are rebooted
1803 Description: kernel BUG accessing uninitialized data structure
1804 Details : When running an obd_echo server it did not start the ping_evictor
1805 thread, and when a client was evicted an uninitialized data
1806 structure was accessed. Start the ping_evictor in the RPC
1807 service startup instead of the OBD startup.
1809 Severity : enhancement
1810 Bugzilla : 10193 (patchless)
1811 Description: Remove dependency on various unexported kernel interfaces.
1812 Details : No longer need reparent_to_init, exit_mm, exit_files,
1813 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1816 Frequency : rare (only users of deprecated and unsupported LDAP config)
1818 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1819 Details : LDAP apparently contains 'lustreName' attributes instead of
1820 'name'. A simple remapping of the name is sufficient.
1823 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1825 Description: waiting_locks_callback trips kernel BUG if client is evicted
1826 Details : Running with the dump_on_timeout debug flag turned on makes
1827 it possible that the waiting_locks_callback() can try to dump
1828 the Lustre kernel debug logs from an interrupt handler. Defer
1829 this log dumping to the expired_lock_main() thread.
1831 Severity : enhancement
1833 Description: Support NFS exporting on 2.6 kernels.
1834 Details : Implement non-rawops metadata methods for NFS server to use without
1835 changing NFS server code.
1838 Frequency : very rare (synthetic metadata workload only)
1840 Description: two racing renames might cause an MDS thread to deadlock
1841 Details : Running the "racer" program may cause one MDS thread to rename
1842 a file from being the source of a rename to being the target of
1843 a rename at exactly the same time that another thread is doing
1844 so, and the second thread has already enqueued these locks after
1845 doing a lookup of the target and is trying to relock them in
1846 order. Ensure that we don't try to re-lock the same resource.
1849 Frequency : only very large systems with liblustre clients
1851 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1852 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1853 all OSTs in parallel. This allows the eviction of stale liblustre
1854 clients to proceed much faster than if they were done in series,
1855 and also offers similar improvements for other set_info RPCs.
1860 Description: excessive CPU usage during initial read phase on client
1861 Details : During the initial read phase on a client, it would agressively
1862 retry readahead on the file, consuming too much CPU and impacting
1863 performance (since 1.4.5.8). Improve the readahead algorithm
1864 to avoid this, and also improve some other common cases (read
1865 of small files in particular, where "small" is files smaller than
1866 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1871 Description: MDS crash when receiving packet with unknown intent.
1872 Details : Do not LBUG in unknown intent case, just return -EFAULT
1874 Severity : enhancement
1875 Bugzilla : 9293, 9385
1876 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1877 Details : Do not serialize getattr (non-intent version) and statfs.
1880 Frequency : occasional, when OST network is overloaded/intermittent
1882 Description: client evicted by OST after bulk IO timeout
1883 Details : If a client sends a bulk IO request (read or write) the OST
1884 may evict the client if it is unresposive to its data GET/PUT
1885 request. This is incorrect if the network is overloaded (takes
1886 too long to transfer the RPC data) or dropped the OST GET/PUT
1887 request. There is no need to evict the client at all, since
1888 the pinger and/or lock callbacks will handle this, and the
1889 client can restart the bulk request.
1892 Frequency : Always when mmapping file with no objects
1894 Description: client crashes when mmapping file with no objects
1895 Details : Check that we actually have objects in a file before doing any
1896 operations on objects in ll_vm_open, ll_vm_close and
1902 Description: Request leak when working with deleted CWD
1903 Details : Introduce advanced request refcount tracking for requests
1904 referenced from lustre intent.
1906 Severity : Enhancement
1908 Description: Cache open file handles on client.
1909 Details : MDS now will return special lock along with openhandle, if
1910 requested and client is allowed to hold openhandle, even if unused,
1911 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1912 closing files for every read/write openration.
1914 Severity : Enhancement
1916 Description: Cache open negative dentries on client when possible.
1917 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1918 negative dentries on lock revocation.
1923 Description: Remounting a client read-only wasn't possible with a zconf mount
1924 Details : It wasn't possible to remount a client read-only with llmount.
1926 Severity : enhancement
1927 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1928 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1929 MPICH is included to add Lustre-specific ADIO interfaces.
1930 This is based closely on the UFS ADIO layer and only differs
1931 in file creation, in order to allow the OST striping to be set.
1932 This is user-contributed code and not supported by CFS.
1937 Description: extended inode attributes (immutable, append-only) work improperly
1938 when 2.4 and 2.6 kernels are used on client/server or vice versa
1939 Details : Introduce kernel-independent values for these flags.
1941 Severity : enhancement
1944 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1945 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1946 granularity. Now, Fractional MB granularity is be supported,
1947 this is very useful for low memory system.
1949 Severity : enhancement
1951 Description: Getattr by fid
1952 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1953 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1954 have locks and avoids getattr rpc requests for it on every stat.
1957 Frequency : Always, for filesystems larger than 2TB
1959 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1960 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1961 filesystems larger than 2TB to be mounted with mballoc enabled.
1964 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1966 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1967 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1968 filesystems larger than 2TB is mounted then the kernel will
1969 silently corrupt the start of the filesystem. CONFIG_LBD is
1970 enabled for all CFS-supported kernels, but the possibility of
1971 this happening with a modified kernel config exists.
1973 Severity : enhancement
1975 Description: add client O_DIRECT support for 2.6 kernels
1976 Details : It is now possible to do O_DIRECT reads and writes to files
1977 in the Lustre client mountpoint on 2.6 kernel clients.
1979 Severity : enhancement
1981 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1982 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1983 parallel, not waiting for response from every OST before sending
1984 a rpc to the next OST.
1989 Description: setattr vs write race when updating file timestamps
1990 Details : Client processes that update a file timestamp into the past
1991 right after writing to the file (e.g. tar) it is possible that
1992 the updated file modification time can be reset to the current
1993 time due to a race between processing the setattr and write RPC.
1995 Severity : enhancement
1997 Description: Bring 'lfs find' closer in line with regular Linux find.
1998 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1999 -print0 options and obtains all the needed info through the lustre
2002 Severity : enhancement
2004 Description: support up to 1024 configured devices on one node
2005 Details : change obd_dev array from statically allocated to dynamically
2006 allocated structs as they are first used to reduce memory usage
2011 Description: Flush dirty partially truncated pages during truncate
2012 Details : Immediatelly flush partially truncated pages in filter_setattr,
2013 this way we completely avoid having any pages in page cache on OST
2014 and can retire ugly workarounds during writes to flush such pages.
2019 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2020 Details : protect inode from truncation within vfs_unlink() context
2021 just take a reference before calling vfs_unlink() and release it
2022 when parent's i_sem is free.
2027 Description: last_id value checked outside lock on OST caused LASSERT failure
2028 Details : If there were multiple MDS->OST object precreate requests in
2029 flight, it was possible that the OST's last object id was checked
2030 outside a lock and incorrectly tripped an assertion. Move checks
2031 inside locks, and discard old precreate requests.
2034 Frequency : always, if extents are used on OSTs
2036 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2037 Details : OSTs using the extents format would not zero the high 16 bits of
2038 the index physical block number. This is not a problem for any
2039 OST filesystems smaller than 16TB, and no kernels support ext3
2040 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2041 new/modified files) and can be fixed for existing filesystems
2042 with e2fsprogs-1.39-cfs1.
2047 Description: import connection selection may be incorrect if timer wraps
2048 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2049 import connections to be ignored if the 32-bit jiffies counter
2050 wraps. Use a 64-bit jiffies counter.
2053 Frequency : very large clusters immediately after boot
2055 Description: LNET request buffers exhausted under heavy short-term load
2056 Details : If a large number of client requests are generated on a service
2057 that has previously never seen so many requests it is possible
2058 that the request buffer growth cannot keep up with the spike in
2059 demand. Instead of dropping incoming requests, they are held in
2060 the LND until the RPC service can accept more requests.
2063 Frequency : Sometimes during replay
2065 Description: Assertion failure in ll_local_open after replay.
2066 Details : If replay happened on an open request reply before we were able
2067 to set replay handler, reply will become not swabbed tripping the
2068 assertion in ll_local_open. Now we set the handler right after
2069 recognising of open request
2072 Frequency : very rare
2074 Description: kernel reports "badness in vsnprintf"
2075 Details : Reading from the "recovery_status" /proc file in small chunks
2076 may cause a negative length in lprocfs_obd_rd_recovery_status()
2077 call to vsnprintf() (which is otherwise harmless). Exit early
2078 if there is no more space in the output buffer.
2080 Severity : enhancement
2082 Description: clear OBD RPC statistics by writing to them
2083 Details : It is now possible to clear the OBD RPC statistics by writing
2084 to the "stats" file.
2089 Description: Client mtime is not the same on different clients after utimes
2090 Details : In some cases, the client was using the utimes() syscall on
2091 a file cached on another node. The clients now validate the
2092 ctime from the MDS + OSTs to determine which one is right.
2097 Description: Inability to activate failout mode
2098 Details : lconf script incorrectly assumed that in pythong string's numeric
2099 value is used in comparisons.
2102 Frequency : always with multiple stripes per file
2104 Description: Inefficient object allocation for mutli-stripe files
2105 Details : When selecting which OSTs to stripe files over, for files with
2106 a stripe count that divides evenly into the number of OSTs,
2107 the MDS is always picking the same starting OST for each file.
2108 Return the OST selection heuristic to the original design.
2113 Description: mount failures may take full timeout to return an error
2114 Details : Under some heavy load conditions it is possible that a
2115 failed mount can wait for the full obd_timeout interval,
2116 possibly several minutes, before reporting an error.
2117 Instead return an error as soon as the status is known.
2119 Frequency : quota enabled and large files being deleted
2121 Description: releasing more than 4GB of quota at once hangs OST
2122 Details : If a user deletes more than 4GB of files on a single OST it
2123 will cause the OST to spin in an infinite loop. Release
2124 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2129 Description: statfs data retrieved from /proc may be stale or zero
2130 Details : When reading per-device statfs data from /proc, in the
2131 {kbytes,files}_{total,free,avail} files, it may appear
2132 as zero or be out of date.
2135 Frequency : systems with MD RAID1 external journal devices
2137 Description: lconf's call to blkid is confused by RAID1 journal devices
2138 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2139 of returning all block devices that match the journal UUID.
2142 Frequency : always, for aggregate stripe size over 4GB
2144 Description: assertion fails when trying to use 4GB stripe size
2145 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2146 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2149 Frequency : always on ppc64
2151 Description: the first write on an ext3 filesystem with mballoc got stuck
2152 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2153 perform endianness conversion.
