3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.53-0.16 (SLES 10),
6 2.6.18-53.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-cfs5
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 Severity : enhancement
17 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
21 Frequency : rare, at shutdown
22 Description: access already free / zero obd_namespace.
23 Details : if client_disconnect_export was called without force flag set,
24 and exist connect request in flight, this can produce access to
25 NULL pointer (or already free pointer) when connect_interpret
26 store ocd flags in obd_namespace.
30 Frequency : only at startup
31 Description: not alloc memory with spinlock held.
32 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
35 Severity : enhancement
37 Description: make lustre randomly fail allocating memory
38 Details : Make lustre randomly failed allocating memory for testing purpose.
40 Severity : enhancement
42 Description: lost problems with lov objid file
43 Details : Fixes some scability and access to not inited memory problems
44 in work with lov objdid file.
49 Description: lfs find does not continue on file error
50 Details : Continue other files processing when a file/dir is absent.
54 Description: Inconsistent usage of lustre_pack_reply()
55 Details : Standardize the usage of lustre_pack_reply() such that it
56 always generate a CERROR on failure.
59 Frequency : occasional
61 Description: MDS hang or stay in waiting lock
62 Details : If client receive lock with CBPENDING flag ldlm need send lock
63 cancel as separate rpc, to avoid situation when cancel request
64 can't processed due all i/o threads stay in wait lock.
68 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
72 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
76 Description: Update to SLES9 kernel-2.6.5-7.287.3.
80 Description: Update to RHEL4 latest kernel.
82 Severity : enhancement
84 Description: Build SLES10 patchless client fails
85 Details : The configure was broken by run ./configure with
86 --with-linux-obj=.... argument for patchless client. When the
87 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
88 can't search header adequately. Use absolute path such as
89 -I($LINUX)/include instead.
93 Description: interrupt oig_wait produce painc on resend.
94 Details : brw_redo_request can be used for resend requests from ptlrpcd and
95 private set, and this produce situation when rq_ptlrpcd_data not
96 copyed to new allocated request and triggered LBUG on assert
97 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
98 ptlrpcd set if request is changed and can be safety changed to use
103 Description: LASSERT_{REQ,REP}SWAB macros are buggy
104 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
105 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
106 these macros and replace them with their difinitions instead.
110 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
111 Details : Update which_patch & target file for SLES10 latest kernel.
114 Description: Remove client patches from SLES 10 kernel.
115 Details : This causes SLES 10 clients to behave as patchless clients
116 even on a Lustre-patched (server) kernel.
118 Severity : enhancement
120 Description: self-adjustable client's lru lists
121 Details : use adaptive algorithm for managing client cached locks lru
122 lists according to current server load, other client's work
123 pattern, memory activities, etc. Both, server and client
124 side namespaces provide number of proc tunables for controlling
127 Severity : enhancement
129 Description: light-weight GSS support
130 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
131 and reduce performance overhead.
133 Severity : enhancement
135 Description: Linux keyring support
136 Details : Support using service of Linux keyring for Lustre GSS internal
137 context refresh/cache mechanism.
141 Description: Fix errors in lfs documentation
142 Details : Fixes man pages
146 Description: don't use GFP_* in generic Lustre code.
147 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
151 Description: obdclass is limited by single OBD_ALLOC(idarray)
152 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
156 Description: warnings with build patchless client with vanila 2.6.19 and up
157 Details : change old ctl_table style and replace ctl_table/ctl_table_header
158 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
162 Frequency : Only for SLES
163 Description: sanity_quota fail test_1
164 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
165 file, which makes TSTID[2] inunique.
169 Frequency : Only for SLES9
170 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
171 Details : Update target/ChangeLog/which_patch .
175 Frequency : Only for test_10 in sanity-quota.sh
176 Description: a bug in quota test code
177 Details : checking a test flag in a wrong place
181 Frequency : Only for test_13 in sanity-quota.sh
182 Description: a bug in quota test script
183 Details : shouldn't check overfull allocated quota
187 Description: jbd statistics
188 Details : Port older jbd statistics patch for sles10
190 Severity : enhancement
192 Description: Add printing inode info into message about error in writepage.
197 Description: reply_lock_interpret crash due to race with it and lock cancel.
198 Details : Do not replay locks that are being cancelled. Do not reference
199 locks by their address during replay, just by their handle.
203 Frequency : When flocks are used.
204 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
205 Details : Flock locks might destroy just granted lock if it could be merged
206 with another existing flock, this is done in completion handler,
207 so teach ldlm_cli_enquque_fini that this is a valid case for
213 Description: Oops in read and write path when failing to allocate lock.
214 Details : Check if lock allocation failed and return error back.
218 Description: lstripe command fails for valid OST index
219 Details : The stripe offset is compared to lov->desc.ld_tgt_count
220 instead of lov->desc.ld_active_tgt_count.
224 Description: sanity.sh failed test 103
225 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
230 Description: df doesn't work properly if diskfs blocksize != 4K
231 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
235 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
236 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
240 Description: obd_health_check_timeout too short
241 Details : set obd_health_check_timeout as 1.5x of obd_timeout
245 Description: llapi_file_create() does not allow some changes
246 Details : add llapi_file_open() that allows specifying the mode and
247 open flags, and also returns an open file handle.
251 Description: merge and cleanup kernel patches.
252 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
256 Description: Add journal checksum support.(Kernel part)
257 Details : The journal checksum feature adds two new flags i.e
258 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
259 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
260 indicates that the commit block contains the checksum for
261 the blocks described by the descriptor blocks. Now commit
262 record can be sent to disk without waiting for descriptor
263 blocks to be written to disk. This behavior is controlled
264 using JBD2_FEATURE_ASYNC_COMMIT flag.
268 Description: OSS needs mutliple precreate threads
269 Details : Add ability to start more than one create thread per OSS.
273 Description: Sanity.sh test_65k failed: setstripe should have succeeded
274 Details : execute lfs setstripe on client
278 Description: mds_obd_create error creating tmp object
279 Details : When the user sets quota on root, llog will be affected and can't
280 create files and write files.
283 Frequency : Always on ia64 patchless client, and possibly others.
285 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
286 Details : This allows the patchless client to be loaded on architectures
291 Description: RedHat Update kernel for RHEL5
292 Details : Add the kernel config file for RHEL5.
296 Description: RedHat Update kernel for RHEL5
297 Details : Modify the kernel config file more closer RHEL5.
301 Description: Build failure against Centos5 (RHEL5)
302 Details : Define PAGE_SIZE when it isn't present.
306 Description: test_7 on sanity-quota.sh can't be used on separate machines
307 Details : add facet to handle it
311 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
312 Details : with NFS, the anon dentry's parent was set to itself in
313 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
314 corresponding dentry other than use rec->ur_name.
316 Severity : enhancement
318 Description: lfs setstripe enhancement
319 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
323 Description: enable data checksumming by default
324 Details : enable checksum by default, allow --disable-checksum
325 configure option and "-o nochecksum" mount option. Checksums
326 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
330 Description: lustre support for RHEL5
331 Details : Add support for RHEL5.
335 Description: Client eviction due to failover config
336 Details : after a connection loss, the lustre client should attempt to
337 reconnect to the last active server first before trying the
338 other potential connections.
342 Description: when mds and osts use different quota unit(32bit and 64bit),
343 quota will be released repeatly.
344 Details : void sending multiple quota reqs to mds, which will keep the
345 status between the reqs.
349 Description: osts not allocated evenly to files
350 Details : change the condition to increase offset_idx
353 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
355 Description: Data corruption for OSTs that are formatted larger than 2TB
357 Details : When generating the bio request for lustre file writes the
358 sector number would overflow a temporary variable before being
359 used for the IO. The data reads correctly from Lustre (which
360 will overflow in a similar manner) but other file data or
361 filesystem metadata may be corrupted in some cases.
365 Description: Tune the kernel for good SCSI performance.
366 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
367 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
372 Description: rewrite ext2-derived code in obdclass/uuid.c
373 Details : rewrite inherited code (uuid parsing code from ext2 utils)
374 from scratch preserving functionality.
378 Description: "lfs find -obd UUID" prints directories
379 Details : "lfs find -obd UUID" will return all directory names instead
380 of just file names. It is incorrect because the directories
381 do not reside on the OSTs.
385 Description: use i_size_read and i_size_write in 2.6 port
386 Details : replace inode->i_size access with i_size_read/write()
389 Frequency : when removing large files
391 Description: scheduling issue during removal of large Lustre files
392 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
393 It causes scheduling issues when removing large files (17TB in the
397 Frequency : only with liblustre clients on XT3
399 Description: evictions taking too long
400 Details : allow llrd to evict clients directly on OSTs
403 Frequency : only on ppc
405 Description: /proc/fs/lustre/devices broken on ppc
406 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
407 We should make sure the type of 'index' is loff_t and then cast
408 explicitly as needed below. Do not assign an explicitly cast
412 Frequency : only for rhel5
414 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
415 Details : Modify the target file & which_kernel.
417 Severity : enhancement
419 Description: omit set fsid for export NFS
420 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
421 as FS_REQUIRES_DEV for avoid problems with generate fsid.
425 Frequency : Always, for kernels after 2.6.16
426 Description: Fix warning idr_remove called for id=.. which is not allocated.
427 Details : Last kernels save old s_dev before kill super and not allow
428 to restore from callback - restore it before call kill_anon_super.
432 Description: buffer overruns could theoretically occur
433 Details : llapi_semantic_traverse() modifies the "path" argument by
434 appending values to the end of the origin string, and a
435 overrun may occur. Adding buffer overrun check in liblustreapi.
439 Description: Fix error on 'ls .' at the top of the Lustre mount.
440 Details : Don't revalidate dentry if it is a root dentry.
444 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
445 Details : Modify vm-tunables-rhel4.patch.
449 Description: Kernel config for 2.6.18-vanilla.
450 Details : Modify targets/2.6-vanilla.target.in.
451 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
452 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
453 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
454 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
458 Description: improve handling recoverable errors
459 Details : if request processig with error which can be recoverable on server
460 request should be resend, otherwise page released from cache and
465 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
466 Details : Modify target file & which_patch.
467 A flaw was found in the IA32 system call emulation provided
468 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
469 value could be stored in the %RAX register, which could trigger an
470 out-of-bounds system call table access. An untrusted local user
471 could exploit this flaw to run code in the kernel
472 (ie a root privilege escalation). (CVE-2007-4573).
476 Description: O_DIRECT bypasses client statistics.
477 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
478 read_ahead_stats were updated. copied the stats section from
479 osc_send_oap_rpc() into async_internal().
483 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
487 Description: change order of libsysio includes
488 Details : '#include sysio.h' should always come before '#include xtio.h'
492 Description: handle "serious error: objid * already exists" more gracefully
493 Details : If LAST_ID value on disk is smaller than the objects existing in
494 the O/0/d* directories, it indicates disk corruption and causes an
495 LBUG(). If the object is 0-length, then we should use the existing
496 object. This will help to avoid a full fsck in most cases.
498 Severity : enhancement
500 Description: adapt the lustre_config script to support the upgrade case
501 Details : Add "-u" option for lustre_config script to support upgrading 1.4
502 server targets to 1.6 in parallel.
506 Description: To avoid grant space > avaible space when the disk is almost
507 full. Without this patch you might see the error "grant XXXX >
508 available" or some LBUG about grant, when the disk is almost
510 Details : In filter_check_grant, for non_grant cache write, we should
511 check the left space by if (*left > ungranted + bytes), instead
512 of (*left > ungranted), because only we are sure the left space
513 is enough for another "bytes", then the ungrant space should be
514 increase. In client, we should update cl_avail_grant only there
515 is OBD_MD_FLGRANT in the reply.
519 Description: Update RHEL 4 kernel to fix local root privilege escalation.
520 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
521 described in CVE-2007-4573. This problem could allow untrusted
522 local users to gain root access.
525 Frequency : when using O_DIRECT and quotas
527 Description: Incorrect file ownership on O_DIRECT output files
528 Details : block usage reported by 'lfs quota' does not take into account
529 files that have been written with O_DIRECT.
534 Description: touch file failed when fs is not full
535 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
536 otherwise we can get ENOSP while fs is not full.
540 Description: parallel lock callbacks
541 Details : Instead of sending blocking and completion callbacks as separated
542 requests, adding them to a set and sending in parallel.
545 Frequency : only for Cray XT3
546 Bugzilla : 12829/13455
547 Description: Changing primary group doesn't change the group lustre assigns to
549 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
550 overridden with the primary group provided by the group upcall,
551 whereas the supplied fsgid can be trusted if it is in the list of
552 supplementary groups returned by the group upcall.
554 --------------------------------------------------------------------------------
556 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
558 * Support for kernels:
559 2.6.5-7.283 (SLES 9),
560 2.6.9-55.EL (RHEL 4),
561 2.6.16.46-0.14 (SLES 10),
562 2.6.18.8 vanilla (kernel.org)
563 * Client support for unpatched kernels:
564 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
565 2.6.16 - 2.6.22 vanilla (kernel.org)
566 * Due to recently discovered recovery problems, we do not recommend
567 using patchless RHEL 4 clients with this or any earlier release.