2155 ------------------------------------------------------------------------------
2157 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2159 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2160 INTEROPERATE with older versions automatically. Please read the
2161 user documentation before upgrading any part of a live system.
2162 * WARNING: Lustre networking configuration changes are required with
2163 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2166 * Support for kernels:
2167 2.6.9-22.0.2.EL (RHEL 4)
2168 2.6.5-7.244 (SLES 9)
2169 2.6.12.6 vanilla (kernel.org)
2172 Severity : enhancement
2173 Bugzilla : 7981/8208
2174 Description: Introduced Lustre Networking (LNET)
2175 Details : LNET is new networking infrastructure for Lustre, it includes
2176 a reorganized network configuration mode (see the user
2177 documentation for full details) as well as support for routing
2178 between different network fabrics. Lustre Networking Devices
2179 (LNDs) for the supported network fabrics have also been
2180 created for this new infrastructure.
2182 Severity : enhancement
2183 Description: Introduced Access control lists
2184 Details : clients can set ACLs on files and directories in order to have
2185 more fine-grained permissions than the standard Unix UGO+RWX.
2186 The MDS must be started with the "-o acl" mount option.
2188 Severity : enhancement
2189 Description: Introduced filesystem quotas
2190 Details : Administrators may now establish per-user quotas on the
2193 Severity : enhancement
2195 Description: Configuration change for the XT3
2196 The PTLLND is now used to run Lustre over Portals on the XT3
2197 The configure option(s) --with-cray-portals are no longer used.
2198 Rather --with-portals=<path-to-portals-includes> is used to
2199 enable building on the XT3. In addition to enable XT3 specific
2200 features the option --enable-cray-xt3 must be used.
2205 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2206 Details : A race between allocating a new llcd and re-getting the llcd_lock
2207 allowed another thread to grab newly-allocated llcd.
2209 Severity : enhancement
2211 Description: 2.6 OST async journal commit and locking fix to improve performance
2212 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2213 2.6 kernels are now async as they already were in 2.4 kernels so
2214 that they can commit concurrently with the network bulk transfer.
2215 For block-allocated files the filter allocation semaphore is held
2216 to avoid filesystem fragmentation during allocation. BKL lock
2217 removed for 2.6 xattr operations where it is no longer needed.
2222 Description: lconf incorrectly determined whether two IP networks could talk
2223 Details : In some more complicated routing and multiple-network
2224 configurations, lconf will avoid trying to make a network
2225 connection to a disjoint part of the IP space. It was doing the
2226 math incorrectly for one set of cases.
2231 Description: Fix for potential infinite loop processing records in an llog.
2232 Details : If an llog record is corrupted/zeroed, it is possible to loop
2233 forever in llog_process(). Validate the llog record length
2234 and skip the remainder of the block on error.
2237 Frequency : occasional (liblustre only)
2239 Description: liblustre could not open files whose last component is a symlink
2240 Details : sysio_path_walk() would incorrectly pass the open intent to
2241 intermediate path components.
2244 Frequency : rare (liblustre only with non-standard tuning)
2245 Bugzilla : 7201 (7350)
2246 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2247 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2248 after it has been released, only do this for referenced locks
2250 Severity : enhancement
2252 Description: specifying an (invalid) directory default stripe_size of -1
2253 would reset the directory default striping
2254 Details : stripe_size -1 was used internally to signal directory stripe
2255 removal, now use "all default" to signal dir stripe removal
2256 as a directory striping of "all default" is not useful
2259 Frequency : common for large clusters running liblustre clients
2261 Description: doing an ls when liblustre clients are running is slow
2262 Details : sending a glimpse AST to a liblustre client waits for every AST
2263 to time out, as liblustre clients will not respond. Since they
2264 cannot cache data we refresh the OST lock LVB from disk instead.
2266 Severity : enhancement
2268 Description: doing an ls at the same time as file IO can be slow
2269 Details : enqueue and other "small" requests can be blocked behind many
2270 large IO requests. Create a new OST IO portal for non-IO
2271 requests so they can be processed faster.
2274 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2276 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2277 created file with mode 000
2278 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2279 allow an ftruncate() as a non-root user to a file with mode 000.
2280 Lustre now respects this flag to disable mode checks when
2281 truncating a file owned by the user
2284 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2287 Description: Revoking locks from clients that went dead or catatonic might take
2289 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2290 cancellation of such locks instant on servers without waiting for
2291 any reply from clients. Clients drops these locks when cancel
2292 notification from server is received without replying.
2295 Frequency : liblustre-only, when liblustre client dies or becomes busy
2297 Description: Doing ls on Linux clients can take a long time with active
2299 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2300 granting such locks to it in the first place if possible. Locks
2301 are taken by proxy on the OST during the read or write and
2302 dropped immediately afterward. Add connect flags handling, do
2303 not grant locks to liblustre clients for glimpse ASTs.
2305 Severity : enhancement
2307 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2308 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2309 sequential reads and behaves badly with applications doing random
2310 reads. Improve it by reading ahead at least read region, and
2311 avoiding excessive large RPC for small reads.
2313 Severity : enhancement
2315 Description: Creating more than 1000 files for a single job may cause a load
2316 imbalance on the OSTs if there are also a large number of OSTs.
2317 Details : qos_prep_create() uses an OST index reseed value that is an
2318 even multiple of the number of available OSTs so that if the
2319 reseed happens in the middle of the object allocation it will
2320 still utilize the OSTs as uniformly as possible.
2325 Description: OST or MDS may oops in ping_evictor_main()
2326 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2327 but doesn't restart at beginning of obd_exports_timed list
2330 Severity : enhancement
2332 Description: improve by-nid export eviction on the MDS and OST
2333 Details : allow multiple exports with the same NID to be evicted at one
2334 time without re-searching the exports list.
2337 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2339 Description: MDS may oops in groups_free()
2340 Details : in rare race conditions a newly allocated group_info struct is
2341 freed again, and this can be NULL. The 2.4 compatibility code
2342 for groups_free() checked for a NULL pointer, but 2.6 did not.
2345 Frequency : common for liblustre clients doing little filesystem IO
2346 Bugzilla : 9352, 7313
2347 Description: server may evict liblustre clients accessing contended locks
2348 Details : if a client is granted a lock or receives a completion AST
2349 with a blocking AST already set it would not reply to the AST
2350 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2353 Frequency : lfs setstripe, only systems with more than 160 OSTs
2355 Description: unable to set striping with a starting offset beyond OST 160
2356 Details : llapi_create_file() incorrectly limited the starting stripe
2357 index to the maximum single-file stripe count.
2360 Frequency : LDAP users only
2362 Description: lconf did not handle in-kernel recovery with LDAP properly
2363 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2365 Severity : enhancement
2367 Description: bind OST threads to NUMA nodes to improve performance
2368 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2369 node and do their allocations there to localize memory access
2371 Severity : enhancement
2373 Description: llmount can determine client NID directly from Myrinet (GM)
2374 Details : the client NID code from gmnalnid was moved directly into
2375 llmount, removing the need to use this or specifying the
2376 client NID explicitly when mounting GM clients with zeroconf
2379 Frequency : if client is started with down MDS
2381 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2382 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2383 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2384 ptlrpc_import_delayed_req() because we didn't check if the
2385 request was interrupted, and we also didn't break out of the
2386 event loop if there was no timeout
2391 Description: data loss during non-page-aligned writes to a single file from
2392 both multiple nodes and multiple threads on one node at same time
2393 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2394 inconsistency led to false short-reads, that were cached and later
2395 used by ->prepare_write() to fill in partially written page,
2396 leading to data loss.
2399 Frequency : always, if lconf --abort_recovery used
2401 Description: lconf --abort_recovery fails with 'Operation not supported'
2402 Details : lconf was attempting to abort recovery on the MDT device and not
2405 Severity : enhancement
2407 Description: remove cleanup logs
2408 Details : replace lconf-generated cleanup logs with lustre internal
2409 cleanup routines. Eliminates the need for client-cleanup and
2412 Severity : enhancement
2414 Description: add support for EAs (user and system) on lustre filesystems
2415 Details : it is now possible to store extended attributes in the Lustre
2416 client filesystem, and with the user_xattr mount option it
2417 is possible to allow users to store EAs on their files also
2419 Severity : enhancement
2421 Description: Add possibility (config option) to show minimal available OST free
2423 Details : When compiled with --enable-mindf configure option, statfs(2)
2424 (and so, df) will return least minimal free space available from
2425 all OSTs as amount of free space on FS, instead of summary of
2426 free spaces of all OSTs.
2428 Severity : enhancement
2430 Description: do not expand extent locks acquired on OST-side
2431 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2432 by server: they are not cached anyway.
2435 Frequency : when mmap is used/binaries executed from Lustre
2437 Description: Unmmap pages before throwing them away from read cache.
2438 Details : llap_shrink cache now attempts to unmap pages before discarding
2439 them (if unmapping failed - do not discard). SLES9 kernel has
2440 extra checks that trigger if this unmapping is not done first.
2445 Description: lconf didn't resolve symlinks before checking to see whether a
2446 given mountpoint was already in use
2449 Frequency : when migrating failover services
2450 Bugzilla : 6395, 9514
2451 Description: When migrating a subset of services from a node (e.g. failback
2452 from a failover service node) the remaining services would
2453 time out and evict clients.
2454 Details : lconf --force (implied by --failover) sets the global obd_timeout
2455 to 5 seconds in order to quickly disconnect, but this caused
2456 other RPCs to time out too quickly. Do not change the global
2457 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2459 Severity : enhancement
2460 Frequency : if MDS is started with down OST
2461 Bugzilla : 9439,5706
2462 Description: Allow startup/shutdown of an MDS without depending on the
2463 availability of the OSTs.
2464 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2465 Add appropriate locking and lov-osc refcounts for safe
2466 cleaning. Add osc abort_inflight calls in case the
2467 synchronize never started.
2470 Frequency : occasional (Cray XT3 only)
2472 Description: root not authorized to access files in CRAY_PORTALS environment
2473 Details : The client process capabilities were not honoured on the MDS in
2474 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2475 been accessed by an authorized user then root was able to access
2476 the file on the local client also. The root user capabilities
2477 are now allowed on the MDS, as this environment has secure UID.
2480 Frequency : occasional
2482 Description: ldiskfs "too long searching" message happens too often
2483 Details : A debugging message (otherwise harmless) prints too often on
2484 the OST console. This has been reduced to only happen when
2485 there are fragmentation problems on the filesystem.
2490 Description: Division by zero in statfs when all OSCs are inactive
2491 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2492 lov_statfs divides by value returned by lov_get_stripecnt().
2496 Bugzilla : 9489, 3273
2497 Description: First write from each client to each OST was only 4kB in size,
2498 to initialize client writeback cache, which caused sub-optimal
2499 RPCs and poor layout on disk for the first writen file.
2500 Details : Clients now request an initial cache grant at (re)connect time
2501 and so that they can start streaming writes to the cache right
2502 away and always do full-sized RPCs if there is enough data.
2503 If the OST is rebooted the client also re-establishes its grant
2504 so that client cached writes will be honoured under the grant.
2509 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2510 Details : Now I/O RPCs go to different portal number and (presumably) fast
2511 lock requests (and glimses) and other RPCs get their own service
2512 threads pool that should be able to service those RPCs
2515 Severity : enhancement
2517 Description: Ability to exchange lustre version between client and servers and
2518 issue warnings at client side if client is too old. Also for
2519 liblustre clients there is ability to refuse connection of too old
2521 Details : New 'version' field is added to connect data structure that is
2522 filled with version info. That info is later checked by server and
2526 Frequency : rare, liblustre only.
2527 Bugzilla : 9296, 9581
2528 Description: Two simultaneous writes from liblustre at offset within same page
2529 might proceed at the same time overwriting eachother with stale
2531 Details : I/O lock withing llu_file_prwv was released too early, before data
2532 actually was hitting the wire. Extended lock-holding time until
2533 server acknowledges receiving data.
2536 Frequency : extremely rare. Never observed in practice.
2538 Description: avoid generating lustre_handle cookie of 0.
2539 Details : class_handle_hash() generates handle cookies by incrementing
2540 global counter, and can hit 0 occasionaly (this is unlikely, but
2541 not impossible, because initial value of cookie counter is
2542 selected randonly). Value of 0 is used as a sentinel meaning
2543 "unassigned handle" --- avoid it. Also coalesce two critical
2544 sections in this function into one.
2546 Severity : enhancement
2548 Description: allow liblustre clients to delegate truncate locking to OST
2549 Details : To avoid overhead of locking, liblustre client instructs OST to
2550 take extent lock in ost_punch() on client's behalf. New connection
2551 flag is added to handle backward compatibility.
2553 Severity : enhancement
2554 Bugzilla : 4928, 7341, 9758
2555 Description: allow number of OST service threads to be specified
2556 Details : a module parameter allows the number of OST service threads
2557 to be specified via "options ost ost_num_threads={N}" in the
2558 OSS's /etc/modules.conf or /etc/modprobe.conf.
2562 Bugzilla : 6146, 9635, 9895
2563 Description: servers crash with bad pointer in target_handle_connect()
2564 Details : In rare cases when a client is reconnecting it was possible that
2565 the connection request was the last reference for that export.
2566 We would temporarily drop the export reference and get a new
2567 one, but this may have been the last reference and the export
2568 was just destroyed. Get new reference before dropping old one.
2570 Severity : enhancement
2571 Frequency : if client is started with failover MDS
2573 Description: Allow multiple MDS hostnames in the mount command
2574 Details : Try to read the configuration from all specified MDS
2575 hostnames during a client mount in case the "primary"
2578 Severity : enhancement
2580 Description: Stop sending data to evicted clients as soon as possible.
2581 Details : Check if the client we are about to send or are sending data to
2582 was evicted already. (Check is done every second of waiting,
2583 for which l_wait_event interface was extended to allow checking
2584 of exit condition at specified intervals).
2587 Frequency : rare, normally only when NFS exporting is done from client
2589 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2591 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2594 Frequency : always, if srand() is called before liblustre initialization
2596 Description: Liblustre uses system PRNG disturbing its usage by user application
2597 Details : Introduce internal to lustre fast and high-quality PRNG for
2598 lustre usage and make liblustre and some other places in generic
2599 lustre code to use it.
2601 Severity : enhancement
2602 Bugzilla : 9477, 9557, 9870
2603 Description: Verify that the MDS configuration logs are updated when xml is
2604 Details : Check if the .xml configuration logs are newer than the config
2605 logs stored on the MDS and report an error if this is the case.
2606 Request --write-conf, or allow starting with --old_conf.
2608 Severity : enhancement
2610 Description: Handle symlinks in the path when checking if Lustre is mounted.
2611 Details : Resolve intermediate symlinks when checking if a client has
2612 mounted a filesystem to avoid duplicate client mounts.
2617 Description: lconf can hit an error exception but still return success.
2618 Details : The lconf command catches the Command error exception at the top
2619 level script context and will exit with the associated exit
2620 status, but doesn't ensure that this exit status is non-zero.
2625 Description: failure of ptlrpc thread startup can cause oops
2626 Details : Starting a ptlrpc service thread can fail if there are a large
2627 number of threads or the server memory is very fragmented.
2628 Handle this without oopsing.
2631 Frequency : always, only if liblustre and non-default acceptor port was used
2633 Description: liblustre cannot connect to servers with non-default acceptor port
2634 Details : tcpnal_set_default_params() was not called and was therefore
2635 ignoring the environment varaible TCPNAL_PORT, as well as other
2636 TCPNAL_ environment variables
2641 Description: two objects could be created on the same OST for a single file
2642 Details : If an OST is down, in some cases it was possible to create two
2643 objects on a single OST for a single file. No problems other
2644 than potential performance impact and spurious error messages.
2648 Bugzilla : 5681, 9562
2649 Description: Client may oops in ll_unhash_aliases
2650 Details : Client dcache may become inconsistent in race condition.
2651 In some cases "getcwd" can fail if the current directory is
2657 Description: Inode refcounting problems in NFS export code
2658 Details : link_raw functions used to call d_instantiate without obtaining
2659 extra inode reference first.
2663 Bugzilla : 9942, 9903
2664 Description: Referencing freed requests leading to crash, memleaks with NFS.
2665 Details : We used to require that call to ll_revalidate_it was always
2666 followed by ll_lookup_it. Also with revalidate_special() it is
2667 possible to call ll_revalidate_it() twice for the same dentry
2668 even if first occurence returned success. This fix changes semantic
2669 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2670 reference on a request referred from the intent.
2671 ll_intent_release() then releases such a request.
2674 Frequency : rare, normally benchmark loads only
2676 Description: unlinked inodes were kept in memory on the client
2677 Details : If a client is repeatedly creating and unlinking files it
2678 can accumulate a lot of stale inodes in the inode slab cache.
2679 If there is no other client load running this can cause the
2680 client node to run out of memory. Instead flush old inodes
2681 from client cache that have the same inode number as a new inode.
2684 Frequency : SLES9 2.6.5 kernel and long filenames only
2685 Bugzilla : 9969, 10379
2686 Description: utime reports stale NFS file handle
2687 Details : SLES9 uses out-of-dentry names in some cases, which confused
2688 the lustre dentry revalidation. Change it to always use the
2692 Frequency : rare, unless heavy write-truncate concurrency is continuous
2693 Bugzilla : 4180, 6984, 7171, 9963, 9331
2694 Description: OST becomes very slow and/or deadlocked during object unlink
2695 Details : filter_destroy() was holding onto the parent directory lock
2696 while truncating+unlinking objects. For very large objects this
2697 may block other threads for a long time and slow overall OST
2698 responsiveness. It may also be possible to get a lock ordering
2699 deadlock in this case, or run out of journal credits because of
2700 the combined truncate+unlink. Solution is to do object truncate
2701 first in one transaction without parent lock, and then do the
2702 final unlink in a new transaction with the parent lock. This
2703 reduces the lock hold time dramatically.
2706 Frequency : rare, 2.4 kernels only
2708 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2709 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2710 before unmounting their filesystem in order to avoid deadlock.
2711 The kernel_locked() function in 2.4 kernels only checks whether
2712 the kernel lock is held, not whether it is this process that is
2713 holding it as 2.6 kernels do.
2718 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2719 Details : The client ptlrpc code may be trying to reconnect to a down
2720 server before a previous connection attempt has timed out.
2721 Increase the reconnect interval to be longer than the connection
2722 timeout interval to avoid sending duplicate connections to
2726 Frequency : echo_client brw_test command
2728 Description: fix echo_client to work with OST preallocated code
2729 Details : OST preallocation code (5137) didn't take echo_client IO path
2730 into account: echo_client calls filter methods outside of any
2731 OST thread and, hence, there is no per-thread preallocated
2732 pages and buffers to use. Solution: hijack pga pages for IO. As
2733 a byproduct, this avoids unnecessary data copying.
2737 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2738 Description: Client can oops in mdc_commit_close() after open replay
2739 Details : It was possible for the MDS to return an open request with no
2740 transaction number in mds_finish_transno() if the client was
2741 evicted, but without actually returning an error. Clients
2742 would later try to replay that open and may trip an assertion
2743 Simplify the client close codepath, and always return an error
2744 from the MDS in case the open is not successful.
2747 Frequency : rare, 2.6 OSTs only
2749 Description: OST may deadlock under high load on fragmented files
2750 Details : If there was a heavy load and highly-fragmented OST filesystems
2751 it was possible to have all the OST threads deadlock waiting on
2752 allocation of biovecs, because the biovecs were not released
2753 until the entire RPC IO was completed. Instead, release biovecs
2754 as soon as they are complete to ensure forward IO progress.
2756 Severity : enhancement
2758 Description: Support for specifying external journal device at mount
2759 Details : If an OST or MDS device is formatted with an external journal
2760 device, this device major/minor is stored in the ext3 superblock
2761 and may not be valid for failover. Allow detecting and
2762 specifying the external journal at mount time.
2767 Description: Mounting an MDS with pending unlinked files may cause oops
2768 Details : target_finish_recovery() calls mds_postrecov() which returned
2769 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2770 considers this an error and immediately begins cleaning up the
2771 lov, just after starting the mds_lov process
2773 Severity : enhancement
2775 Description: Implement 'lfs df' to report actual free space on per-OST basis
2776 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2777 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2778 usage of objects; '-h' to report in human readable format.
2780 ------------------------------------------------------------------------------
2782 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2789 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2790 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2791 at startup during orphan unlinking. Instead, skip buddy update
2792 before setup, as it will be regenerated after recovery is complete.
2797 Description: If an OST is inactive, its locks might reference stale inodes.
2798 Details : lov_change_cbdata() must iterate over all namespaces, even if
2799 they are inactive to clear inode references from the lock.
2801 Severity : enhancement
2802 Frequency : occasional, if non-standard max_dirty_mb used
2804 Description: Client will block write RPCs if not enough grant
2805 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2806 then the client will block writes while waiting for another RPC
2807 to complete instead of consuming its dirty limit. With change
2808 we get improved performance when max_dirty_mb is small.