568 * Recommended e2fsprogs version: 1.39.cfs8
569 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
570 * Starting with this release, the ldiskfs backing filesystem required
571 by Lustre is now in its own package, lustre-ldiskfs. This package
572 should be installed. It is versioned separately from Lustre and
573 may be released separately in future.
577 Description: block reactivating mgc import until all deactivates complete
578 Details : Fix race when failing back MDT/MGS to itself (testing)
581 Frequency : at statup only
583 Description: mds_lov_synchronize race leads to various problems
584 Details : simultaneous MDT->OST connections at startup can cause the
585 sync to abort, leaving the OSC in a bad state.
587 Severity : enhancement
589 Description: add optional extra BUILD_VERSION info
590 Details : add a new environment variable (namely LUSTRE_VERS) which allows
591 to override the lustre version.
594 Frequency : 2.6.18 servers only
596 Description: ll_kern_mount() doesn't release the module reference
597 Details : The ldiskfs module reference count never drops down to 0
598 because ll_kern_mount() doesn't release the module reference.
603 Description: server LBUG when using old ost_num_threads parameter
604 Details : Accept the old ost_num_threads parameter but warn that it
605 is deprecated, and fix an off-by-one error that caused an LBUG.
610 Description: Transient SCSI error results in persistent IO issue
611 Details : iobuf->dr_error is not reinitialized to 0 between two
615 Frequency : sometimes when underlying device returns I/O errors
617 Description: OSTs not going read-only during write failures
618 Details : OSTs are not remounted read-only when the journal commit threads
619 get I/O errors because fsfilt_ext3 calls journal_start/stop()
620 instead of the ext3 wrappers.
624 Description: poor connect scaling with increasing client count
625 Details : Don't run filter_grant_sanity_check for more than 100 exports
626 to improve scaling for large numbers of clients.
629 Frequency : SLES10 only
631 Description: sanity-quota.sh quotacheck failed: rc = -22
632 Details : Quotas cannot be enabled on SLES10.
635 Frequency : liblustre clients only
637 Description: getdirentries does not give error when run on compute nodes
638 Details : getdirentries does not fail when the size specified as an argument
639 is too small to contain at least one entry
641 Severity : enhancement
643 Description: Add LNET router traceability for debug purposes
644 Details : If a checksum failure occurs with a router as part of the
645 IO path, the NID of the last router that forwarded the bulk data
646 is printed so it can be identified.
651 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
652 Details : Due to a race condition, liblustre clients were occasionally
655 Severity : enhancement
657 Description: lfs setstripe use optional parameters instead of postional
660 Severity : enhancement
662 Description: Nanosecond timestamp support for ldiskfs
663 Details : The on-disk ldiskfs filesystem has added support for nanosecond
664 resolution timestamps. There is not yet support for this at
665 the Lustre filesystem level.
668 Frequency : during server recovery
670 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
671 Details : request with rq_no_resend flag not awake l_wait_event if they get a
675 Frequency : nfs export on patchless client
677 Description: connectathon hang when test nfs export over patchless client
678 Details : Disconnected dentry cannot be found with lookup, so we do not need
679 to unhash it or make it invalid
682 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
683 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
684 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
689 Description: open req refcounting wrong on reconnect
690 Details : If reconnect happened between getting open reply from server and
691 call to mdc_set_replay_data in ll_file_open, we will schedule
692 replay for unreferenced request that we are about to free.
693 Subsequent close will crash in variety of ways.
694 Check that request is still eligible for replay in
695 mdc_set_replay_data().
700 Description: disable writes to filesystem when reading health_check file
701 Details : the default for reading the health_check proc file has changed
702 to NOT do a journal transaction and write to disk, because this
703 can cause reads of the /proc file to hang and block HA state
704 checking on a healthy but otherwise heavily loaded system. It
705 is possible to return to the previous behaviour during configure
706 with --enable-health-write.
708 Severity : enhancement
710 Description: 64-bit inode version
711 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
712 to the inode. This will be required for version-based recovery.
717 Description: MDS fails to start if a duplicate client export is detected
718 Details : in some rare cases it was possible for a client to connect to
719 an MDS multiple times. Upon recovery the MDS would detect this
720 and fail during startup. Handle this more gracefully.
722 Severity : enhancement
724 Description: Add -o localflock option to simulate old noflock
726 Details : This will achieve local-only flock/fcntl locks
732 Description: log_commit_thread vs filter_destroy race leads to crash
733 Details : Take import reference before releasing llog record semaphore
738 Description: Wrong request locking in request set processing
739 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
740 imp_delayed_list, in this place should be used imp_lock.
743 Frequency : when reconnection
745 Description: Grant Leak when osc reconnect to OST
746 Details : When osc reconnect ost, OST(filter) should check whether it
747 should grant more space to client by comparing fed_grant and
748 cl_avail_grant, and return the granted space to client instead
749 of "new granted" space, because client will call osc_init_grant
750 to update the client grant space info.
753 Frequency : when client reconnect to OST
755 Description: Grant Leak when osc do resend and replay bulk write
756 Details : When osc reconnect to OST, OST(filter)should clear grant info of
757 bulk write request, because the grant info while be sync between
758 OSC and OST when reconnect, and we should ignore the grant info
759 these of resend/replay write req.
764 Description: Grant space more than avaiable left space sometimes.
765 Details : When then OST is about to be full, if two bulk writing from
766 different clients came to OST. Accord the avaliable space of the
767 OST, the first req should be permitted, and the second one
768 should be denied by ENOSPC. But if the seconde arrived before
769 the first one is commited. The OST might wrongly permit second
770 writing, which will cause grant space > avaiable space.
773 Frequency : when client is evicted
775 Description: Grant might be wrongly erased when osc is evicted by OST
776 Details : when the import is evicted by server, it will fork another
777 thread ptlrpc_invalidate_import_thread to invalidate the
778 import, where the grant will be set to 0. While the original
779 thread will update the grant it got when connecting. So if
780 the former happened latter, the grant will be wrongly errased
781 because of this race.
786 Description: Checking Stale with correct fid
787 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
788 to check whether it is stale, because sometimes, we want the
789 enqueue happened anyway, and op_data.fid2 will not be initialized.
791 Severity : enhancement
793 Description: update patchless client
794 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
797 Frequency : only with 2.4 kernel
799 Description: random memory corruption
800 Details : size of struct ll_inode_info is to big for union inode.u and this
801 can be cause of random memory corruption.
806 Description: Memory leak in recovery
807 Details : Lov_mds_md was not free in an error handler in mds_create_object.
808 It should also check obd_fail before fsfilt_start, otherwise if
809 fsfilt_start return -EROFS,(failover mds during mds recovery).
810 then the req will return with repmsg->transno = 0 and rc = EROFS.
811 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
812 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
813 be freed no matter whether fsfilt_commit success or not.
816 Frequency : only with huge count clients
818 Description: Prevents from taking the superblock lock in llap_from_page for
820 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
821 for llap_from_page call in ll_removepage prevents from taking the
822 superblock lock for a soon died page.
827 Description: Not check open intent error before release open handle
828 Details : in some rare cases, the open intent error is not checked before
829 release open handle, which may cause
830 ASSERTION(open_req->rq_transno != 0), because it tries to release
831 the failed open handle.
836 Description: Set cat log bitmap only after create log success.
837 Details : in some rare cases, the cat log bitmap is set too early. and it
838 should be set only after create log success.
842 Description: Accessing a block bevice can re-enable I/O when Lustre is
843 tearing down a device.
844 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
849 Description: service threads may hog cpus when there are a lot of requests
850 Details : Insert cond_resched to give other threads a chance to use some CPU
855 Description: the cat log was not initialized in recovery
856 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
857 unlink log on mds will not be initialized until mds post
858 recovery. And also in mds post recovery, the unlink log will
859 initialization will be done asynchronausly, so there will be race
860 between add unlink log and unlink log initialization.
864 Description: brw_stats were being printed incorrectly
865 Details : brw_stats were being printed as log2 but all of them were not
866 recorded as log2. Also remove some code duplication arising from
867 filter_tally_{read,write}.
871 Frequency : rare, only in recovery.
872 Description: ASSERTION(req->rq_type != LI_POISON) failed
873 Details : imp_lock should be held while iterating over imp_sending_list for
874 prevent destroy request after get timeout in ptlrpc_queue_wait.
878 Description: replay-single.sh test 52 fails
879 Details : A lock's skiplist need to be cleanup when it being unlinked
880 from its resource list.
884 Description: Short directio read returns full requested size rather than
886 Details : Direct I/O operations should return actual amount of bytes
887 transferred rather than requested size.
889 Severity : enhancement
891 Description: metadata RPC reduction (e.g. for rm performance)
892 Details : decrease the amount of synchronous RPC between clients and servers
893 by canceling conflicing lock before the operation on the client
894 and packing thier handles into the main operation RPC to server.
896 Severity : enhancement
898 Description: Async OSC create to avoid the blocking unnecessarily.
899 Details : If a OST has no remain object, system will block on the creating
900 when need to create a new object on this OST. Now, ways use
901 pre-created objects when available, instead of blocking on an
902 empty osc while others are not empty. If we must block, we block
903 for the shortest possible period of time.
905 Severity : enhancement
907 Description: refine locking for avoid write wrong info into lov_objid file
908 Details : fix possible races with add new target and write/update data in
911 --------------------------------------------------------------------------------
913 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
918 Frequency : on some architectures
920 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
921 Details : Uninitialized flags sometimes cause configuration commands to
925 Frequency : patchless clients only
927 Description: missing __iget() symbol export
928 Details : The __iget() symbol export is missing. To avoid the need for
929 this on patchless clients the deathrow inode reaper is turned
930 off, and we depend on the VM to clean up old inodes. This
931 dependency was during via the fix for bug 12181.
933 --------------------------------------------------------------------------------
935 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
937 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
938 INTEROPERATE with older versions automatically. In many cases a
939 special upgrade step is needed. Please read the
940 user documentation before upgrading any part of a live system.
941 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
942 version will not interoperate with 1.6 betas before beta5 (1.5.95).
943 * WARNING: Lustre configuration and startup changes are required with
944 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
950 Severity : enhancement
952 Description: MountConf
953 Details : Lustre configuration is now managed via mkfs and mount
954 commands instead of lmc and lconf. New obd types (MGS, MGC)
955 are added for dynamic configuration management. See
956 https://mail.clusterfs.com/wikis/lustre/MountConf for
959 Severity : enhancement
961 Description: dynamic OST addition
962 Details : OSTs can now be added to a live filesystem
964 Severity : enhancement
966 Description: startup order invariance
967 Details : MDTs and OSTs can be started in any order. Clients only
968 require the MDT to complete startup.
970 Severity : enhancement
972 Description: parallel, asynchronous orphan cleanup
973 Details : orphan cleanup is now performed in separate threads for each
974 OST, allowing parallel non-blocking operation.
976 Severity : enhancement
978 Description: optimized stripe assignment
979 Details : stripe assignments are now made based on ost space available,
980 ost previous usage, and OSS previous usage, in order to try
981 to optimize storage space and networking resources.
983 Severity : enhancement
985 Description: Permanently set tunables
986 Details : All writable /proc/fs/lustre tunables can now be permanently
987 set on a per-server basis, at mkfs time or on a live system.
989 Severity : enhancement
991 Description: Lustre message v2
992 Details : Add lustre message format v2.
994 Severity : enhancement
996 Description: client OST exclusion list
997 Details : Clients can be started with a list of OSTs that should be
998 declared "inactive" for known non-responsive OSTs.
1002 Description: SPEC SFS validation failure on NFS v2 over lustre.
1003 Details : Changes the blocksize for regular files to be 2x RPC size,
1004 and not depend on stripe size.
1006 Severity : enhancement
1008 Description: Multiple MD RPCs in flight.
1009 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1010 To avoid overly-overloading MDS, introduce a limit on number of
1011 MDS RPCs in flight for a single client and add /proc controls
1012 to adjust this limit.
1014 Severity : enhancement
1016 Description: client read/write statistics
1017 Details : Add client read/write call usage stats for performance
1018 analysis of user processes.
1019 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1020 file access. extents_stats shows chunk size distribution.
1021 extents_stats_per_process show chunk size distribution per
1024 Severity : enhancement
1026 Description: mds statistics
1027 Details : Add detailed mds operations statistics in
1028 /proc/fs/lustre/mds/*/stats.
1032 Description: Failure of copying files with lustre special EAs.
1033 Details : Client side always return success for setxattr call for lustre
1034 special xattr (currently only "trusted.lov").
1039 Description: Refcount LNET uuids
1040 Details : The global LNET uuid list grew linearly with every startup;
1041 refcount repeated list entries instead of always adding to
1044 Severity : enhancement
1046 Description: Dynamic service threads
1047 Details : Within a small range, start extra service threads
1048 automatically when the request queue builds up.
1051 Frequency : mixed-endian client/server environments
1053 Description: mixed-endian crashes
1054 Details : The new msg_v2 system had some failures in mixed-endian
1057 Severity : enhancement
1059 Description: Easy OST removal
1060 Details : OSTs can be permanently deactivated with e.g. 'lctl
1061 conf_param lustre-OST0001.osc.active=0'
1063 Severity : enhancement
1065 Description: MGS proc entries
1066 Details : Added basic proc entries for the MGS showing what filesystems
1069 Severity : enhancement
1071 Description: provide MGS failover
1072 Details : Added config lock reacquisition after MGS server failover.
1074 Severity : enhancement
1076 Description: add Linux 2.4 support
1077 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1081 Description: a large application tries to do I/O to the same resource and dies
1082 in the middle of it.