2810 Severity : enhancement
2811 Bugzilla : 3389, 6253
2812 Description: Add support for supplementary groups on the MDS.
2813 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2814 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2815 lookups for user supplementary groups into a cache.
2819 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2820 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2821 for liblustre O_EXCL is enforced by MDS.
2825 Description: readdir never returns NULL in liblustre.
2826 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2827 dentry in struct dirent.
2831 Frequency : liblustre only, depends on application IO pattern
2832 Description: liblustre clients evicted if not contacting servers
2833 Details : Don't put liblustre clients into the ping_evictor list, so
2834 they will not be evicted by the pinger ever.
2836 Severity : enhancement
2838 Description: Add ability to evict clients by NID from MDS.
2839 Details : By echoing "nid:$NID" string into
2840 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2841 $NID would be instantly evicted from this MDS and from all active
2842 OSTs connected to it.
2846 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2847 Details : lookup_it_finish() used to query file size from OSTs that was not
2852 Description: service threads change working directory to that of init
2853 Details : Starting lustre service threads may pin the working directory
2854 of the parent thread, making that filesystem busy. Threads
2855 now change to the working directory of init to avoid this.
2859 Frequency : during shutdown only
2860 Description: shutdown with a failed MDS or OST can cause unmount to hang
2861 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2866 Frequency : 2.6 only
2867 Description: chmod/chown may include an extra supplementary group
2868 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2869 supplementary group and if none is specified this is used.
2872 Bugzilla : 5479 (6816)
2874 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2875 Details : If lookup is in progress on a file that is unlinked we might try
2876 to revalidate the inode and fail in revalidate after lookup is
2877 complete and ll_file_open() enqueues the open again but
2878 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2881 Frequency : always, if lconf --abort_recovery used
2883 Description: lconf --abort_recovery fails with 'Operation not supported'
2884 Details : lconf was attempting to abort recovery on the MDT device and not
2887 ------------------------------------------------------------------------------
2889 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2894 Frequency : rare (only unsupported configurations with a node running as an
2896 Bugzilla : 6514, 5137
2897 Description: Mounting a Lustre file system on a node running as an OST could
2899 Details : OSTs now preallocates memory needed to write out data at
2900 startup, instead of when needed, to avoid having to
2901 allocate memory in possibly low memory situations.
2902 Specifically, if the file system is mounted on on OST,
2903 memory pressure could force it to try to write out data,
2904 which it needed to allocate memory to do. Due to the low
2905 memory, it would be unable to do so and the node would
2906 become unresponsive.
2908 Severity : enhancement
2910 Description: Addition of lconf --service command line option
2911 Details : lconf now accepts a '--service <arg>' option, which is
2912 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2914 Severity : enhancement
2916 Description: Failover mode is now the default for OSTs.
2917 Details : By default, OSTs will now run in failover mode. To return to
2918 the old behaviour, add '--failout' to the lmc line for OSTs.
2920 Severity : enhancement
2922 Description: Health checks are now provided for MDS and OSTs
2923 Details : Additional detailed health check information on MSD and OSTs
2924 is now provided through the procfs health_check value.
2927 Frequency : occasional, depends on IO load
2929 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2930 after numerous create/delete cycles
2931 Details : The ext3 inode allocation policy would not allocate new inodes
2932 very well on the OSTs because there are no new directories
2933 being created. Instead we look for groups with free space if
2934 the parent directories are nearly full.
2939 Description: Network or server problems during mount may cause partially
2940 mounted clients instead of returning an error.
2941 Details : The config llog parsing code may overwrite the error return
2942 code during mount error handling, returning success instead
2948 Description: MDS can fail to allocate large reply buffers
2949 Details : After long uptimes the MDS can fail to allocate large reply
2950 buffers (e.g. zconf client mount config records) due to memory
2951 fragmentation or consumption by the buffer cache. Preallocate
2952 some large reply buffers so that these replies can be sent even
2953 under memory pressure.
2957 Frequency : rare (liblustre)
2958 Description: fsx running with liblustre complained that using truncate() to
2959 extend the file doesn't work. This patch corrects that issue.
2960 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2961 ATTR_SIZE and lsm use in llu_setattr_raw.
2965 Frequency : rare, only 2.6 kernels
2966 Description: Unusual file access patterns on the MDS may result in inode
2967 data being lost in very rare circumstances.
2968 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2969 code in the 2.6 kernel caused the inode and/or EA data not to
2970 be read from disk, causing single-file corruption.
2974 Frequency : rare, only 2.6 filesystems using extents
2975 Description: Heavy concurrent write and delete load may cause data corruption.
2976 Details : It was possible under high-load situations to have an extent
2977 metadata block in the block device cache from a just-unlinked
2978 file overwrite a newly-allocated data block. We now unmap any
2979 metadata buffers that alias just-allocated data blocks.
2983 Frequency : filesystems with default stripe_count larger than 77
2984 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2985 Details : lconf specifies an inode size of 4096 bytes when the default
2986 stripe_count is larger than 77. This conflicts with the default
2987 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2988 this case to avoid pinning too much memory for large EAs.
2990 ------------------------------------------------------------------------------
2992 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2997 Frequency : rare (extremely heavy IO load with hundreds of clients)
2999 Description: Client is evicted, gets IO error writing to file
3000 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3001 caused clients to be evicted for AST timeouts. The fixes in
3002 bug 5192 mean we no longer need to have such short AST timeouts
3003 so ldlm_timeout has been increased.
3006 Frequency : occasional during --force or --failover shutdown under load
3007 Bugzilla : 5949, 4834
3008 Description: Server oops/LBUG if stopped with --force or --failover under load
3009 Details : a collection of import/export refcount and cleanup ordering
3010 issues fixed for safer force cleanup
3013 Frequency : only filesystems larger than 120 OSTs
3014 Bugzilla : 5990, 6223
3015 Description: lfs getstripe would oops on a very large filesystem
3016 Details : lov_getconfig used kfree on vmalloc'd memory
3019 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3020 Bugzilla : 6242, 6243
3021 Description: reading from files that had been truncated to a non-zero size
3022 but never opened returned no data
3023 Details : ll_file_read() reads zeros from no-object files to EOF
3028 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3029 Details : There's little harm in aborting MDS/OSS recovery and letting it
3030 try again, so I removed the LASSERT and return an error instead.
3032 Severity : enhancement
3034 Description: New debugging infrastructure for tracking down data corruption
3035 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3036 (b) cover more of the client-side code path, and (c) try to narrow
3037 down where problems occurred
3041 Bugzilla : 3819, 4364, 4397, 6313
3042 Description: Racing close and eviction MDS could cause assertion in mds_close
3043 Details : It was possible to get multiple mfd references during close and
3044 client eviction, leading to one thread referencing a freed mfd.
3046 Severity: : enhancement
3047 Bugzilla : 3262, 6359
3048 Description: Attempts to reconnect to servers are now more aggressive.
3049 Details : This builds on the enhanced upcall-less recovery that was added
3050 in 1.4.2. When trying to reconnect to servers, clients will
3051 now try each server in the failover group every 10 seconds. By
3052 default, clients would previously try one server every 25 seconds.
3057 Description: After recovery, certain operations trigger a failed
3058 assertion on a client.
3059 Details : Failing over an mds, using lconf -d --failover, while a
3060 client was doing a readdir() call would cause the client to
3061 LBUG after recovery completed and the readdir() was resent.
3063 Severity : enhancement
3065 Description: Default groups are now added by lconf
3066 Details : You can now run lconf --group <servicename> without having to
3067 manually add groups with lmc.
3070 Frequency : occasional
3072 Description: Nodes with an elan id of 0 trigger a failed assertion
3075 Frequency : always when accessing e.g. tty/console device nodes
3077 Description: tty and some other devices nodes cannot be used on lustre
3078 Details : file's private_data field is used by device data and lustre
3079 values in there got lost. New field was added to struct file to
3080 store fs-specific private data.
3083 Frequency : when exporting Lustre via NFS
3085 Description: NFSD failed occasionally when looking up a path component
3086 Details : NFSD is looking up ".." which was broken in ext3 directories
3087 that had grown large enough to become hashed.
3090 Frequency : Clusters with multiple interfaces not on the same subnet
3092 Description: Nodes will repeatedly try to reconnect to an interface which it
3093 cannot reach and report an error to the log.
3094 Details : Extra peer list entries will be created by lconf with some peers
3095 unreachable. lconf now validates the peer before adding it.
3098 Frequency : Only if a default stripe is set on the filesystem root.
3100 Description: Setting a default stripe on the filesystem root prevented the
3101 filesystem from being remounted.
3102 Details : The client was sending extra request flags in the root getattr
3103 request and did not allocate a reply buffer for the dir EA.
3106 Frequency : occasional, higher if lots of files are accessed by one client
3107 Bugzilla : 6159, 6097
3108 Description: Client trips assertion regarding lsm mismatch/magic
3109 Details : While revalidating inodes the VFS looks up inodes with ifind()
3110 and in rare cases can find an inode that is being freed.
3111 The ll_test_inode() code will free the lsm during ifind()
3112 when it finds an existing inode and then the VFS later attaches
3113 this free lsm to a new inode.
3117 Bugzilla : 6422, 7030
3118 Description: MDS deadlock between mkdir and client eviction
3119 Details : Creating a new file via mkdir or mknod (starting a transaction
3120 and getting the ns lock) can deadlock with client eviction
3121 (gets ns lock and trying to finish a synchronous transaction).
3124 Frequency : occasional
3125 Description: While starting a server, the fsfilt_ext3 module could not be
3127 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3128 kernels. Previously, lconf would still use the ext3 name
3129 when trying to load modules. Now, it will correctly use
3130 ext3 on 2.4 and ldiskfs on 2.6.
3132 Severity : enhancement
3133 Description: The default stripe count has been changed to 1
3134 Details : The interpretation of the default stripe count (0, to lfs
3135 or lmc) has been changed to mean striping across a single
3136 OST, rather than all available. For general usage we have
3137 found a stripe count of 1 or 2 works best.
3139 Severity : enhancement
3140 Description: Add support for compiling against Cray portals.
3141 Details : Conditional compiling for some areas that are different
3145 Frequency : occasional
3146 Bugzilla : 6409, 6834
3147 Description: Creating files with an explicit stripe count may lead to
3148 a failed assertion on the MDS
3149 Details : If some OSTs are full or unavailable, creating files may
3150 trigger a failed assertion on the MDS. Now, Lustre will
3151 try to use other servers or return an error to the
3155 Frequency : occasional
3157 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3158 caused lock timeout during readahead (since 1.4.2).