1083 Details : Check the req->rq_arrival time after the call to
1084 ost_brw_lock_get(), but before we do anything about
1085 processing it & sending the BULK transfer request. This
1086 should help move old stale pending locks off the queue as
1087 quickly as obd_timeout.
1090 Frequency : when an incorrect nid is specified during startup
1092 Description: ptlrpc connect to non-existant node causes kernel crash
1093 Details : LNET can't be re-entered from an event callback, which
1094 happened when we expire a message after the export has been
1095 cleaned up. Instead, hand the zombie cleanup off to another
1098 Severity : enhancement
1100 Description: plain/inodebits lock performance improvement
1101 Details : Grouping plain/inodebits in granted list by their request modes
1102 and bits policy, thus improving the performance of search through
1106 Frequency : only if OST filesystem is corrupted
1108 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1109 Details : for a short time RPCs with bulk IO are in the replay list,
1110 but replay of bulk IOs is unimplemented. If the OST filesystem
1111 is corrupted due to disk cache incoherency and then replay is
1112 started it is possible to trip an assertion. Avoid putting
1113 committed RPCs into the replay list at all to avoid this issue.
1116 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1118 Description: System hang on startup
1119 Details : This bug allowed the liblustre (e.g. catamount) client to
1120 return to the app before handling all startup RPCs. This
1121 could leave the node unresponsive to lustre network traffic
1122 and manifested as a server ptllnd timeout.
1124 Severity : enhancement
1126 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1127 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1128 boolean to control whether to print peer debug info when a
1129 client's RPC times out.
1132 Frequency : only for kernels with patches from Lustre below 1.4.3
1134 Description: Remove old rdonly API
1135 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1138 Frequency : only for devices with external journals
1140 Description: Set external device read-only also
1141 Details : During a commanded failover stop, we set the disk device
1142 read-only while the server shuts down. We now also set any
1143 external journal device read-only at the same time.
1146 Frequency : when upgrading from 1.4 while trying to change parameters
1148 Description: The wrong (new) MDC name was used when setting parameters for
1149 upgraded MDT's. Also allows changing of OSC (and MDC)
1150 parameters if --writeconf is specified at tunefs upgrade time.
1153 Frequency : when setting specific ost indicies
1155 Description: QOS code breaks on skipped indicies
1156 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1157 created with --index need not be sequential.
1159 Severity : enhancement
1161 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1162 Details : The uninit_groups feature works in conjunction with the kernel
1163 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1164 up the pass1 processing of e2fsck. This is a read-only feature
1165 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1166 mount filesystems that have had this feature enabled.
1168 Severity : enhancement
1170 Description: Improve multi-block allocation algorithm to avoid fragmentation
1171 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1172 allocation locality and avoid filesystem fragmentation.
1174 ------------------------------------------------------------------------------
1176 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1178 * Support for kernels:
1179 2.4.21-47.0.1.EL (RHEL 3)
1180 2.6.5-7.283 (SLES 9)
1181 2.6.9-42.0.10.EL (RHEL 4)
1182 2.6.12.6 vanilla (kernel.org)
1183 2.6.16.27-0.9 (SLES 10)
1184 * Recommended e2fsprogs version: 1.39.cfs5
1186 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1190 Frequency : occasional, depends on client load and configuration
1191 Bugzilla : 12181, 12203
1192 Description: data loss for recently-modified files
1194 Details : In some cases it is possible that recently written or created
1195 files may not be written to disk in a timely manner (this should
1196 normally be within 30s unless client IO load is very high).
1197 The problem appears as zero-length files or files that are a
1198 multiple of 1MB in size after a client crash or client eviction
1199 that are missing data at the end of the file.
1201 This problem is more likely to be hit on clients where files are
1202 repeatedly created and unlinked in the same directory, clients
1203 have a large amount of RAM, have many CPUs, the filesystem has
1204 many OSTs, the clients are rebooted frequently, and/or the files
1205 are not accessed by other nodes after being written.
1207 The presence of the problem can be detected by looking at
1208 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1209 smaller than the second (nr_unused) then dirty files will not
1210 be flushed automatically to disk. "sync; sleep 10" should be
1211 run several times on the node before unmounting it to update
1212 Lustre (this is also safe to run on nodes without this problem).
1214 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1215 that can cause this same problem, so customers using that kernel
1216 also need to update the kernel in addition to Lustre. In order
1217 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1219 It is normal that files written just before a client crash (less
1220 than 30s) may not yet have been flushed to disk, even for local
1224 Frequency : frequent on thin XT3 nodes
1226 Description: UUID collision on thin XT3 Linux nodes
1227 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1228 randomly, since we relied on an insufficiently-seeded PRNG.
1233 Description: OSS hangs after "All ost request buffers busy"
1234 Details : A deadlock between quota and journal operations caused OSS
1235 hangs after printing "All ost request buffers busy."
1238 Frequency : always on liblustre builds
1240 Description: Cleanup compiler warnings on liblustre
1243 Frequency : always on liblustre builds on XT3
1245 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1249 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1250 > 32000 subdirectories
1251 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1252 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1253 subdirectory count crosses 32000. This will aid e2fsck to
1254 correctly handle more than 32000 subdirectories.
1257 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1259 Description: System hang on startup
1260 Details : This bug allowed the liblustre (e.g. catamount) client to
1261 return to the app before handling all startup RPCs. This
1262 could leave the node unresponsive to lustre network traffic
1263 and manifested as a server ptllnd timeout.
1265 Severity : enhancement
1267 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1268 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1269 boolean to control whether to print peer debug info when a
1270 client's RPC times out.
1275 Description: make O_SYNC working on 2.6 kernels
1276 Details : 2.6 kernels use different method for mark pages for write,
1277 so need add a code to lustre for O_SYNC work.
1282 Description: Failure to close file and release space on NFS
1283 Details : Put inode details into lock acquired in ll_intent_file_open.
1284 Use mdc_intent_lock in ll_intent_open to properly
1285 detect all kind of errors unhandled by mdc_enqueue.
1290 Description: proc file read during shutdown sometimes raced obd removal,
1292 Details : Add lock to prevent obd access after proc file removal.
1295 Frequency : Only for files larger than 4GB on 32-bit clients.
1297 Description: improperly doing page alignment of locks
1298 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1299 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1304 Description: RPCs being resent when they shouldn't be
1305 Details : Some RPCs that should not be resent are being resent. This
1306 can cause inconsistencies in the RPC state machine. Do not
1307 resend such requests.
1310 Frequency : rare, only with NFS export
1312 Description: Crash on NFS re-export node
1313 Details : under very unusual load conditions an assertion is hit in
1314 ll_intent_file_open()
1317 Frequency : only if OST filesystem is corrupted
1319 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1320 Details : for a short time RPCs with bulk IO are in the replay list,
1321 but replay of bulk IOs is unimplemented. If the OST filesystem
1322 is corrupted due to disk cache incoherency and then replay is
1323 started it is possible to trip an assertion. Avoid putting
1324 committed RPCs into the replay list at all to avoid this issue.
1329 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1330 Details : Large single O_DIRECT read and write calls can fail to allocate
1331 a sufficiently large buffer to process the request. In case of
1332 allocation failure the allocation is retried with a smaller
1333 buffer and broken into smaller requests.
1335 Severity : enhancement
1337 Description: Add -o localflock option to simulate old noflock behaviour.
1338 Details : This will achieve local-only flock/fcntl locks coherentness.
1343 Description: versioning check is incomplete
1344 Details : Checking the version difference of client vs. server, report
1345 error if the gap is too big.
1350 Description: add support PG_writeback bit
1351 Details : add support for PG_writeback bit for Lustre, for more carefull
1352 work with page cache in 2.6 kernel. This also fix some deadlocks
1353 and remove hack for work O_SYNC with 2.6 kernel.
1355 Severity : enhancement
1357 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1358 Details : The uninit_groups feature works in conjunction with the kernel
1359 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1360 up the pass1 processing of e2fsck. This is a read-only feature
1361 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1362 mount filesystems that have had this feature enabled.
1364 Severity : enhancement
1366 Description: Improve multi-block allocation algorithm to avoid fragmentation
1367 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1368 allocation locality and avoid filesystem fragmentation.
1370 ------------------------------------------------------------------------------
1372 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1374 * Support for kernels:
1375 2.6.9-42.0.3.EL (RHEL 4)
1376 2.6.5-7.276 (SLES 9)
1377 2.4.21-47.0.1.EL (RHEL 3)
1378 2.6.12.6 vanilla (kernel.org)
1379 2.6.16.21-0.8 (SLES10)
1380 * Recommended e2fsprogs version: 1.39.cfs2-0
1382 * The backwards-compatible /proc/sys/portals symlink has been removed
1383 in this release. Before upgrading, please ensure that you change
1384 any configuration scripts or /etc/sysctl.conf files that access
1385 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1386 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1387 in advance of the upgrade on any system running Lustre 1.4.6 or
1388 newer, since /proc/sys/lnet was added in that version.
1389 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1393 Frequency : only when quota is used
1395 Description: avoid scanning export list for quota master
1396 Details : Change the algorithms to avoid scanning export list in order
1397 to improve the efficiency.
1400 Frequency : MDS failover only, very rarely
1402 Description: "went back in time" messages on mds failover
1403 Details : The greatest transno may be lost when the current operation
1404 finishes with an error (transno==0) and the client's last_rcvd
1405 record is over-written. Save the greatest transno in the
1406 mds_last_transno for this case.
1409 Frequency : always for specific kernels and striping counts
1411 Description: client may get "Matching packet too big" without ACL support
1412 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1413 when trying to access files in certain configurations. The
1414 clients should in fact be denied when mounting because they do
1415 not understand ACLs.
1418 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1420 Description: many clients connecting with IO in progress causes connect timeouts
1421 Details : Avoid synchronous journal commits to avoid delays caused by many
1422 clients connecting/disconnecting when bulk IO is in progress.
1423 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1424 OST_IO_PORTAL to avoid delays behind potentially many pending
1428 Frequency : occasionally with multiple writers to a single file
1430 Description: shared writes to file may result in wrong size reported by stat()
1431 Details : Allow growing of kms when extent lock is cancelled
1434 Frequency : always with random mmap IO to multi-striped file
1436 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1437 Details : Only if the hole is at the end of OST object so that kms is too
1438 small. Fix is to increase kms accordingly in ll_nopage.
1441 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1443 Description: writes to a missing object would leak memory on the OST
1444 Details : If there is an inconsistency between the MDS and OST filesystems,
1445 such that the MDS references an object that doesn't exist, writes
1446 to that object will leak memory due to incorrect cleanup in the
1447 error handling path, eventually running out of memory on the OST.
1452 Description: Creating too long symlink causes lustre errors
1453 Details : Check symlink and name lengths before sending requests to MDS.
1456 Frequency : only if flock is enabled (not on by default)
1458 Description: posix locks not released on fd closure on 2.6.9+
1459 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1460 kernels, this caused such locks not to be released on fd close and
1461 then assertions on fs unmount about still used locks.
1464 Frequency : MDS failover only, very rarely
1466 Description: clients may get ASSERTION(granted_lock != NULL)
1467 Details : When request was taking a long time, and a client was resending
1468 a getattr by name lock request. The were multiple lock requests
1469 with the same client lock handle and
1470 mds_getattr_name->fixup_handle_for_resent_request found one of the
1471 lock handles but later failed with ASSERTION(granted_lock != NULL).
1476 Description: handle->h_buffer_credits > 0, assertion failure
1477 Details : h_buffer_credits is zero after truncate, causing assertion
1478 failure. This patch extends the transaction or creates a new
1482 Frequency : NFS re-export or patchless client
1483 Bugzilla : 11179, 10796
1484 Description: Crash on NFS re-export node (__d_move)
1485 Details : We do not want to hash the dentry if we don't have a lock.
1486 But if this dentry is later used in d_move, we'd hit uninitialised
1487 list head d_hash, so we just do this to init d_hash field but
1488 leave dentry unhashed.
1491 Frequency : NFS re-export or patchless client
1493 Description: NFS exports has problem with symbolic link
1494 Details : lustre client didn't properly install dentry when re-exported
1495 to NFS or running patchless client.
1498 Frequency : NFS re-export or patchless client
1500 Description: Various nfs/patchless fixes.
1501 Details : fixes reuse disconected alias for lookup process - this fixes
1502 warning "find_exported_dentry: npd != pd",
1503 fix permission error with open files at nfs.
1504 fix apply umask when do revalidate.
1507 Frequency : occasional
1509 Description: Crash on NFS re-export node
1510 Details : calling clear_page() on the wrong pointer triggered oops in
1511 generic_mapping_read().
1514 Frequency : rarely, using O_DIRECT IO
1516 Description: unaligned directio crashes client with LASSERT
1517 Details : check for unaligned buffers before trying any requests.
1520 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1522 Description: stale data returned from RAID cache
1523 Details : If only a small amount of IO is done to the RAID device before
1524 reading it again it is possible to get stale data from the RAID
1525 cache instead of reading it from disk.
1528 Frequency : always for sles10 kernel
1530 Description: sles10 support
1531 Details : ll_follow_link: compile fixes and using of nd_set_link
1532 under newer kernels.
1535 Frequency : depends on arch, kernel and compiler version, always on sles10
1538 Description: recursive or deep enough symlinks cause stack overflow
1539 Details : getting rid of large stack-allocated variable in
1543 Frequency : depends on hardware
1545 Description: lustre write performance loss in the SLES10 kernel
1546 Details : the performance loss is caused by using of write barriers in the
1547 ext3 code. The SLES10 kernel turns barrier support on by
1548 default. The fix is to undo that change for ldiskfs.
1550 ------------------------------------------------------------------------------
1552 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1554 * Support for kernels:
1555 2.6.9-42.0.3EL (RHEL 4)
1556 2.6.5-7.276 (SLES 9)
1557 2.4.21-47.0.1.EL (RHEL 3)
1558 2.6.12.6 vanilla (kernel.org)
1562 Frequency : quota enabled and large files being deleted
1564 Description: releasing more than 4GB of quota at once hangs OST
1565 Details : If a user deletes more than 4GB of files on a single OST it
1566 will cause the OST to spin in an infinite loop. Release
1567 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1572 Description: statfs data retrieved from /proc may be stale or zero
1573 Details : When reading per-device statfs data from /proc, in the
1574 {kbytes,files}_{total,free,avail} files, it may appear
1575 as zero or be out of date.
1578 Frequency : systems with MD RAID1 external journal devices
1580 Description: lconf's call to blkid is confused by RAID1 journal devices
1581 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1582 of returning all block devices that match the journal UUID.
1585 Frequency : always, for aggregate stripe size over 4GB
1587 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1588 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1589 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1592 Frequency : always if "lfs find" used on a local file/directory
1594 Description: "lfs find" segfaults if used on a local file/directory
1595 Details : The case where a directory component was not specified wasn't
1596 handled correctly. Handle this properly.
1599 Frequency : always on ppc64
1601 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1602 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1603 perform endianness conversion.
1606 Frequency : rarely (truncate to non-zero file size after write under load)
1607 Bugzilla : 10730, 10687
1608 Description: Files padded with zeros to next 4K multiple
1609 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1610 it is possible that files that are truncated to a non-zero size
1611 immediately after being written are filled with zero bytes beyond
1612 the truncated size. No file data is lost.
1614 Severity : enhancement
1616 Description: Allow recovery/failover for liblustre clients.
1617 Details : liblustre clients were unaware of failover configurations until
1620 Severity : enhancement
1622 Description: user file locks should fail when not mounting with flock option
1623 Details : Set up an error-returning stub in ll_file_operations.lock field
1624 to prevent incorrect behaviour when client is mounted without
1625 flock option. Also, set up properly f_op->flock field for
1629 Frequency : always on ia64
1631 Description: "lfs df" loops on printing out MDS statfs information
1632 Details : The obd_ioctl_data was not initialized and in some systems
1633 this caused a failure during the ioctl that did not return
1634 an error. Initialize the struct and return an error on failure.
1637 Frequency : SLES 9 only
1639 Description: Error of copying files with lustre special EAs as root
1640 Details : Client side always return success for setxattr call for lustre
1641 special xattr (currently only "trusted.lov").
1644 Frequency : rarely on clusters with both ia64+i386 clients
1646 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1647 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1648 clients are doing concurrent writes to the same file it is
1649 possible that the ia64 clients may LASSERT because the OST
1650 extent locks are not PAGE_SIZE aligned. Ensure that grown
1651 locks are always aligned on the request boundary.
1654 Frequency : specific use, occasional
1656 Description: Overwriting in use executable truncates on-disk binary image
1657 Details : If one node attempts to overwrite an executable in use by
1658 another node, we now correctly return ETXTBSY instead of
1659 truncating the file.
1661 Severity : enhancement
1663 Description: Async OSC create to avoid the blocking unnecessarily.
1664 Details : If a OST has no remain object, system will block on the creating
1665 when need to create a new object on this OST. Now, ways use
1666 pre-created objects when available, instead of blocking on an
1667 empty osc while others are not empty. If we must block, we block
1668 for the shortest possible period of time.
1673 Description: chmod on Lustre root is propagated to other clients
1674 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1675 invalid by the follow_mount time.
1678 Frequency : liblustre clients only
1680 Description: Race in 'instant cancel' lock handling could lead to such locks
1681 never to be granted in case of SMP MDS
1682 Details : Do not destroy not yet granted but cbpending locks in
1686 Frequency : replay/resend of open
1688 Description: non null lock assetion failure in mds_intent_policy
1689 Details : Trying to replay/resend lockless open requests resulted in
1690 mds_open() returning 0 with no lock. Now it sets a flag if
1691 a lock is going to be returned.
1693 Severity : enhancement
1695 Description: Checksum enhancements
1696 Details : New checksum enhancements allow for resending RPCs that failed
1699 Severity : enhancement
1701 Description: Tunables on number of dirty pages in cacche
1702 Details : Allow to set limit on number of dirty pages cached.
1707 Description: client crash on unmount - lock still has references
1708 Details : In some error handling cases it was possible to leak a lock
1709 reference on a client while accessing a file. This was not
1710 harmful to the client during operation, but would cause the
1711 client to crash when the filesystem is unmounted.
1714 Frequency : specific case, rare
1716 Description: ETXTBSY on mds though file not in use by client
1717 Details : ETXTBSY is no longer incorrectly returned when attempting to
1718 chmod or chown a directory that the user previously tried to
1719 execute or a currently-executing binary.
1722 Frequency : extremely rare except on liblustre-based clients
1724 Description: Lustre space not freed when files are deleted
1725 Details : Clean up open-unlinked files after client eviction. Previously
1726 the unlink was skipped and the files remained as orphans.
1731 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1732 Details : In some cases it was possible to send a blocking callback to
1733 a client doing a glimpse, even though that client didn't get
1734 a lock granted. When the glimpse lock is cancelled on the OST
1735 the freed lock is left on the waiting list and corrupted the list.
1738 Frequency : all core dumps
1740 Description: Broke core dumps to lustre
1741 Details : Negative dentry may be unhashed if parent does not have UPDATE
1742 lock, but some callers, e.g. do_coredump, expect dentry to be
1743 hashed after successful create, hash it in ll_create_it.
1745 ------------------------------------------------------------------------------
1747 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1749 * Support for kernels:
1750 2.6.9-42.0.2.EL (RHEL 4)
1751 2.6.5-7.276 (SLES 9)
1752 2.4.21-40.EL (RHEL 3)
1753 2.6.12.6 vanilla (kernel.org)
1757 Frequency : always on RHEL 3
1759 Description: Number of open files grows over time
1760 Details : The number of open files grows over time, whether or not
1761 Lustre is started. This was due to a filp leak introduced
1762 by one of our kernel patches.
1764 ------------------------------------------------------------------------------
1766 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1768 * Support for kernels:
1769 2.6.9-42.EL (RHEL 4)
1770 2.6.5-7.276 (SLES 9)
1771 2.4.21-40.EL (RHEL 3)
1772 2.6.12.6 vanilla (kernel.org)
1777 Bugzilla : 5719, 9635, 9792, 9684
1778 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1779 Details : If a server is under heavy load and cannot reply to new
1780 connection requests before the client resends the (re)connect,
1781 the connection handling code can behave badly if two service
1782 threads are concurrently handing separate (re)connections from
1783 the same client. Add better locking to the connection handling
1784 code, and ensure that only a single connection will be processed
1785 for a given client UUID, even if the lock is dropped.
1787 Severity : enhancement
1789 Description: add TCP zero-copy support to kernel
1790 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1791 sends if the hardware supports scatter-gather and checksumming.
1792 This allows socklnd to do client-write and server-read more
1793 efficiently and reduce CPU utilization from skbuf copying.
1796 Frequency : only if NFS exporting from client
1798 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1799 Details : If setting attributes on a file created under NFS that had
1800 never been opened it would be possible to oops the client
1801 if the file had no objects.
1804 Frequency : always for liblustre
1806 Description: liblustre client does MDS+OSTs setattr RPC for each write
1807 Details : When doing a write from a liblustre client, the client
1808 incorrectly issued an RPC to the MDS and each OST the file was
1809 striped over in order to update the timestamps. When writing
1810 with small chunks and many clients this could overwhelm the MDS
1811 with RPCs. In all cases it would slow down the write because
1812 these RPCs are unnecessary.
1814 Severity : enhancement
1816 Description: allow number of MDS service threads to be changed at module load
1817 Details : It is now possible to change the number of MDS service threads
1818 running. Adding "options mds mds_num_threads={N}" to the MDS's
1819 /etc/modprobe.conf will set the number of threads for the next
1820 time Lustre is restarted (assuming the "mds" module is also
1821 reloaded at that time). The default number of threads will
1822 stay the same, 32 for most systems.
1827 Description: OST crash if filesystem is unformatted or corrupt
1828 Details : If an OST is started on a device that has never been formatted
1829 or if the filesystem is corrupt and cannot even mount then the
1830 error handling cleanup routines would dereference a NULL pointer.
1835 Description: NULL pointer deref in llap_from_page.
1836 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1837 incorrect) mapping if the page was truncated/reclaimed while it was
1838 searched for. Check for this condition and skip such pages when
1839 doing readahead. Introduce extra check to llap_from_page() to
1840 verify page->mapping->host is non-NULL (so page is not anonymous).
1843 Frequency : Sometimes when using sys_sendfile
1845 Description: "page not covered by a lock" warnings from ll_readpage
1846 Details : sendfile called ll_readpage without right page locks present.
1847 Now we introduced ll_file_sendfile that does necessary locking
1848 around call to generic_file_sendfile() much like we do in
1852 Frequency : with certain MDS communication failures at client mount time
1854 Description: NULL pointer deref after failed client mount
1855 Details : a client connection request may delayed by the network layer
1856 and not be sent until after the PTLRPC layer has timed out the
1857 request. If the client fails the mount immediately it will try
1858 to clean up before the network times out the request. Add a
1859 reference from the request import to the obd device and delay
1860 the cleanup until the network drops the request.
1863 Frequency : occasionally during client (re)connect
1865 Description: assertion failure during client (re)connect
1866 Details : processing a client connection request may be delayed by the
1867 client or server longer than the client connect timeout. This
1868 causes the client to resend the connection request. If the
1869 original connection request is replied in this interval, the
1870 client may trip an assertion failure in ptlrpc_connect_interpret()
1871 which thought it would be the only running connect process.
1874 Frequency : only with obd_echo servers and clients that are rebooted
1876 Description: kernel BUG accessing uninitialized data structure
1877 Details : When running an obd_echo server it did not start the ping_evictor
1878 thread, and when a client was evicted an uninitialized data
1879 structure was accessed. Start the ping_evictor in the RPC
1880 service startup instead of the OBD startup.
1882 Severity : enhancement
1883 Bugzilla : 10193 (patchless)
1884 Description: Remove dependency on various unexported kernel interfaces.
1885 Details : No longer need reparent_to_init, exit_mm, exit_files,
1886 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1889 Frequency : rare (only users of deprecated and unsupported LDAP config)
1891 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1892 Details : LDAP apparently contains 'lustreName' attributes instead of
1893 'name'. A simple remapping of the name is sufficient.
1896 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1898 Description: waiting_locks_callback trips kernel BUG if client is evicted
1899 Details : Running with the dump_on_timeout debug flag turned on makes
1900 it possible that the waiting_locks_callback() can try to dump
1901 the Lustre kernel debug logs from an interrupt handler. Defer
1902 this log dumping to the expired_lock_main() thread.
1904 Severity : enhancement
1906 Description: Support NFS exporting on 2.6 kernels.
1907 Details : Implement non-rawops metadata methods for NFS server to use without
1908 changing NFS server code.
1911 Frequency : very rare (synthetic metadata workload only)
1913 Description: two racing renames might cause an MDS thread to deadlock
1914 Details : Running the "racer" program may cause one MDS thread to rename
1915 a file from being the source of a rename to being the target of
1916 a rename at exactly the same time that another thread is doing
1917 so, and the second thread has already enqueued these locks after
1918 doing a lookup of the target and is trying to relock them in
1919 order. Ensure that we don't try to re-lock the same resource.
1922 Frequency : only very large systems with liblustre clients
1924 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1925 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1926 all OSTs in parallel. This allows the eviction of stale liblustre
1927 clients to proceed much faster than if they were done in series,
1928 and also offers similar improvements for other set_info RPCs.
1933 Description: excessive CPU usage during initial read phase on client
1934 Details : During the initial read phase on a client, it would agressively
1935 retry readahead on the file, consuming too much CPU and impacting
1936 performance (since 1.4.5.8). Improve the readahead algorithm
1937 to avoid this, and also improve some other common cases (read
1938 of small files in particular, where "small" is files smaller than
1939 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1944 Description: MDS crash when receiving packet with unknown intent.