3159 Details : Processes doing readahead might match a lock that hasn't been
3160 granted yet if there are overlapping and conflicting lock
3161 requests. The readahead process waits on ungranted lock
3162 (original lock is CBPENDING), while OST waits for that process
3163 to cancel CBPENDING read lock and eventually evicts client.
3165 Severity : enhancement
3167 Description: Initial enabling of flock support for clients
3168 Details : Implements fcntl advisory locking and file status functions.
3169 This feature is provided as an optional mount flag (default
3170 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3171 locking are implemented yet, and those that are are not guaranteed
3172 to be completely correct in production environments.
3173 mount -t lustre -o [flock|noflock] ...
3176 Frequency : occasional
3178 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3179 assertion in the ext3 JBD (journaling) layer.
3180 Details : The b_committed_data struct is protected by the big kernel lock
3181 in 2.4 kernels, serializing journal_commit_transaction() and
3182 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3183 access to this struct. In 2.6 kernels there is finer grained
3184 locking to improve SMP performance of the JBD layer.
3188 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3193 Description: merge and cleanup kernel patches.
3196 -----------------------------------------------------------------------------
3198 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3200 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3201 versions of Lustre. Please update all servers and clients to
3202 version 1.4.2 or later at the same time. You must also run
3203 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3204 to update the configuration logs.
3206 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3207 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3208 - swab llog records of type '0' so we get proper header size/idx (5861)
3209 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3210 - fix rename of one directory over another leaking an inode (5953)
3211 - avoid SetPageDirty on 2.6 (5981)
3212 - don't re-add just-being-destroyed locks to the waiting list (5653)
3213 - when creating new directories, inherit the parent's custom
3214 striping settings if present parent (3048)
3215 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3216 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3217 - don't hold client locks on temporary worklist from l_lru (5666)
3218 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3219 - automatically evict dead clients (5921)
3220 - Update file size properly in create+truncate+fstat case (6196)
3221 - Do not unhash mountpoint dentries, do not allow removal of
3223 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3224 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3225 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3226 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3227 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3228 - Correctly handle reads of files with no objects (6243)
3229 - lctl recover will also mark a device active if deactivate used (5933)
3231 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3232 - allow --write-conf on an MDS with different nettype than client (5619)
3233 - don't write config llogs to MDS for mounts not from that MDS (5617)
3234 - lconf should create multiple TCP connections from a client (5201)
3235 - init scripts are now turned off by default; run chkconfig --on
3236 lustre and chkconfig --on lustrefs to use them
3237 - upcalls are no longer needed for clients to recover to failover
3239 - add --abort-recovery option to lconf to abort recovery on device
3241 - add support for an arbitrary number of OSTs (3026)
3242 - Quota support protocol changes.
3243 - forward compatibility changes to wire structs (6007)
3244 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3245 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3246 - improved "lustre status" script
3247 - initialize blocksize for non-regular files (6062)
3248 - added --disable-server and --disable-client configure options (5782)
3249 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3250 - Vanilla 2.4.29 support
3251 - increase maximum number of obd devices to 520 (6242)
3252 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3253 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3254 SLES 9 2.6.5 kernels
3255 - limit stripes per file to 160 (the maximum EA size) (6093)
3257 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3260 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3261 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3262 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3263 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3264 - small ext3 extents cleanups and fixes (5733)
3265 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3266 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3267 - avoid needless client->OST connect, fix handle mismatch (5317)
3268 - fix DLM error path that led to out-of-sync client, long delays (5779)
3269 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3270 - fix several locking issues related to i_size (5492,5624,5654,5672)
3271 - don't move pending lock onto export if it is already evicted (5683)
3272 - fix kernel oops when creating .foo in unlinked directory (5548)
3273 - fix deadlock in obdfilter statistics vs. object create (5811)
3274 - use time_{before,after} to avoid timer jiffies wrap (5882)
3275 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3276 - Do not leak request if server was not able to process it (5154)
3277 - If mds_open unable to find parent dir, make that negative lookup(5154)
3278 - don't create new directories with extent-mapping (5909, 5936)
3280 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3281 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3282 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3283 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3284 - support for RHEL4 kernel on i686 (5773)
3285 - provide error messages when incompatible logs are encountered (5898)
3287 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3288 * version 1.4.0.10 (1.4.1 release candidate 1)
3290 - don't keep a lock reference when lock is not granted (4238)
3291 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3292 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3293 - drop import inflight refcount on signal_completed_replay error (5255)
3294 - unlock page after async write error during send (3677)
3295 - handle missing objects in filter_preprw_read properly (5265)
3296 - no transno return for symlink open, don't save no-trasno open (3440)
3297 - don't try to complete elan receive that already failed (4012)
3298 - free RPC server reply state on error (5406)
3299 - clean up thread from ptlrpc_start_thread() on error (5160)
3300 - readahead could read extra page into cache that wasn't ejected (5388)
3301 - prevent races in class_attach/setup/cleanup/detach (5260)
3302 - don't dereference de->d_inode after l_dput of de (5458)
3303 - use "int" for stripe value returned from lock_to_stripe (5544)
3304 - mballoc allocation and error-checking fixes in 2.6 (5504)
3305 - block device patches to fix I/O request sizes in 2.6 (5482)
3306 - look up hostnames for IB nals (5602)
3307 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3308 - don't start multiple acceptors for the same port (5277)
3309 - fix incorrect LASSERT in mds_getattr_name (5635)
3310 - export a proc file for general "ping" checking (5628)
3311 - fix "lfs check" to not block when the MDS is down (5628)
3313 - service request history (4965)
3314 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3315 - create an "evict_client" /proc entry on OSTs, like the MDS has
3316 - fix mount usage message, return errors per mount(8) (5168)
3317 - change grep [] to grep "[]" in tests so they work in more UMLs
3318 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3319 - remove ancient LOV_MAGIC_V0 EA support (5047)
3320 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3321 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3322 - allow readahead allocations to fail when low on memory (5383)
3323 - mmap locking landed again, after considerable improvement (2828)
3324 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3326 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3329 - send OST transaction number in read/write reply to free req (4966)
3330 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3331 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3332 - the watchdog thread now runs as interruptible (5246)
3333 - flock/lockf fixes (but it's still disabled, pending 5135)
3334 - don't use EXT3 constants in llite code (5094)
3335 - memory shortage at startup could cause assertion (5176)
3337 - reorganization of lov code
3338 - single portals codebase
3340 - add extents/mballoc support (5025)
3341 - direct I/O reads in the obdfilter (4048)
3342 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3344 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3347 - send OST transaction number in read/write reply to free req (4966)
3348 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3349 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3350 - the watchdog thread now runs as interruptible (5246)
3351 - handle missing objects in filter_preprw_read properly (5265)
3352 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3353 - drop import inflight refcount on signal_completed_replay error (5255)
3354 - unlock page after async write error during send (3677)
3355 - return original error code on reconstructed replies (3761)
3356 - no transno return for symlink open, don't save no-trasno open (3440)
3358 - add pid to ldlm debugging output (4922)
3359 - bump the watchdog timeouts -- we can't handle 30sec yet
3360 - extra debugging for orphan dentry/inode bug (5259)
3362 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3365 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3366 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3367 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3368 - fix NULL dereference in /proc/sys/portals/routes (4827)
3369 - allow failed mdc_close() operations to be interrupted (4561)
3370 - stop precreate on OST before MDS would time out on it (4778)
3371 - don't send partial-page writes before EOF from client (4410)
3372 - discard client grant for sub-page writes on large-page clients (4520)
3373 - don't free dentries not owned by NFS code, check generation (4806)
3374 - fix lsm leak if mds_create_objects() fails (4801)
3375 - limit debug_daemon file size, always print CERROR messages (4789)
3376 - use transno after validating reply (3892)
3377 - process timed out requests if import state changes (3754)
3378 - update mtime on OST during writes, return in glimpse (4829)
3379 - add mkfsoptions to LDAP (4679)
3380 - use ->max_readahead method instead of zapping global ra (5039)
3381 - don't interrupt __l_wait_event() during strace
3383 - add software watchdogs to catch hung threads quickly (4941)
3384 - make lustrefs init script start after nfs is mounted
3385 - fix CWARN/ERROR duplication (4930)
3386 - return async write errors to application if possible (2248)
3387 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3388 - print NAL number in %x format (4645)
3389 - update barely-supported suse-2.4.21-171 series (4842)
3390 - support for sles 9 %post scripts
3391 - support for building 2.6 kernel-source packages
3392 - support for sles km_* packages
3394 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3397 - ignore -ENOENT errors in osc_destroy (3639)
3398 - notify osc create thread that OSC is being cleaned up (4600)
3399 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3400 - reconstruct ost_handle() like mds_handle() (4657)
3401 - create a new thread to do import eviction to avoid deadlock (3969)
3402 - let lconf resolve symlinked-to devices (4629)
3403 - don't unlink "objects" from directory with default EA (4554)
3404 - hold socknal file ref over connect in case target is down (4394)
3405 - allow more than 32000 subdirectories in a single directory (3244)
3406 - fix blocks count for O_DIRECT writes (3751)
3407 - OST returns ENOSPC from object create when no space left (4539)
3408 - don't send truncate RPC if file size isn't changing (4410)
3409 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3410 - bind to privileged port in socknal and tcpnal (3689)
3412 - rate limit CERROR/CWARN console message to avoid overload (4519)
3413 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3414 - basic mmap support (3918)
3415 - kernel patch series update from b1_4 (4711)
3417 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3420 - avoid crash during MDS cleanup with OST shut down (2775)
3421 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3422 - don't use bad inodes on the MDS (3744)
3423 - dynamic object preallocation to improve recovery speed (4236)