1945 Details : Do not LBUG in unknown intent case, just return -EFAULT
1947 Severity : enhancement
1948 Bugzilla : 9293, 9385
1949 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1950 Details : Do not serialize getattr (non-intent version) and statfs.
1953 Frequency : occasional, when OST network is overloaded/intermittent
1955 Description: client evicted by OST after bulk IO timeout
1956 Details : If a client sends a bulk IO request (read or write) the OST
1957 may evict the client if it is unresposive to its data GET/PUT
1958 request. This is incorrect if the network is overloaded (takes
1959 too long to transfer the RPC data) or dropped the OST GET/PUT
1960 request. There is no need to evict the client at all, since
1961 the pinger and/or lock callbacks will handle this, and the
1962 client can restart the bulk request.
1965 Frequency : Always when mmapping file with no objects
1967 Description: client crashes when mmapping file with no objects
1968 Details : Check that we actually have objects in a file before doing any
1969 operations on objects in ll_vm_open, ll_vm_close and
1975 Description: Request leak when working with deleted CWD
1976 Details : Introduce advanced request refcount tracking for requests
1977 referenced from lustre intent.
1979 Severity : Enhancement
1981 Description: Cache open file handles on client.
1982 Details : MDS now will return special lock along with openhandle, if
1983 requested and client is allowed to hold openhandle, even if unused,
1984 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1985 closing files for every read/write openration.
1987 Severity : Enhancement
1989 Description: Cache open negative dentries on client when possible.
1990 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1991 negative dentries on lock revocation.
1996 Description: Remounting a client read-only wasn't possible with a zconf mount
1997 Details : It wasn't possible to remount a client read-only with llmount.
1999 Severity : enhancement
2000 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2001 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2002 MPICH is included to add Lustre-specific ADIO interfaces.
2003 This is based closely on the UFS ADIO layer and only differs
2004 in file creation, in order to allow the OST striping to be set.
2005 This is user-contributed code and not supported by CFS.
2010 Description: extended inode attributes (immutable, append-only) work improperly
2011 when 2.4 and 2.6 kernels are used on client/server or vice versa
2012 Details : Introduce kernel-independent values for these flags.
2014 Severity : enhancement
2017 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2018 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2019 granularity. Now, Fractional MB granularity is be supported,
2020 this is very useful for low memory system.
2022 Severity : enhancement
2024 Description: Getattr by fid
2025 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2026 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2027 have locks and avoids getattr rpc requests for it on every stat.
2030 Frequency : Always, for filesystems larger than 2TB
2032 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2033 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2034 filesystems larger than 2TB to be mounted with mballoc enabled.
2037 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2039 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2040 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2041 filesystems larger than 2TB is mounted then the kernel will
2042 silently corrupt the start of the filesystem. CONFIG_LBD is
2043 enabled for all CFS-supported kernels, but the possibility of
2044 this happening with a modified kernel config exists.
2046 Severity : enhancement
2048 Description: add client O_DIRECT support for 2.6 kernels
2049 Details : It is now possible to do O_DIRECT reads and writes to files
2050 in the Lustre client mountpoint on 2.6 kernel clients.
2052 Severity : enhancement
2054 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2055 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2056 parallel, not waiting for response from every OST before sending
2057 a rpc to the next OST.
2062 Description: setattr vs write race when updating file timestamps
2063 Details : Client processes that update a file timestamp into the past
2064 right after writing to the file (e.g. tar) it is possible that
2065 the updated file modification time can be reset to the current
2066 time due to a race between processing the setattr and write RPC.
2068 Severity : enhancement
2070 Description: Bring 'lfs find' closer in line with regular Linux find.
2071 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2072 -print0 options and obtains all the needed info through the lustre
2075 Severity : enhancement
2077 Description: support up to 1024 configured devices on one node
2078 Details : change obd_dev array from statically allocated to dynamically
2079 allocated structs as they are first used to reduce memory usage
2084 Description: Flush dirty partially truncated pages during truncate
2085 Details : Immediatelly flush partially truncated pages in filter_setattr,
2086 this way we completely avoid having any pages in page cache on OST
2087 and can retire ugly workarounds during writes to flush such pages.
2092 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2093 Details : protect inode from truncation within vfs_unlink() context
2094 just take a reference before calling vfs_unlink() and release it
2095 when parent's i_sem is free.
2100 Description: last_id value checked outside lock on OST caused LASSERT failure
2101 Details : If there were multiple MDS->OST object precreate requests in
2102 flight, it was possible that the OST's last object id was checked
2103 outside a lock and incorrectly tripped an assertion. Move checks
2104 inside locks, and discard old precreate requests.
2107 Frequency : always, if extents are used on OSTs
2109 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2110 Details : OSTs using the extents format would not zero the high 16 bits of
2111 the index physical block number. This is not a problem for any
2112 OST filesystems smaller than 16TB, and no kernels support ext3
2113 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2114 new/modified files) and can be fixed for existing filesystems
2115 with e2fsprogs-1.39-cfs1.
2120 Description: import connection selection may be incorrect if timer wraps
2121 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2122 import connections to be ignored if the 32-bit jiffies counter
2123 wraps. Use a 64-bit jiffies counter.
2126 Frequency : very large clusters immediately after boot
2128 Description: LNET request buffers exhausted under heavy short-term load
2129 Details : If a large number of client requests are generated on a service
2130 that has previously never seen so many requests it is possible
2131 that the request buffer growth cannot keep up with the spike in
2132 demand. Instead of dropping incoming requests, they are held in
2133 the LND until the RPC service can accept more requests.
2136 Frequency : Sometimes during replay
2138 Description: Assertion failure in ll_local_open after replay.
2139 Details : If replay happened on an open request reply before we were able
2140 to set replay handler, reply will become not swabbed tripping the
2141 assertion in ll_local_open. Now we set the handler right after
2142 recognising of open request
2145 Frequency : very rare
2147 Description: kernel reports "badness in vsnprintf"
2148 Details : Reading from the "recovery_status" /proc file in small chunks
2149 may cause a negative length in lprocfs_obd_rd_recovery_status()
2150 call to vsnprintf() (which is otherwise harmless). Exit early
2151 if there is no more space in the output buffer.
2153 Severity : enhancement
2155 Description: clear OBD RPC statistics by writing to them
2156 Details : It is now possible to clear the OBD RPC statistics by writing
2157 to the "stats" file.
2162 Description: Client mtime is not the same on different clients after utimes
2163 Details : In some cases, the client was using the utimes() syscall on
2164 a file cached on another node. The clients now validate the
2165 ctime from the MDS + OSTs to determine which one is right.
2170 Description: Inability to activate failout mode
2171 Details : lconf script incorrectly assumed that in pythong string's numeric
2172 value is used in comparisons.
2175 Frequency : always with multiple stripes per file
2177 Description: Inefficient object allocation for mutli-stripe files
2178 Details : When selecting which OSTs to stripe files over, for files with
2179 a stripe count that divides evenly into the number of OSTs,
2180 the MDS is always picking the same starting OST for each file.
2181 Return the OST selection heuristic to the original design.
2186 Description: mount failures may take full timeout to return an error
2187 Details : Under some heavy load conditions it is possible that a
2188 failed mount can wait for the full obd_timeout interval,
2189 possibly several minutes, before reporting an error.
2190 Instead return an error as soon as the status is known.
2192 Frequency : quota enabled and large files being deleted
2194 Description: releasing more than 4GB of quota at once hangs OST
2195 Details : If a user deletes more than 4GB of files on a single OST it
2196 will cause the OST to spin in an infinite loop. Release
2197 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2202 Description: statfs data retrieved from /proc may be stale or zero
2203 Details : When reading per-device statfs data from /proc, in the
2204 {kbytes,files}_{total,free,avail} files, it may appear
2205 as zero or be out of date.
2208 Frequency : systems with MD RAID1 external journal devices
2210 Description: lconf's call to blkid is confused by RAID1 journal devices
2211 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2212 of returning all block devices that match the journal UUID.
2215 Frequency : always, for aggregate stripe size over 4GB
2217 Description: assertion fails when trying to use 4GB stripe size
2218 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2219 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2222 Frequency : always on ppc64
2224 Description: the first write on an ext3 filesystem with mballoc got stuck
2225 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2226 perform endianness conversion.
2228 ------------------------------------------------------------------------------
2230 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2232 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2233 INTEROPERATE with older versions automatically. Please read the
2234 user documentation before upgrading any part of a live system.
2235 * WARNING: Lustre networking configuration changes are required with
2236 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2239 * Support for kernels:
2240 2.6.9-22.0.2.EL (RHEL 4)
2241 2.6.5-7.244 (SLES 9)
2242 2.6.12.6 vanilla (kernel.org)
2245 Severity : enhancement
2246 Bugzilla : 7981/8208
2247 Description: Introduced Lustre Networking (LNET)
2248 Details : LNET is new networking infrastructure for Lustre, it includes
2249 a reorganized network configuration mode (see the user
2250 documentation for full details) as well as support for routing
2251 between different network fabrics. Lustre Networking Devices
2252 (LNDs) for the supported network fabrics have also been
2253 created for this new infrastructure.
2255 Severity : enhancement
2256 Description: Introduced Access control lists
2257 Details : clients can set ACLs on files and directories in order to have
2258 more fine-grained permissions than the standard Unix UGO+RWX.
2259 The MDS must be started with the "-o acl" mount option.
2261 Severity : enhancement
2262 Description: Introduced filesystem quotas
2263 Details : Administrators may now establish per-user quotas on the
2266 Severity : enhancement
2268 Description: Configuration change for the XT3
2269 The PTLLND is now used to run Lustre over Portals on the XT3
2270 The configure option(s) --with-cray-portals are no longer used.
2271 Rather --with-portals=<path-to-portals-includes> is used to
2272 enable building on the XT3. In addition to enable XT3 specific
2273 features the option --enable-cray-xt3 must be used.
2278 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2279 Details : A race between allocating a new llcd and re-getting the llcd_lock
2280 allowed another thread to grab newly-allocated llcd.
2282 Severity : enhancement
2284 Description: 2.6 OST async journal commit and locking fix to improve performance
2285 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2286 2.6 kernels are now async as they already were in 2.4 kernels so
2287 that they can commit concurrently with the network bulk transfer.
2288 For block-allocated files the filter allocation semaphore is held
2289 to avoid filesystem fragmentation during allocation. BKL lock
2290 removed for 2.6 xattr operations where it is no longer needed.
2295 Description: lconf incorrectly determined whether two IP networks could talk
2296 Details : In some more complicated routing and multiple-network
2297 configurations, lconf will avoid trying to make a network
2298 connection to a disjoint part of the IP space. It was doing the
2299 math incorrectly for one set of cases.
2304 Description: Fix for potential infinite loop processing records in an llog.
2305 Details : If an llog record is corrupted/zeroed, it is possible to loop
2306 forever in llog_process(). Validate the llog record length
2307 and skip the remainder of the block on error.
2310 Frequency : occasional (liblustre only)
2312 Description: liblustre could not open files whose last component is a symlink
2313 Details : sysio_path_walk() would incorrectly pass the open intent to
2314 intermediate path components.
2317 Frequency : rare (liblustre only with non-standard tuning)
2318 Bugzilla : 7201 (7350)
2319 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2320 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2321 after it has been released, only do this for referenced locks
2323 Severity : enhancement
2325 Description: specifying an (invalid) directory default stripe_size of -1
2326 would reset the directory default striping
2327 Details : stripe_size -1 was used internally to signal directory stripe
2328 removal, now use "all default" to signal dir stripe removal
2329 as a directory striping of "all default" is not useful
2332 Frequency : common for large clusters running liblustre clients
2334 Description: doing an ls when liblustre clients are running is slow
2335 Details : sending a glimpse AST to a liblustre client waits for every AST
2336 to time out, as liblustre clients will not respond. Since they
2337 cannot cache data we refresh the OST lock LVB from disk instead.
2339 Severity : enhancement
2341 Description: doing an ls at the same time as file IO can be slow
2342 Details : enqueue and other "small" requests can be blocked behind many
2343 large IO requests. Create a new OST IO portal for non-IO
2344 requests so they can be processed faster.
2347 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2349 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2350 created file with mode 000
2351 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2352 allow an ftruncate() as a non-root user to a file with mode 000.
2353 Lustre now respects this flag to disable mode checks when
2354 truncating a file owned by the user
2357 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2360 Description: Revoking locks from clients that went dead or catatonic might take
2362 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2363 cancellation of such locks instant on servers without waiting for
2364 any reply from clients. Clients drops these locks when cancel
2365 notification from server is received without replying.
2368 Frequency : liblustre-only, when liblustre client dies or becomes busy
2370 Description: Doing ls on Linux clients can take a long time with active
2372 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2373 granting such locks to it in the first place if possible. Locks
2374 are taken by proxy on the OST during the read or write and
2375 dropped immediately afterward. Add connect flags handling, do
2376 not grant locks to liblustre clients for glimpse ASTs.
2378 Severity : enhancement
2380 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2381 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2382 sequential reads and behaves badly with applications doing random
2383 reads. Improve it by reading ahead at least read region, and
2384 avoiding excessive large RPC for small reads.
2386 Severity : enhancement
2388 Description: Creating more than 1000 files for a single job may cause a load
2389 imbalance on the OSTs if there are also a large number of OSTs.
2390 Details : qos_prep_create() uses an OST index reseed value that is an
2391 even multiple of the number of available OSTs so that if the
2392 reseed happens in the middle of the object allocation it will
2393 still utilize the OSTs as uniformly as possible.
2398 Description: OST or MDS may oops in ping_evictor_main()
2399 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2400 but doesn't restart at beginning of obd_exports_timed list
2403 Severity : enhancement
2405 Description: improve by-nid export eviction on the MDS and OST
2406 Details : allow multiple exports with the same NID to be evicted at one
2407 time without re-searching the exports list.
2410 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2412 Description: MDS may oops in groups_free()
2413 Details : in rare race conditions a newly allocated group_info struct is
2414 freed again, and this can be NULL. The 2.4 compatibility code
2415 for groups_free() checked for a NULL pointer, but 2.6 did not.
2418 Frequency : common for liblustre clients doing little filesystem IO
2419 Bugzilla : 9352, 7313
2420 Description: server may evict liblustre clients accessing contended locks
2421 Details : if a client is granted a lock or receives a completion AST
2422 with a blocking AST already set it would not reply to the AST
2423 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2426 Frequency : lfs setstripe, only systems with more than 160 OSTs
2428 Description: unable to set striping with a starting offset beyond OST 160
2429 Details : llapi_create_file() incorrectly limited the starting stripe
2430 index to the maximum single-file stripe count.
2433 Frequency : LDAP users only
2435 Description: lconf did not handle in-kernel recovery with LDAP properly
2436 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2438 Severity : enhancement
2440 Description: bind OST threads to NUMA nodes to improve performance
2441 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2442 node and do their allocations there to localize memory access
2444 Severity : enhancement
2446 Description: llmount can determine client NID directly from Myrinet (GM)
2447 Details : the client NID code from gmnalnid was moved directly into
2448 llmount, removing the need to use this or specifying the
2449 client NID explicitly when mounting GM clients with zeroconf
2452 Frequency : if client is started with down MDS
2454 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2455 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2456 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2457 ptlrpc_import_delayed_req() because we didn't check if the
2458 request was interrupted, and we also didn't break out of the
2459 event loop if there was no timeout
2464 Description: data loss during non-page-aligned writes to a single file from
2465 both multiple nodes and multiple threads on one node at same time
2466 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2467 inconsistency led to false short-reads, that were cached and later
2468 used by ->prepare_write() to fill in partially written page,
2469 leading to data loss.
2472 Frequency : always, if lconf --abort_recovery used
2474 Description: lconf --abort_recovery fails with 'Operation not supported'
2475 Details : lconf was attempting to abort recovery on the MDT device and not
2478 Severity : enhancement
2480 Description: remove cleanup logs
2481 Details : replace lconf-generated cleanup logs with lustre internal
2482 cleanup routines. Eliminates the need for client-cleanup and
2485 Severity : enhancement
2487 Description: add support for EAs (user and system) on lustre filesystems
2488 Details : it is now possible to store extended attributes in the Lustre
2489 client filesystem, and with the user_xattr mount option it
2490 is possible to allow users to store EAs on their files also
2492 Severity : enhancement
2494 Description: Add possibility (config option) to show minimal available OST free
2496 Details : When compiled with --enable-mindf configure option, statfs(2)
2497 (and so, df) will return least minimal free space available from
2498 all OSTs as amount of free space on FS, instead of summary of
2499 free spaces of all OSTs.
2501 Severity : enhancement
2503 Description: do not expand extent locks acquired on OST-side
2504 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2505 by server: they are not cached anyway.
2508 Frequency : when mmap is used/binaries executed from Lustre
2510 Description: Unmmap pages before throwing them away from read cache.
2511 Details : llap_shrink cache now attempts to unmap pages before discarding
2512 them (if unmapping failed - do not discard). SLES9 kernel has
2513 extra checks that trigger if this unmapping is not done first.
2518 Description: lconf didn't resolve symlinks before checking to see whether a
2519 given mountpoint was already in use
2522 Frequency : when migrating failover services
2523 Bugzilla : 6395, 9514
2524 Description: When migrating a subset of services from a node (e.g. failback
2525 from a failover service node) the remaining services would
2526 time out and evict clients.
2527 Details : lconf --force (implied by --failover) sets the global obd_timeout
2528 to 5 seconds in order to quickly disconnect, but this caused
2529 other RPCs to time out too quickly. Do not change the global
2530 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2532 Severity : enhancement
2533 Frequency : if MDS is started with down OST
2534 Bugzilla : 9439,5706
2535 Description: Allow startup/shutdown of an MDS without depending on the
2536 availability of the OSTs.
2537 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2538 Add appropriate locking and lov-osc refcounts for safe
2539 cleaning. Add osc abort_inflight calls in case the
2540 synchronize never started.
2543 Frequency : occasional (Cray XT3 only)
2545 Description: root not authorized to access files in CRAY_PORTALS environment
2546 Details : The client process capabilities were not honoured on the MDS in
2547 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2548 been accessed by an authorized user then root was able to access
2549 the file on the local client also. The root user capabilities
2550 are now allowed on the MDS, as this environment has secure UID.
2553 Frequency : occasional
2555 Description: ldiskfs "too long searching" message happens too often
2556 Details : A debugging message (otherwise harmless) prints too often on
2557 the OST console. This has been reduced to only happen when
2558 there are fragmentation problems on the filesystem.
2563 Description: Division by zero in statfs when all OSCs are inactive
2564 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2565 lov_statfs divides by value returned by lov_get_stripecnt().
2569 Bugzilla : 9489, 3273
2570 Description: First write from each client to each OST was only 4kB in size,
2571 to initialize client writeback cache, which caused sub-optimal
2572 RPCs and poor layout on disk for the first writen file.
2573 Details : Clients now request an initial cache grant at (re)connect time
2574 and so that they can start streaming writes to the cache right
2575 away and always do full-sized RPCs if there is enough data.
2576 If the OST is rebooted the client also re-establishes its grant
2577 so that client cached writes will be honoured under the grant.
2582 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2583 Details : Now I/O RPCs go to different portal number and (presumably) fast
2584 lock requests (and glimses) and other RPCs get their own service
2585 threads pool that should be able to service those RPCs
2588 Severity : enhancement
2590 Description: Ability to exchange lustre version between client and servers and
2591 issue warnings at client side if client is too old. Also for
2592 liblustre clients there is ability to refuse connection of too old
2594 Details : New 'version' field is added to connect data structure that is
2595 filled with version info. That info is later checked by server and
2599 Frequency : rare, liblustre only.
2600 Bugzilla : 9296, 9581
2601 Description: Two simultaneous writes from liblustre at offset within same page
2602 might proceed at the same time overwriting eachother with stale
2604 Details : I/O lock withing llu_file_prwv was released too early, before data
2605 actually was hitting the wire. Extended lock-holding time until
2606 server acknowledges receiving data.
2609 Frequency : extremely rare. Never observed in practice.
2611 Description: avoid generating lustre_handle cookie of 0.
2612 Details : class_handle_hash() generates handle cookies by incrementing
2613 global counter, and can hit 0 occasionaly (this is unlikely, but
2614 not impossible, because initial value of cookie counter is
2615 selected randonly). Value of 0 is used as a sentinel meaning
2616 "unassigned handle" --- avoid it. Also coalesce two critical
2617 sections in this function into one.
2619 Severity : enhancement
2621 Description: allow liblustre clients to delegate truncate locking to OST
2622 Details : To avoid overhead of locking, liblustre client instructs OST to
2623 take extent lock in ost_punch() on client's behalf. New connection
2624 flag is added to handle backward compatibility.
2626 Severity : enhancement
2627 Bugzilla : 4928, 7341, 9758
2628 Description: allow number of OST service threads to be specified
2629 Details : a module parameter allows the number of OST service threads
2630 to be specified via "options ost ost_num_threads={N}" in the
2631 OSS's /etc/modules.conf or /etc/modprobe.conf.
2635 Bugzilla : 6146, 9635, 9895
2636 Description: servers crash with bad pointer in target_handle_connect()
2637 Details : In rare cases when a client is reconnecting it was possible that
2638 the connection request was the last reference for that export.
2639 We would temporarily drop the export reference and get a new
2640 one, but this may have been the last reference and the export
2641 was just destroyed. Get new reference before dropping old one.
2643 Severity : enhancement
2644 Frequency : if client is started with failover MDS
2646 Description: Allow multiple MDS hostnames in the mount command
2647 Details : Try to read the configuration from all specified MDS
2648 hostnames during a client mount in case the "primary"
2651 Severity : enhancement
2653 Description: Stop sending data to evicted clients as soon as possible.
2654 Details : Check if the client we are about to send or are sending data to
2655 was evicted already. (Check is done every second of waiting,
2656 for which l_wait_event interface was extended to allow checking
2657 of exit condition at specified intervals).
2660 Frequency : rare, normally only when NFS exporting is done from client
2662 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2664 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2667 Frequency : always, if srand() is called before liblustre initialization
2669 Description: Liblustre uses system PRNG disturbing its usage by user application
2670 Details : Introduce internal to lustre fast and high-quality PRNG for
2671 lustre usage and make liblustre and some other places in generic
2672 lustre code to use it.
2674 Severity : enhancement
2675 Bugzilla : 9477, 9557, 9870
2676 Description: Verify that the MDS configuration logs are updated when xml is
2677 Details : Check if the .xml configuration logs are newer than the config
2678 logs stored on the MDS and report an error if this is the case.
2679 Request --write-conf, or allow starting with --old_conf.
2681 Severity : enhancement
2683 Description: Handle symlinks in the path when checking if Lustre is mounted.
2684 Details : Resolve intermediate symlinks when checking if a client has
2685 mounted a filesystem to avoid duplicate client mounts.
2690 Description: lconf can hit an error exception but still return success.
2691 Details : The lconf command catches the Command error exception at the top
2692 level script context and will exit with the associated exit
2693 status, but doesn't ensure that this exit status is non-zero.
2698 Description: failure of ptlrpc thread startup can cause oops
2699 Details : Starting a ptlrpc service thread can fail if there are a large
2700 number of threads or the server memory is very fragmented.
2701 Handle this without oopsing.
2704 Frequency : always, only if liblustre and non-default acceptor port was used
2706 Description: liblustre cannot connect to servers with non-default acceptor port
2707 Details : tcpnal_set_default_params() was not called and was therefore
2708 ignoring the environment varaible TCPNAL_PORT, as well as other
2709 TCPNAL_ environment variables
2714 Description: two objects could be created on the same OST for a single file
2715 Details : If an OST is down, in some cases it was possible to create two
2716 objects on a single OST for a single file. No problems other
2717 than potential performance impact and spurious error messages.
2721 Bugzilla : 5681, 9562
2722 Description: Client may oops in ll_unhash_aliases
2723 Details : Client dcache may become inconsistent in race condition.
2724 In some cases "getcwd" can fail if the current directory is
2730 Description: Inode refcounting problems in NFS export code
2731 Details : link_raw functions used to call d_instantiate without obtaining
2732 extra inode reference first.
2736 Bugzilla : 9942, 9903
2737 Description: Referencing freed requests leading to crash, memleaks with NFS.
2738 Details : We used to require that call to ll_revalidate_it was always
2739 followed by ll_lookup_it. Also with revalidate_special() it is
2740 possible to call ll_revalidate_it() twice for the same dentry
2741 even if first occurence returned success. This fix changes semantic
2742 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2743 reference on a request referred from the intent.
2744 ll_intent_release() then releases such a request.
2747 Frequency : rare, normally benchmark loads only
2749 Description: unlinked inodes were kept in memory on the client
2750 Details : If a client is repeatedly creating and unlinking files it
2751 can accumulate a lot of stale inodes in the inode slab cache.
2752 If there is no other client load running this can cause the
2753 client node to run out of memory. Instead flush old inodes
2754 from client cache that have the same inode number as a new inode.
2757 Frequency : SLES9 2.6.5 kernel and long filenames only
2758 Bugzilla : 9969, 10379
2759 Description: utime reports stale NFS file handle
2760 Details : SLES9 uses out-of-dentry names in some cases, which confused
2761 the lustre dentry revalidation. Change it to always use the
2765 Frequency : rare, unless heavy write-truncate concurrency is continuous
2766 Bugzilla : 4180, 6984, 7171, 9963, 9331
2767 Description: OST becomes very slow and/or deadlocked during object unlink
2768 Details : filter_destroy() was holding onto the parent directory lock
2769 while truncating+unlinking objects. For very large objects this
2770 may block other threads for a long time and slow overall OST
2771 responsiveness. It may also be possible to get a lock ordering
2772 deadlock in this case, or run out of journal credits because of
2773 the combined truncate+unlink. Solution is to do object truncate
2774 first in one transaction without parent lock, and then do the
2775 final unlink in a new transaction with the parent lock. This
2776 reduces the lock hold time dramatically.
2779 Frequency : rare, 2.4 kernels only
2781 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2782 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2783 before unmounting their filesystem in order to avoid deadlock.
2784 The kernel_locked() function in 2.4 kernels only checks whether
2785 the kernel lock is held, not whether it is this process that is
2786 holding it as 2.6 kernels do.