3424 - don't hold spinlock over lock dumping or change debug flags (4401)
3425 - don't zero obd_dev when it is force cleaned (3651)
3426 - print grants to console if they go negative (4431)
3427 - "lctl deactivate" will stop automatic recovery attempts (3406)
3428 - look for existing locks in ldlm_handle_enqueue() (3764)
3429 - don't resolve lock handle twice in recovery avoiding race (4401)
3430 - revalidate should check working dir is a directory (4134)
3432 - don't always mark "slow" obdfilter messages as errors (4418)
3434 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3437 - don't close LustreDB during write_conf until it is done (3860)
3438 - fix typo in lconf for_each_profile (3821)
3439 - allow dumping logs from multiple threads at one time (3820)
3440 - don't allow multiple threads in OSC recovery (3812)
3441 - fix debug_size parameters (3864)
3442 - fix mds_postrecov to initialize import for llog ctxt (3121)
3443 - replace config semaphore with spinlock (3306)
3444 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3445 - don't allow enqueue to complete on a destroyed export (3822)
3446 - down write_lock before checking llog header bitmap (3825)
3447 - recover from lock replay timeout (3764)
3448 - up llog sem before sending rpc (3652)
3449 - reduce ns lock hold times when setting kms (3267)
3450 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3451 - fix NULL deref and obd_dev leak on setup error (3312)
3452 - replace some LBUG about llog ops with error handling (3841)
3453 - don't match INVALID dentries from d_lookup and spin (3784)
3454 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3455 - fix invalid assertion in ptlrpc_set_wait (3880)
3457 - add libwrap support for the TCP acceptor (3996)
3458 - add /proc/sys/portals/routes for non-root route listing (3994)
3459 - allow setting MDS UUID in .xml (2580)
3460 - print the stack of a process that LBUGs (4228)
3462 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3465 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3466 - make sure to unset replay flag from failed open requests (3440)
3467 - if default stripe count is 0, use OST count for inode size (3636)
3468 - update parent mtime/ctime on client for create/unlink (2611)
3469 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3470 - free recovery state on server during a forced cleanup (3571)
3471 - unregister_reply for resent reqs (3063)
3472 - loop back devices mounting and status check on 2.6 (3563)
3473 - fix resource-creation race that can provoke i_size == 0 (3513)
3474 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3475 - more debugging for page-accounting assertion (3746)
3476 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3477 - avoid deadlock after precreation failure (3758)
3478 - fix race and lock order deadlock in orphan handling (3450, 3750)
3479 - add validity checks when grabbing inodes from l_ast_data (3599)
3481 - add /proc/.../recovery_status to obdfilter (3428)
3482 - lightweight CDEBUG infrastructure, debug daemon (3668)
3483 - change default OSC RPC parameters to be better on small clusters
3484 - turn off OST read cache for files smaller than 32MB
3485 - install man pages and include them in rpms (3100)
3486 - add new init script for (un)mounting lustre filesystems (2593)
3487 - run chkconfig in %post for init scripts (3701)
3488 - drop scimac NAL (unmaintained)
3490 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3493 - clean kiobufs before and after use (3485)
3494 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3495 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3496 - change init script to fail more clearly if not run as root (1528)
3497 - allow clients to reconnect during replay (1742)
3498 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3499 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3500 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3501 - fix cond_resched (3554)
3502 - only evict unfinished clients after recovery (3515)
3503 - allow bulk resend, prevent data loss (3570)
3504 - dynamic ptlrpc request buffer allocation (2102)
3505 - don't allow unlinking open directory if it isn't empty (2904)
3506 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3507 - remove extraneous obd dereference causing LASSERT failure (3334)
3508 - don't use get_cycles() when creating temp. files on the mds (3156)
3509 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3510 - handle EEXIST for set-stripe, set proper directory name (3336)
3512 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3513 - fix ksocknal_fmb_callback() error messages (2918)
3515 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3518 - don't copy lvb into (possibly NULL) reply on error (2983)
3519 - don't deref dentry after dput, don't free lvb on error (2922)
3520 - use the kms to determine writeback rpc length (2947)
3521 - increment oti_logcookies when osc is inactive (2948)
3522 - update client's i_blocks count via lvb messages (2543)
3523 - handle intent open/close of special files properly (1557)
3524 - mount MDS with errors=remount-ro, like obdfilter (2009)
3525 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3526 - don't use cancelling-locks' kms values (2947)
3527 - use highest lock extent for kms, not last one (2925)
3528 - don't dereference ERR_PTR() dentry in error handling path (3107)
3529 - fix thread race in portals_debug_dumplog() (3122)
3530 - create lprocfs device entries at setup instead of at attach (1519)
3531 - common AST error handler, don't evict client on completion race (3145)
3532 - zero nameidata in detach_mnt in 2.6 (3118)
3533 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3534 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3535 - initialize RPC timeout timer earlier for 2.6 (3219)
3536 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3537 - print nal/nid for unknown nid (3258)
3538 - additional checks for oscc recovery before doing precreate (3284)
3539 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3540 - don't crash in mdc_close for bad permissions on open (3285)
3541 - zero i_rdev for non-device files (3147)
3542 - clear page->private before handing to FS, better assertion (3119)
3543 - tune the read pipeline (3236)
3544 - fix incorrect decref of invalidated dentry (2350)
3545 - provide read-ahead stats and refine rpc in flight stats (3328)
3546 - don't hold journal transaction open across create RPC (3313)
3547 - update atime on MDS at close time (3265)
3548 - close LDAP connection when recovering to avoid server load (3315)
3549 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3550 - don't leak open file on MDS after open resend (3325)
3551 - serialize filter_precreate and filter_destroy_precreated (3329)
3552 - loop device shouldn't call sync_dev() for nul device (3092)
3553 - clear page cache after eviction (2766)
3554 - resynchronize MDS->OST in background (2824)
3555 - refuse to mount the same filesystem twice on same mountpoint (3394)
3556 - allow llmount to create routes for mounting behind routers (3320)
3557 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3558 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3559 - fix rare problem with rename on htree directories (3417)
3561 - allow default OST striping configuration per directory (1414)
3562 - fix compilation for qswnal for 2.6 kernels (3125)
3563 - increase maximum number of MDS request buffers for large systems
3564 - change liblustreapi to be useful for external progs like lfsck (3098)
3565 - increase local configuration timeout for slow disks (3353)
3566 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3568 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3571 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3572 - don't overwrite extent policy data in reply if lock was blocked (2901)
3573 - drop filter export grants atomically with removal from device (2663)
3574 - del obd_self_export from work_list in class_disconnect_exports (2908)
3575 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3576 - swab reply message in mdc_close, other PPC fixes (2464)
3577 - fix destroying of named logs (2325)
3578 - overwrite old logs when running lconf --write_conf (2264)
3579 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3580 - fix race in target_handle_connect (2898)
3581 - mds_reint_create() should take same inode create lock (2926)
3582 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3583 - don't close files for self_export to avoid uninitialized obd (2936)
3584 - allow MDS with the same name as client node (2939)
3585 - hold dentry reference for closed log files for unlink (2325)
3586 - reserve space for all logs during transactions (2059)
3587 - don't evict page beyond end of stripe extent (2925)
3588 - don't oops on a deleted current working directory (2399)
3589 - handle hard links to targets without a parent properly (2517)
3590 - don't dereference NULL lock when racing during eviction (2867)
3591 - don't grow lock extents when lots of conflicting locks (2919)
3593 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3596 - account for cache space usage on clients to avoid data loss (974)
3597 - lfsck support in lustre kernel code (2349)
3598 - reduce journal credits needed for BRW writes (2370)
3599 - orphan handling to avoid losing space on client/server crashes
3600 - ptlrpcd can be blocked, stopping ALL progress (2477)
3601 - use lock value blocks to assist in proper KMS, faster stat (1021)
3602 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3603 - recovery for initial connections (2355)
3604 - fixes for mds_cleanup_orphans (1934)
3605 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3606 - block all file creations until orphan recovery completes (1901)
3607 - client remove rq_connection from request struct (2423)
3608 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3609 - recovery timer race (2670)
3610 - mdc_close recovey bug (2532)
3611 - ptlrpc cleanup bug (2710)
3612 - mds timeout on local locks (2588)
3613 - namespace lock held during RPCs (2431)
3614 - handle interrupted sync write properly (2503)
3615 - don't try to handle a message that hasn't been replied to (2699)
3616 - client assert failure during cleanup after abort recovery (2701)
3617 - leak mdc device after failed mount (2712)
3618 - ptlrpc_check_set allows timedout requests to complete (2714)
3619 - wait for inflight reqs when ptlrpcd finishes (2710)
3620 - make sure unregistered services are removed from the srv_list
3621 - reset bulk XID's when resending them (caught by 1138 test)
3622 - unregister_bulk after timeout
3623 - fix lconf error (2694)
3624 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3625 - readahead locks pages, leaves pending causing memory pressure (2673)
3626 - increase OST request buffers to 4096 on large machines (2729)
3627 - fix up permission of existing directories in simple_mkdir (2661)
3628 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3629 - don't assign transno to errored transactions (2742)
3630 - don't delete objects on OST if given a bogus objid from MDS (2751)
3631 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3632 - if rq_no_resend, then timeout request after recovery (2432)
3633 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3634 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3635 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3636 - MDS recovery shouldn't delete objects that it has given out (2730)
3637 - if enqueue arrives after completion, don't clobber LVB (2819)
3638 - don't unlock pages twice when trigger_group_io returns error (2814)
3639 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3640 - don't write pages to disk if there was an error (1450)
3641 - don't ping imports that have recovery disabled (2676)
3642 - take buffered bytes into account when balancing socknal conn (2817)
3643 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3644 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3645 - remove little-endian swabbing of llog records (1987)
3646 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3647 - retry reposting request buffers if they fail (1191)
3648 - grow extent at grant time to avoid granting a revoked lock (2809)
3649 - lock revoke doesn't evict page if covered by a second lock (2765)
3650 - disable VM readahead to avoid reading outside lock extents (2805)
3652 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3653 - updated kernel patches for hp-2.4.20 kernel (2681)
3655 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3658 - fix truncated write corruption (2366)
3659 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3661 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3662 - ensure len(uuid) < 37 in lmc (1171)
3663 - fix ia64 OOPS in llog_test (2255)
3664 - zero end of page at obdfilter for partial page writes (2648)
3665 - don't leave stale dentries around after renames (bug 2428)
3666 - fix timeouts when evicting a client with a single lock held (2642)
3667 - set deadline for the initial HELLO message to drain (2634)
3668 - print out dotted-quad IP addresses in the socknal (2302)