2791 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2792 Details : The client ptlrpc code may be trying to reconnect to a down
2793 server before a previous connection attempt has timed out.
2794 Increase the reconnect interval to be longer than the connection
2795 timeout interval to avoid sending duplicate connections to
2799 Frequency : echo_client brw_test command
2801 Description: fix echo_client to work with OST preallocated code
2802 Details : OST preallocation code (5137) didn't take echo_client IO path
2803 into account: echo_client calls filter methods outside of any
2804 OST thread and, hence, there is no per-thread preallocated
2805 pages and buffers to use. Solution: hijack pga pages for IO. As
2806 a byproduct, this avoids unnecessary data copying.
2810 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2811 Description: Client can oops in mdc_commit_close() after open replay
2812 Details : It was possible for the MDS to return an open request with no
2813 transaction number in mds_finish_transno() if the client was
2814 evicted, but without actually returning an error. Clients
2815 would later try to replay that open and may trip an assertion
2816 Simplify the client close codepath, and always return an error
2817 from the MDS in case the open is not successful.
2820 Frequency : rare, 2.6 OSTs only
2822 Description: OST may deadlock under high load on fragmented files
2823 Details : If there was a heavy load and highly-fragmented OST filesystems
2824 it was possible to have all the OST threads deadlock waiting on
2825 allocation of biovecs, because the biovecs were not released
2826 until the entire RPC IO was completed. Instead, release biovecs
2827 as soon as they are complete to ensure forward IO progress.
2829 Severity : enhancement
2831 Description: Support for specifying external journal device at mount
2832 Details : If an OST or MDS device is formatted with an external journal
2833 device, this device major/minor is stored in the ext3 superblock
2834 and may not be valid for failover. Allow detecting and
2835 specifying the external journal at mount time.
2840 Description: Mounting an MDS with pending unlinked files may cause oops
2841 Details : target_finish_recovery() calls mds_postrecov() which returned
2842 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2843 considers this an error and immediately begins cleaning up the
2844 lov, just after starting the mds_lov process
2846 Severity : enhancement
2848 Description: Implement 'lfs df' to report actual free space on per-OST basis
2849 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2850 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2851 usage of objects; '-h' to report in human readable format.
2853 ------------------------------------------------------------------------------
2855 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2862 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2863 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2864 at startup during orphan unlinking. Instead, skip buddy update
2865 before setup, as it will be regenerated after recovery is complete.
2870 Description: If an OST is inactive, its locks might reference stale inodes.
2871 Details : lov_change_cbdata() must iterate over all namespaces, even if
2872 they are inactive to clear inode references from the lock.
2874 Severity : enhancement
2875 Frequency : occasional, if non-standard max_dirty_mb used
2877 Description: Client will block write RPCs if not enough grant
2878 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2879 then the client will block writes while waiting for another RPC
2880 to complete instead of consuming its dirty limit. With change
2881 we get improved performance when max_dirty_mb is small.
2883 Severity : enhancement
2884 Bugzilla : 3389, 6253
2885 Description: Add support for supplementary groups on the MDS.
2886 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2887 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2888 lookups for user supplementary groups into a cache.
2892 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2893 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2894 for liblustre O_EXCL is enforced by MDS.
2898 Description: readdir never returns NULL in liblustre.
2899 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2900 dentry in struct dirent.
2904 Frequency : liblustre only, depends on application IO pattern
2905 Description: liblustre clients evicted if not contacting servers
2906 Details : Don't put liblustre clients into the ping_evictor list, so
2907 they will not be evicted by the pinger ever.
2909 Severity : enhancement
2911 Description: Add ability to evict clients by NID from MDS.
2912 Details : By echoing "nid:$NID" string into
2913 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2914 $NID would be instantly evicted from this MDS and from all active
2915 OSTs connected to it.
2919 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2920 Details : lookup_it_finish() used to query file size from OSTs that was not
2925 Description: service threads change working directory to that of init
2926 Details : Starting lustre service threads may pin the working directory
2927 of the parent thread, making that filesystem busy. Threads
2928 now change to the working directory of init to avoid this.
2932 Frequency : during shutdown only
2933 Description: shutdown with a failed MDS or OST can cause unmount to hang
2934 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2939 Frequency : 2.6 only
2940 Description: chmod/chown may include an extra supplementary group
2941 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2942 supplementary group and if none is specified this is used.
2945 Bugzilla : 5479 (6816)
2947 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2948 Details : If lookup is in progress on a file that is unlinked we might try
2949 to revalidate the inode and fail in revalidate after lookup is
2950 complete and ll_file_open() enqueues the open again but
2951 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2954 Frequency : always, if lconf --abort_recovery used
2956 Description: lconf --abort_recovery fails with 'Operation not supported'
2957 Details : lconf was attempting to abort recovery on the MDT device and not
2960 ------------------------------------------------------------------------------
2962 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2967 Frequency : rare (only unsupported configurations with a node running as an
2969 Bugzilla : 6514, 5137
2970 Description: Mounting a Lustre file system on a node running as an OST could
2972 Details : OSTs now preallocates memory needed to write out data at
2973 startup, instead of when needed, to avoid having to
2974 allocate memory in possibly low memory situations.
2975 Specifically, if the file system is mounted on on OST,
2976 memory pressure could force it to try to write out data,
2977 which it needed to allocate memory to do. Due to the low
2978 memory, it would be unable to do so and the node would
2979 become unresponsive.
2981 Severity : enhancement
2983 Description: Addition of lconf --service command line option
2984 Details : lconf now accepts a '--service <arg>' option, which is
2985 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2987 Severity : enhancement
2989 Description: Failover mode is now the default for OSTs.
2990 Details : By default, OSTs will now run in failover mode. To return to
2991 the old behaviour, add '--failout' to the lmc line for OSTs.
2993 Severity : enhancement
2995 Description: Health checks are now provided for MDS and OSTs
2996 Details : Additional detailed health check information on MSD and OSTs
2997 is now provided through the procfs health_check value.
3000 Frequency : occasional, depends on IO load
3002 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3003 after numerous create/delete cycles
3004 Details : The ext3 inode allocation policy would not allocate new inodes
3005 very well on the OSTs because there are no new directories
3006 being created. Instead we look for groups with free space if
3007 the parent directories are nearly full.
3012 Description: Network or server problems during mount may cause partially
3013 mounted clients instead of returning an error.
3014 Details : The config llog parsing code may overwrite the error return
3015 code during mount error handling, returning success instead
3021 Description: MDS can fail to allocate large reply buffers
3022 Details : After long uptimes the MDS can fail to allocate large reply
3023 buffers (e.g. zconf client mount config records) due to memory
3024 fragmentation or consumption by the buffer cache. Preallocate
3025 some large reply buffers so that these replies can be sent even
3026 under memory pressure.
3030 Frequency : rare (liblustre)
3031 Description: fsx running with liblustre complained that using truncate() to
3032 extend the file doesn't work. This patch corrects that issue.
3033 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3034 ATTR_SIZE and lsm use in llu_setattr_raw.
3038 Frequency : rare, only 2.6 kernels
3039 Description: Unusual file access patterns on the MDS may result in inode
3040 data being lost in very rare circumstances.
3041 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3042 code in the 2.6 kernel caused the inode and/or EA data not to
3043 be read from disk, causing single-file corruption.
3047 Frequency : rare, only 2.6 filesystems using extents
3048 Description: Heavy concurrent write and delete load may cause data corruption.
3049 Details : It was possible under high-load situations to have an extent
3050 metadata block in the block device cache from a just-unlinked
3051 file overwrite a newly-allocated data block. We now unmap any
3052 metadata buffers that alias just-allocated data blocks.
3056 Frequency : filesystems with default stripe_count larger than 77
3057 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3058 Details : lconf specifies an inode size of 4096 bytes when the default
3059 stripe_count is larger than 77. This conflicts with the default
3060 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3061 this case to avoid pinning too much memory for large EAs.
3063 ------------------------------------------------------------------------------
3065 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3070 Frequency : rare (extremely heavy IO load with hundreds of clients)
3072 Description: Client is evicted, gets IO error writing to file
3073 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3074 caused clients to be evicted for AST timeouts. The fixes in
3075 bug 5192 mean we no longer need to have such short AST timeouts
3076 so ldlm_timeout has been increased.
3079 Frequency : occasional during --force or --failover shutdown under load
3080 Bugzilla : 5949, 4834
3081 Description: Server oops/LBUG if stopped with --force or --failover under load
3082 Details : a collection of import/export refcount and cleanup ordering
3083 issues fixed for safer force cleanup
3086 Frequency : only filesystems larger than 120 OSTs
3087 Bugzilla : 5990, 6223
3088 Description: lfs getstripe would oops on a very large filesystem
3089 Details : lov_getconfig used kfree on vmalloc'd memory
3092 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3093 Bugzilla : 6242, 6243
3094 Description: reading from files that had been truncated to a non-zero size
3095 but never opened returned no data
3096 Details : ll_file_read() reads zeros from no-object files to EOF
3101 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3102 Details : There's little harm in aborting MDS/OSS recovery and letting it
3103 try again, so I removed the LASSERT and return an error instead.
3105 Severity : enhancement
3107 Description: New debugging infrastructure for tracking down data corruption
3108 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3109 (b) cover more of the client-side code path, and (c) try to narrow
3110 down where problems occurred
3114 Bugzilla : 3819, 4364, 4397, 6313
3115 Description: Racing close and eviction MDS could cause assertion in mds_close
3116 Details : It was possible to get multiple mfd references during close and
3117 client eviction, leading to one thread referencing a freed mfd.
3119 Severity: : enhancement
3120 Bugzilla : 3262, 6359
3121 Description: Attempts to reconnect to servers are now more aggressive.
3122 Details : This builds on the enhanced upcall-less recovery that was added
3123 in 1.4.2. When trying to reconnect to servers, clients will
3124 now try each server in the failover group every 10 seconds. By
3125 default, clients would previously try one server every 25 seconds.
3130 Description: After recovery, certain operations trigger a failed
3131 assertion on a client.
3132 Details : Failing over an mds, using lconf -d --failover, while a
3133 client was doing a readdir() call would cause the client to
3134 LBUG after recovery completed and the readdir() was resent.
3136 Severity : enhancement
3138 Description: Default groups are now added by lconf
3139 Details : You can now run lconf --group <servicename> without having to
3140 manually add groups with lmc.
3143 Frequency : occasional
3145 Description: Nodes with an elan id of 0 trigger a failed assertion
3148 Frequency : always when accessing e.g. tty/console device nodes
3150 Description: tty and some other devices nodes cannot be used on lustre
3151 Details : file's private_data field is used by device data and lustre
3152 values in there got lost. New field was added to struct file to
3153 store fs-specific private data.
3156 Frequency : when exporting Lustre via NFS
3158 Description: NFSD failed occasionally when looking up a path component
3159 Details : NFSD is looking up ".." which was broken in ext3 directories
3160 that had grown large enough to become hashed.
3163 Frequency : Clusters with multiple interfaces not on the same subnet
3165 Description: Nodes will repeatedly try to reconnect to an interface which it
3166 cannot reach and report an error to the log.
3167 Details : Extra peer list entries will be created by lconf with some peers
3168 unreachable. lconf now validates the peer before adding it.
3171 Frequency : Only if a default stripe is set on the filesystem root.
3173 Description: Setting a default stripe on the filesystem root prevented the
3174 filesystem from being remounted.
3175 Details : The client was sending extra request flags in the root getattr
3176 request and did not allocate a reply buffer for the dir EA.
3179 Frequency : occasional, higher if lots of files are accessed by one client
3180 Bugzilla : 6159, 6097
3181 Description: Client trips assertion regarding lsm mismatch/magic
3182 Details : While revalidating inodes the VFS looks up inodes with ifind()
3183 and in rare cases can find an inode that is being freed.
3184 The ll_test_inode() code will free the lsm during ifind()
3185 when it finds an existing inode and then the VFS later attaches
3186 this free lsm to a new inode.
3190 Bugzilla : 6422, 7030
3191 Description: MDS deadlock between mkdir and client eviction
3192 Details : Creating a new file via mkdir or mknod (starting a transaction
3193 and getting the ns lock) can deadlock with client eviction
3194 (gets ns lock and trying to finish a synchronous transaction).
3197 Frequency : occasional
3198 Description: While starting a server, the fsfilt_ext3 module could not be
3200 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3201 kernels. Previously, lconf would still use the ext3 name
3202 when trying to load modules. Now, it will correctly use
3203 ext3 on 2.4 and ldiskfs on 2.6.
3205 Severity : enhancement
3206 Description: The default stripe count has been changed to 1
3207 Details : The interpretation of the default stripe count (0, to lfs
3208 or lmc) has been changed to mean striping across a single
3209 OST, rather than all available. For general usage we have
3210 found a stripe count of 1 or 2 works best.
3212 Severity : enhancement
3213 Description: Add support for compiling against Cray portals.
3214 Details : Conditional compiling for some areas that are different
3218 Frequency : occasional
3219 Bugzilla : 6409, 6834
3220 Description: Creating files with an explicit stripe count may lead to
3221 a failed assertion on the MDS
3222 Details : If some OSTs are full or unavailable, creating files may
3223 trigger a failed assertion on the MDS. Now, Lustre will
3224 try to use other servers or return an error to the
3228 Frequency : occasional
3230 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3231 caused lock timeout during readahead (since 1.4.2).
3232 Details : Processes doing readahead might match a lock that hasn't been
3233 granted yet if there are overlapping and conflicting lock
3234 requests. The readahead process waits on ungranted lock
3235 (original lock is CBPENDING), while OST waits for that process
3236 to cancel CBPENDING read lock and eventually evicts client.
3238 Severity : enhancement
3240 Description: Initial enabling of flock support for clients
3241 Details : Implements fcntl advisory locking and file status functions.
3242 This feature is provided as an optional mount flag (default
3243 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3244 locking are implemented yet, and those that are are not guaranteed
3245 to be completely correct in production environments.
3246 mount -t lustre -o [flock|noflock] ...
3249 Frequency : occasional
3251 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3252 assertion in the ext3 JBD (journaling) layer.
3253 Details : The b_committed_data struct is protected by the big kernel lock
3254 in 2.4 kernels, serializing journal_commit_transaction() and
3255 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3256 access to this struct. In 2.6 kernels there is finer grained
3257 locking to improve SMP performance of the JBD layer.
3261 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3266 Description: merge and cleanup kernel patches.
3269 -----------------------------------------------------------------------------
3271 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3273 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3274 versions of Lustre. Please update all servers and clients to
3275 version 1.4.2 or later at the same time. You must also run
3276 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3277 to update the configuration logs.
3279 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3280 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3281 - swab llog records of type '0' so we get proper header size/idx (5861)
3282 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3283 - fix rename of one directory over another leaking an inode (5953)
3284 - avoid SetPageDirty on 2.6 (5981)
3285 - don't re-add just-being-destroyed locks to the waiting list (5653)
3286 - when creating new directories, inherit the parent's custom
3287 striping settings if present parent (3048)
3288 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3289 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3290 - don't hold client locks on temporary worklist from l_lru (5666)
3291 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3292 - automatically evict dead clients (5921)
3293 - Update file size properly in create+truncate+fstat case (6196)
3294 - Do not unhash mountpoint dentries, do not allow removal of
3296 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3297 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3298 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3299 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3300 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3301 - Correctly handle reads of files with no objects (6243)
3302 - lctl recover will also mark a device active if deactivate used (5933)
3304 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3305 - allow --write-conf on an MDS with different nettype than client (5619)
3306 - don't write config llogs to MDS for mounts not from that MDS (5617)
3307 - lconf should create multiple TCP connections from a client (5201)
3308 - init scripts are now turned off by default; run chkconfig --on
3309 lustre and chkconfig --on lustrefs to use them
3310 - upcalls are no longer needed for clients to recover to failover
3312 - add --abort-recovery option to lconf to abort recovery on device
3314 - add support for an arbitrary number of OSTs (3026)
3315 - Quota support protocol changes.
3316 - forward compatibility changes to wire structs (6007)
3317 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3318 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3319 - improved "lustre status" script
3320 - initialize blocksize for non-regular files (6062)
3321 - added --disable-server and --disable-client configure options (5782)
3322 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3323 - Vanilla 2.4.29 support
3324 - increase maximum number of obd devices to 520 (6242)
3325 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3326 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3327 SLES 9 2.6.5 kernels
3328 - limit stripes per file to 160 (the maximum EA size) (6093)
3330 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3333 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3334 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3335 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3336 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3337 - small ext3 extents cleanups and fixes (5733)
3338 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3339 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3340 - avoid needless client->OST connect, fix handle mismatch (5317)
3341 - fix DLM error path that led to out-of-sync client, long delays (5779)
3342 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3343 - fix several locking issues related to i_size (5492,5624,5654,5672)
3344 - don't move pending lock onto export if it is already evicted (5683)
3345 - fix kernel oops when creating .foo in unlinked directory (5548)
3346 - fix deadlock in obdfilter statistics vs. object create (5811)
3347 - use time_{before,after} to avoid timer jiffies wrap (5882)
3348 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3349 - Do not leak request if server was not able to process it (5154)
3350 - If mds_open unable to find parent dir, make that negative lookup(5154)
3351 - don't create new directories with extent-mapping (5909, 5936)
3353 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3354 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3355 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3356 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3357 - support for RHEL4 kernel on i686 (5773)
3358 - provide error messages when incompatible logs are encountered (5898)
3360 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3361 * version 1.4.0.10 (1.4.1 release candidate 1)
3363 - don't keep a lock reference when lock is not granted (4238)
3364 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3365 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3366 - drop import inflight refcount on signal_completed_replay error (5255)
3367 - unlock page after async write error during send (3677)
3368 - handle missing objects in filter_preprw_read properly (5265)
3369 - no transno return for symlink open, don't save no-trasno open (3440)
3370 - don't try to complete elan receive that already failed (4012)
3371 - free RPC server reply state on error (5406)
3372 - clean up thread from ptlrpc_start_thread() on error (5160)
3373 - readahead could read extra page into cache that wasn't ejected (5388)
3374 - prevent races in class_attach/setup/cleanup/detach (5260)
3375 - don't dereference de->d_inode after l_dput of de (5458)
3376 - use "int" for stripe value returned from lock_to_stripe (5544)
3377 - mballoc allocation and error-checking fixes in 2.6 (5504)
3378 - block device patches to fix I/O request sizes in 2.6 (5482)
3379 - look up hostnames for IB nals (5602)
3380 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3381 - don't start multiple acceptors for the same port (5277)
3382 - fix incorrect LASSERT in mds_getattr_name (5635)
3383 - export a proc file for general "ping" checking (5628)
3384 - fix "lfs check" to not block when the MDS is down (5628)
3386 - service request history (4965)
3387 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3388 - create an "evict_client" /proc entry on OSTs, like the MDS has
3389 - fix mount usage message, return errors per mount(8) (5168)
3390 - change grep [] to grep "[]" in tests so they work in more UMLs
3391 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3392 - remove ancient LOV_MAGIC_V0 EA support (5047)
3393 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3394 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3395 - allow readahead allocations to fail when low on memory (5383)
3396 - mmap locking landed again, after considerable improvement (2828)
3397 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3399 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3402 - send OST transaction number in read/write reply to free req (4966)
3403 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3404 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3405 - the watchdog thread now runs as interruptible (5246)
3406 - flock/lockf fixes (but it's still disabled, pending 5135)
3407 - don't use EXT3 constants in llite code (5094)
3408 - memory shortage at startup could cause assertion (5176)
3410 - reorganization of lov code
3411 - single portals codebase
3413 - add extents/mballoc support (5025)
3414 - direct I/O reads in the obdfilter (4048)
3415 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3417 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3420 - send OST transaction number in read/write reply to free req (4966)
3421 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3422 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3423 - the watchdog thread now runs as interruptible (5246)
3424 - handle missing objects in filter_preprw_read properly (5265)
3425 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3426 - drop import inflight refcount on signal_completed_replay error (5255)
3427 - unlock page after async write error during send (3677)
3428 - return original error code on reconstructed replies (3761)
3429 - no transno return for symlink open, don't save no-trasno open (3440)
3431 - add pid to ldlm debugging output (4922)
3432 - bump the watchdog timeouts -- we can't handle 30sec yet
3433 - extra debugging for orphan dentry/inode bug (5259)
3435 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3438 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3439 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3440 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3441 - fix NULL dereference in /proc/sys/portals/routes (4827)
3442 - allow failed mdc_close() operations to be interrupted (4561)
3443 - stop precreate on OST before MDS would time out on it (4778)
3444 - don't send partial-page writes before EOF from client (4410)
3445 - discard client grant for sub-page writes on large-page clients (4520)
3446 - don't free dentries not owned by NFS code, check generation (4806)
3447 - fix lsm leak if mds_create_objects() fails (4801)
3448 - limit debug_daemon file size, always print CERROR messages (4789)
3449 - use transno after validating reply (3892)
3450 - process timed out requests if import state changes (3754)
3451 - update mtime on OST during writes, return in glimpse (4829)
3452 - add mkfsoptions to LDAP (4679)
3453 - use ->max_readahead method instead of zapping global ra (5039)
3454 - don't interrupt __l_wait_event() during strace
3456 - add software watchdogs to catch hung threads quickly (4941)
3457 - make lustrefs init script start after nfs is mounted
3458 - fix CWARN/ERROR duplication (4930)
3459 - return async write errors to application if possible (2248)
3460 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3461 - print NAL number in %x format (4645)
3462 - update barely-supported suse-2.4.21-171 series (4842)
3463 - support for sles 9 %post scripts
3464 - support for building 2.6 kernel-source packages
3465 - support for sles km_* packages
3467 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3470 - ignore -ENOENT errors in osc_destroy (3639)
3471 - notify osc create thread that OSC is being cleaned up (4600)
3472 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3473 - reconstruct ost_handle() like mds_handle() (4657)
3474 - create a new thread to do import eviction to avoid deadlock (3969)
3475 - let lconf resolve symlinked-to devices (4629)
3476 - don't unlink "objects" from directory with default EA (4554)
3477 - hold socknal file ref over connect in case target is down (4394)
3478 - allow more than 32000 subdirectories in a single directory (3244)
3479 - fix blocks count for O_DIRECT writes (3751)
3480 - OST returns ENOSPC from object create when no space left (4539)
3481 - don't send truncate RPC if file size isn't changing (4410)
3482 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3483 - bind to privileged port in socknal and tcpnal (3689)
3485 - rate limit CERROR/CWARN console message to avoid overload (4519)
3486 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3487 - basic mmap support (3918)
3488 - kernel patch series update from b1_4 (4711)
3490 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3493 - avoid crash during MDS cleanup with OST shut down (2775)
3494 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3495 - don't use bad inodes on the MDS (3744)
3496 - dynamic object preallocation to improve recovery speed (4236)
3497 - don't hold spinlock over lock dumping or change debug flags (4401)
3498 - don't zero obd_dev when it is force cleaned (3651)
3499 - print grants to console if they go negative (4431)
3500 - "lctl deactivate" will stop automatic recovery attempts (3406)
3501 - look for existing locks in ldlm_handle_enqueue() (3764)
3502 - don't resolve lock handle twice in recovery avoiding race (4401)
3503 - revalidate should check working dir is a directory (4134)
3505 - don't always mark "slow" obdfilter messages as errors (4418)
3507 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3510 - don't close LustreDB during write_conf until it is done (3860)
3511 - fix typo in lconf for_each_profile (3821)
3512 - allow dumping logs from multiple threads at one time (3820)
3513 - don't allow multiple threads in OSC recovery (3812)
3514 - fix debug_size parameters (3864)
3515 - fix mds_postrecov to initialize import for llog ctxt (3121)
3516 - replace config semaphore with spinlock (3306)
3517 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3518 - don't allow enqueue to complete on a destroyed export (3822)
3519 - down write_lock before checking llog header bitmap (3825)
3520 - recover from lock replay timeout (3764)
3521 - up llog sem before sending rpc (3652)
3522 - reduce ns lock hold times when setting kms (3267)
3523 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3524 - fix NULL deref and obd_dev leak on setup error (3312)
3525 - replace some LBUG about llog ops with error handling (3841)
3526 - don't match INVALID dentries from d_lookup and spin (3784)
3527 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3528 - fix invalid assertion in ptlrpc_set_wait (3880)
3530 - add libwrap support for the TCP acceptor (3996)
3531 - add /proc/sys/portals/routes for non-root route listing (3994)
3532 - allow setting MDS UUID in .xml (2580)
3533 - print the stack of a process that LBUGs (4228)
3535 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3538 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3539 - make sure to unset replay flag from failed open requests (3440)
3540 - if default stripe count is 0, use OST count for inode size (3636)
3541 - update parent mtime/ctime on client for create/unlink (2611)
3542 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3543 - free recovery state on server during a forced cleanup (3571)
3544 - unregister_reply for resent reqs (3063)
3545 - loop back devices mounting and status check on 2.6 (3563)
3546 - fix resource-creation race that can provoke i_size == 0 (3513)
3547 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3548 - more debugging for page-accounting assertion (3746)
3549 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3550 - avoid deadlock after precreation failure (3758)
3551 - fix race and lock order deadlock in orphan handling (3450, 3750)
3552 - add validity checks when grabbing inodes from l_ast_data (3599)
3554 - add /proc/.../recovery_status to obdfilter (3428)
3555 - lightweight CDEBUG infrastructure, debug daemon (3668)
3556 - change default OSC RPC parameters to be better on small clusters
3557 - turn off OST read cache for files smaller than 32MB
3558 - install man pages and include them in rpms (3100)
3559 - add new init script for (un)mounting lustre filesystems (2593)
3560 - run chkconfig in %post for init scripts (3701)
3561 - drop scimac NAL (unmaintained)
3563 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3566 - clean kiobufs before and after use (3485)
3567 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3568 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3569 - change init script to fail more clearly if not run as root (1528)
3570 - allow clients to reconnect during replay (1742)
3571 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3572 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3573 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3574 - fix cond_resched (3554)
3575 - only evict unfinished clients after recovery (3515)
3576 - allow bulk resend, prevent data