3670 - additional debugging for MDS client eviction problem (2443)
3671 - fix mkfsoptions support for osts (2603, 2604)
3673 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3676 - add series for the vanilla 2.6.0 kernel
3677 - add series for the vanilla 2.4.24 kernel
3678 - add series for a cray x86/64 UL kernel drop
3679 - fix xattr patches for the vanilla 2.4.19 series
3681 - generate true UUIDs in lmc (1171)
3682 - have portals stack dumping break in UML (2466)
3683 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3684 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3685 - wake callback waiting threads on client eviction (2460)
3686 - Add --ptldebug and --subsystem to lmc (1719)
3687 - update assertion to allow safe interrupt allocation
3688 - set rq_no_resend for cancel requests (2432)
3689 - recalculate ptlrpcd timeout after resend (2494)
3690 - call vfs_rmdir when removing pending directories (2368)
3691 - fix renaming a file to itself (2429)
3692 - lmc creates a default one-stripe lov (2454)
3693 - expand procfs space to handle large clusters (2326)
3694 - increase UML stack to avoid overflow
3695 - update lconf's list of debug and subsystem masks
3696 - fix lfs find --obd (2510)
3697 - /proc tunable for disabling filter read caching (2591)
3698 - stop rpm packages from altering slapd.conf (2301)
3699 - disable nagle in the socknal under 0conf (2578)
3700 - choose mds inode size based on stripe count (2572)
3701 - fix kernel-source rpm problems (2516)
3703 - add --disable-doc to avoid pdf generation (2421)
3704 - update documentation, tests, type-os, comments
3705 - avoid format warnings on ia64
3706 - remove the TOE NAL
3707 - tiny code cleanups by removing unused fields
3709 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3712 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3713 - remove the most blatant lies from BUILDING file (2371)
3714 - change default debug level to reasonable production setting
3715 - reduce client side cache size to reduce cache flush time
3716 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3717 - make TCP zerocopy and pinger support enabled by default (2476)
3718 - sync writes completed after process exits caused crashes (2319)
3719 - maintain correct mount count on the MDS (2356)
3720 - backout 1557, because 2316 wasn't really fixed
3721 - better file I/O statistics gathering in /proc
3722 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3723 - another kernel patch to fix zero-copy TCP function export
3724 - don't take duplicate lock when processing re-sent getattr (2420)
3725 - lctl uses obd_self_export instead of creating new conn (2353)
3726 - MDS/OST recovery case which requires object creation asserted (2425)
3727 - move lfs from /usr/sbin to /usr/bin in packages
3728 - fix race between mds_client_add and mds_client_free (2417)
3729 - use kmalloc instead of slabs in portals (2430)
3730 - don't create duplicate records when a failover MDS is present (2442)
3731 - remove unnecessary mount age check (2332)
3732 - don't remove directory inodes from locks prematurely (2451)
3733 - don't break if MDS service name is the same as hostname (2103)
3734 - fix races in client write RPC generation when cache full (2482)
3736 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3739 - remove now-unused request->rq_obd (278)
3740 - if an allocation fails, print out how much memory we've used (1933)
3741 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3742 - add the "configurable stack size" patch to most series files (1256)
3743 - ability to write large log records, for 100+ OST configs (2306)
3744 - fix NULL deref when filter_prep fails (2314)
3745 - fix operator precedence error in filter_sync
3746 - dynamic allocation of socknal TX descriptors (2315)
3747 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3748 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3749 - ensure that CWARN messages in Portals always get to the syslog
3750 - __init/__exit are not for prototype decls (ldlm_init/exit)
3751 - x86-64 compile warning fixes
3752 - fix gateway LMC keyword conflict (2318)
3753 - fix MDS lock inversions in getattr/reint paths (1844)
3754 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3755 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3756 - DLM race condition prevented some lock evictions (2328)
3757 - ENOMEM detection and retry on socknal sends (2230)
3758 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3759 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3760 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3761 - better error messages when a client is rejected during recovery (1505)
3762 - avoid cancelling locks which were never granted, after failure (2330)
3763 - fix i_sem/journal inversion in mds_client_add (2333)
3764 - fix truncate/getattr lock cycle deadlock (2334)
3765 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3766 - fix two rare exit paths which could leak an l_lock() ref (2321)
3767 - fencepost error in MDS/OST orphan recovery (2226)
3768 - make log record alignment 8 bytes (1988)
3769 - lstripe now fails when requested offset > ost_count (2237)
3770 - ensure that all kernel series have a complete list.h (1607)
3771 - fix crashes in special-file operations (2316)
3772 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3774 - allow configurable automake binary, for testing new versions
3775 - small update to the lfs documentation
3777 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3779 * fix negative export reference count in fsfilt_sync (2312)
3781 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3782 * release candidate 0.9.1
3784 - orphans are moved into the PENDING directory for possible recovery
3785 - replayed opens now open by fid for orphan/rename safety (1042)
3786 - last close of an orphan inode generates a transno (683)
3787 - chdir() and mount() now pin the directory entry (1020)
3788 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3789 - discard very old requests without processing them (1502)
3790 - remove some common, well-understood CERRORs (1505)
3791 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3792 - clear "grant" flags in OST replies until OST grant code lands (1644)
3793 - fix read performance by not clobbering i_blksize on client (1598)
3794 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3795 - make LRU size a /proc tunable, clears locks when reduced (707)
3796 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3797 - server locks take references on exports now (1558)
3798 - build fixes for 2.4.20-rh trees (1663)
3799 - return an error from lov_create if all OSCs are inactive (1751)
3800 - fix import levels when a reconnect happens without a timeout (1597)
3801 - exit early from mds_open if we get a lookup error (1749)
3802 - partial page read at EOF wouldn't wait for disk before sending (1642)
3803 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3804 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3805 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3806 - write/truncate lock inversion (1639)
3807 - Don't auto-load obdclass, portals modules during cleanup (1495)
3808 - fix timestamps from jumping to "now" (1763)
3809 - extra journal assertions (1648)
3810 - add an extra multiunlink test (1771)
3811 - fix read_record/write_record API (1776)
3812 - fix leak of offset_extent, possible incorrect i_size later (1772)
3813 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3814 - Debugging for the kqswnal_get_idle_tx problems (1820)
3815 - Allow recovery to be attempted multiple times (1536)
3816 - Write out MDS last_rcvd file after it is first created (1600)
3817 - Fix tx_descriptor leak in failed transmit situations (1827)
3818 - ext3 journaling fixes for assertion failure after IO error (1871)
3819 - class_export_put() on freed export after completion AST error (1896)
3820 - Fix revalidate looping in VFS (1322)
3821 - Don't access a freed export during MDS_REINT timeout (1521)
3822 - Add open-unlink recovery support on the MDS (1673,1764)
3823 - Return an error if no MDS data was read from last_rcvd (1946)
3824 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3825 - Don't setup a disk device that doesn't match exported UUID (317)
3826 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3827 - avoid committing NULL handle in force close
3828 - local.sh is now a one-stripe LOV configuration
3829 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3830 - fix ext3 htree duplicate directory entry corruption (1516)
3831 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3832 - update to new LOV EA format (2097)
3833 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3835 2003-06-15 Phil Schwan <phil@clusterfs.com>
3838 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3839 - per-import/export recovery handling (958, 931, 959)
3840 - multiple last-rcvd slots, for serving multiple FSes (949)
3841 - connections are again shared between multiple imp/exports (963, 964)
3842 - "umount -f" would hang if any requests needed to be sent (393, 978)
3843 - avoid pinning large req buffer by copying for queued messages (989)
3844 - add "uuid" to "lctl device" command to help upcalls (991)
3845 - "open" RPCs with transnos would confuse recovery counters (1037)
3846 - do proper endian conversion of all wire messages (288, 340, 891)
3847 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3848 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3849 - fix ptlrpc_connection leak in target_handle_connect (1174)
3850 - fix import refcounting bug in OST and MDS cleanup (1134)
3851 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3852 - fix very unlikely obd_types race condition (501)
3853 - remove osc_open hack for echo_client (1187)
3854 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3855 - a failure in read_inode2 leads to deadlock (1139)
3856 - cancel ack-locks as soon as transaction is committed (1072)
3857 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3858 - make sure to commitrw after any preprw to avoid deadlock (1162)
3859 - failing to execute a file in a lustre FS would lock inode (1203)
3860 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3861 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3862 - obd_uuid2tgt would walk off the end of the list (1255)
3863 - on IA64 the osc would give portals incorrect bulk size (1258)
3864 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3865 - fix lock inversion caused by new llite matching code (1282)
3866 - limit the number of dirty pages on a client to 10MB (1286)
3867 - timed out locks were not being corrected cancelled (1289)
3868 - fix O_DIRECT above 4GB on IA-32 (1292)
3869 * major user-visible changes
3870 - fail out/fail over policy now controlled by the upcall (993)
3872 - add OBD_PING to check server availability and failure (954)
3873 - lustre messages are now sent in sending host order (288, 340, 891)
3874 - add eadatalen to MDS getattr reply (340)
3875 - OST read replies may contain second buffer, with per-page status (593)
3877 2003-03-11 Phil Schwan <phil@clusterfs.com>
3880 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3881 - failed open()s could cause deadlock; fixed (867, 869)
3882 - stop cancelling OST locks when files are closed (481)
3883 - overlapping XID spaces caused network corruption (851, 853)
3884 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3885 - setattr_raw added, to do single-RPC, server-side setattrs
3886 - lmc/lconf syntax change for OST UUIDs
3887 - fix crashy race condition between ptlrpc_free_req and osc_close
3888 - don't use request in mdc_enqueue if we hit a timeout (889)
3889 - don't set the inode i_size for regular files from the MDS (896)
3890 - handle out of order completion AST (842)
3891 - don't LBUG if a lock request times out after receiving AST (913)
3892 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3893 - if a bad lock AST arrives, send an error instead of dropping entirely
3894 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3895 - fix leak in bulk IO when only partially completed (899, 900, 926)
3896 - fix O_DIRECT for ia64 (55)
3897 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3898 - C-z after timeout could hang a process forever; fixed (977)
3900 - client-side I/O cache (678, 924, 929, 941, 970)
3902 - READPAGE and SETATTRs which don't take server-side locks get
3905 2003-02-11 Phil Schwan <phil@clusterfs.com>
3908 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3909 - Fix fsfilt_extN_readpage() to read a full page of directory
3910 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3911 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3912 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3913 - Most intent operations were removed, in favour of a new RPC mode
3914 that does a single RPC to the server and bypasses most of the VFS
3915 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3916 - Aggressively cancel local locks on DLM servers
3917 - mds_reint_unlink sends EA to the client if it's the last nlink.
3918 client uses that EA to unlink OST objects.
3919 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3920 - recursive symlinks were fixed (439)
3921 - fixed NULL deref in DEBUG_REQ
3922 - filter_update_lastobjid no longer calls sync, which annoyed extN
3923 - fixed multi-client small-writes to a single file problem (445)
3924 - fixed mtime updates during file writes (607)
3925 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3926 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3927 - fixed Linux OST size reporting problem (444, 656)
3928 - OST now updates object mtime with writes or setattr (607, 619)
3929 - client verifies file size before zeroing page past EOF (445)
3930 - OST now writes last allocated objid to disk with allocation (108)
3931 - LOV on echo now works (409)
3933 - mds_reint_unlink sends a new buffer, with the EA included. this
3934 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3935 set if a regular file was being unlinked, and it was the last link
3936 - use PtlGet from the target for bulk writes (315)
3937 - OST now updates object mtime with writes or setattr (607, 619)
3938 - LDLM now has a grant-time callback to revalidate locked items, if
3940 - Many MDS operations were reorganized to combat race conditions
3942 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3943 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3945 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3948 - Fully reactivate OST imports after reconnection (512, others)
3949 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3950 - More graceful error handling for truncating on dead OST (515)
3951 - Don't error out unless we're actually accessing dead stripes (474)
3952 - Fix garbage sizes when stripes are missing (410)
3953 - LRU counters were broken, causing constant lock purge (433, 432)
3954 - garbage on read from stripes with failed OSTs (441)
3955 - mark OSCs as active before reconnecting during recovery (438)
3956 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3957 - lfind did not preserve OST order in output (443)
3958 - symlinks cause hung clients, incorrect data (439)
3959 - stop dereferencing request after dropping refcount (457)
3960 - don't LASSERT(spin_is_locked) on non-SMP (455)
3961 - fixes for many rename() bugs
3962 - fstat didn't correctly synchronize attributes (399)
3963 - server must handle lock cancellation during blocking AST prep (487)
3964 - bulk descriptors were free()d too soon (511)
3965 - fix paths in lconf, which would load incorrect modules (451, 507)
3966 - fix confusing lconf 'host not found' error message (386)
3967 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3968 - fix race condition in mdc_blocking_ast() for inode access (526)
3969 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3970 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3971 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3972 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3973 - fix lstripe in several ways (kernel side) (527)
3974 - fix request leak in ldlm_cli_enqueue (262)
3975 - incorrect OSC was marked inactive after OST failure
3976 - call mds_fs_cleanup before unmounting filesystem (524)
3977 - fix races between taking ns_lock and ldlm_lock_change_resource
3978 - fix races updating LOV export open file list
3979 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3980 - fix recovery NULL deref in ldlm_cli_cancel_unused
3981 - fix some DLM races by using new hash table for lock handles (419)
3982 - permit the client to specify desired inodes, at replay
3983 - duplicate requests when we queue them for replay reintegration
3984 - fix last_rcvd offset calculation
3985 - sync after each recovered transaction, so we always make progress
3986 - never, not always, ERESTART requests without transnos
3987 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3989 - skip replay if the MDS says that the client is already connected
3990 - don't check for a recovery-enabled export to match lctl's UUID
3991 - don't INC_USE_COUNT for phantom exports
3992 - don't crash when cleaning up phantom exports (567)
3993 - don't double-finish or set replay data for errored mdc_open requests
3994 - abort requests when they time out, so we don't get old replies
3995 - send/receive replies for AST messages again
3996 - if the client says that it doesn't have the lock, cancel it on the
3998 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3999 mark it as destroyed, it will all work out when decref is called
4000 - fix module use counts (22, 581)
4002 - ASTs now expect a reply (server cancels lock on error reply)
4004 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4007 - fix many simultaneous client startup (392)
4008 - fix dentry->d_it clobbering
4009 - credentials weren't being shipped for readdir/getattr operations
4010 - remove invalid assertions triggered during some concurrent MD
4012 - proper Lustre versions added (336, 389)
4013 - fix memory leak for create error case (398)
4014 - fix LOV locking bug that would get cli/srv out of sync
4015 - fix echo client over LOV (409)
4016 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4017 - fix double-O_EXCL intent crash (424)
4018 - avoid sending multiple lock CANCELs (352)
4020 - MDS can do multi-client recovery (modulo bugs in new code)
4022 - many updates, edits, cleanups
4024 2002-11-18 Phil Schwan <phil@clusterfs.com>
4027 - fix null d_it dereference (346)
4028 - fix full OST/dbench hang (333)
4029 - fix permission problem with file removal (286)
4030 - fix removal of OSCs from LOV when they fail
4031 - fix NULL deref during bulk timeout (214)
4032 - fix problems related to multiple filesystems on one MDS (241)
4033 - fixed serious subtle metadata locking bugs
4034 - free locks on clients when inodes are removed due to memory
4036 - fix inode pointer in lock data (285)
4037 - partial support for multiple MDS on a single host (241)
4038 - data locks weren't cancelled at clear_inode time (290, 311)
4039 - intent locks could lead to unbounded lock growth (205)
4040 - added a maximum lock count, an LRU list, and a flusher
4041 - fix multiple rename (365)
4042 - properly abstracted the echo client
4043 - OSC locked 1 byte too many; fixed
4044 - rewrote brw callback code:
4045 - fixed recovery bugs related to LOVs (306)
4046 - fixed too-many-pages-in-one-write crash (191)
4047 - fixed (again) crash in sync_io_timeout (214)
4048 - probably fixed callback-related race (385)
4050 - Add capability to MDS protocol
4051 - LDLM cancellations and callbacks on different portals
4053 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4056 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4057 - defer open object destruction to close time (601981, 138)
4058 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4059 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4060 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4061 - don't call obd_setattr in ll_file_release for destroyed objects
4062 * protocol change to lustre_msg: move |version| and add |flags|
4063 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4064 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4065 * added replay of create, unlink, link and rename operations during
4066 MDS failover; recovery should be much more robust now
4067 * remove failed OSCs from LOVs (only lov_create uses this so far)
4068 * the lustre-HOWTO was brought (more) up to date (582544)
4070 2002-10-23 Phil Schwan <phil@clusterfs.com>
4073 - in-use dentries weren't being reused properly (617851)
4074 - prevent multiple LDLM setup (599178)
4075 - fix LOV size calculations for truncate (617853)
4076 - fix client handling of MDS intent errors (POSIX)
4077 - fix permission bug in lovstripe.c test (624321)
4078 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4079 - truncate past end of file could corrupt data
4080 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4081 - a race in recovery could return ETIMEDOUT to apps (623947)
4082 - building outside the source directory was fixed
4083 * the lustre-HOWTO was brought (more) up to date (582544)
4084 * major progress was made on recovery functionality
4086 2002-10-10 Phil Schwan <phil@clusterfs.com>
4089 - recovery deadlock fix
4090 - rm -rf causes LBUG fix (617817)
4091 - file open by multiple tasks fix (618962)
4092 - directory permissions bugs (602707 and 620007)
4093 - journal_stop fixed with locking (611313)
4094 - O_APPEND failures resolved (618273, perhaps 614459)
4095 - lconf PATH fix (619770)
4096 - IA64 build fix (621450)
4097 - RPC buffer sizes scale with amount of memory
4099 2002-10-01 Phil Schwan <phil@clusterfs.com>
4102 - locks would be cancelled without throwing away data pages,
4103 resulting in inconsistent data (605627)
4104 - inode attributes were not always being refreshed (605627, 612449)
4105 - lconf now continues to cleanup after lctl reports an error
4106 - MDS now enforces user permissions (602707)
4107 - lprocfs cleanup fixed, but not yet enabled (614157)
4108 - fixed infinite server hang, should a client not respond to an AST
4109 - avoid going into recovery if user calls readlink() with a buffer
4110 that's too small (613941)
4111 - AST RPCs no longer require replies (614867) -- this may be changed
4112 - don't crash server if client sends an IOV that's too big (611336)
4113 - fixed lock conversion deadlock (611892)
4114 - fixed the following of symlinks (614622)
4115 * recovery: the server can remove locks from a client that dies, other
4116 clients can make progress
4117 * more extN patch fixes
4118 * compile-time configurable ptlrpc buffer allocations
4120 - collaborative read cache document
4121 - Lustre Lite Performance CDR document-in-progress
4123 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4126 - fix typo in patch-2.4.18
4128 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4131 - clear ptlrpc request each time in handle_incoming_request()
4132 - unlink of files now destroys the object on the OST
4134 2002-09-19 Peter Braam <braam@clusterfs.com>
4136 * add hard link support
4137 * change obdfile creation method
4138 * kernel patch changed
4140 2002-09-19 Peter Braam <braam@clusterfs.com>
4143 - stack overflow bug in extN fixed
4145 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4147 * documentation updates
4148 - add man pages for config tools
4149 - update tests/README to describe testing with new config tools
4150 - finish metadata API descriptions
4151 * bug fixes and cleanups
4152 - statfs workaround for 16TB limit
4153 - LOV stripe allocation improved, can stripe on subset of OSTs
4154 - LOV file size/IO offset was wrong for files > 4GB in size
4155 - object EA data was being dropped, caused files to be unreadable
4156 - memory overflow with non-LOV OST caused memory corruption
4157 - fixed regression tests to work with new config tools, obdfilter
4158 - fixed bug when directory size became larger than 1 block
4159 - fixed bug (for single client case) when PWD was deleted
4160 - invalidate local directory pages when doing intent-based ops
4161 - avoid LDLM oops when lock callback contained bad data
4163 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4165 * documentation updates
4166 * bug fixes and cleanups
4167 - configuration tools
4170 - 64-bit compile warnings
4171 - 64-bit internal statfs data
4173 * test_brw on persistent OST devices
4175 * lprocfs (disabled)
4177 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4179 * documentation updates
4180 * bug fixes and cleanups
4181 * configuration tools
4183 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4186 * many small fixes to 0_5_4
4187 * io/network handling
4188 * thinkos in MDS operations
4190 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4193 * crucial basic fixes to 0.5.3
4194 * IOR, Iozone work over Elan
4197 2002-08-07 Phil Schwan <phil@clusterfs.com>
4198 * version 0_5_3, our first alpha
4199 * we use the new Portals iovs
4200 * documentation updates
4201 * bug fixes and cleanups
4202 * small changes in the DLM wire protocol
4204 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4205 * version 0_5_1 with some initial stability,
4206 * locking on MD and file I/O.
4207 * documentation updates
4208 * several bug fixes since 0.5.0
4209 * small changes in wire protocol
4211 2002-07-18 Phil Schwan <phil@clusterfs.com>
4213 * delivered as Lustre Light Alpha
4214 * fixed a crash after handling invalid MDS requests
4215 * fixed directory pages for architectures with non-4k pages sizes
4217 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4218 * release version v0_4_4
4219 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4221 * Fixes a number of interruption problems with OST operations.
4222 * Update documentation for portals header changes
4223 * Move all wire protocol structs/defines to lustre_idl.h
4224 * Fixes symlink length bug.
4225 * Add tcpdump to repository.
4227 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4228 * release version v0_4_3
4229 * Fixes statfs for inodes on extN.
4230 * Fixes bug in runtests which would delete /etc/hosts.
4231 * Use 64-bit object IDs wherever possible (not into VFS though)
4232 Remove ost_get_info, which is unused by lustre, and out of date.
4234 2002-07-03 Peter Braam <braam@clusterfs.com>
4235 * release version v0_4_2 Fixes a lookup error (type not passed)
4236 * move forward to head of Portals
4237 * move forward to latest Lustre kernel
4239 2002-06-25 Peter Braam <braam@clusterfs.com>
4240 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob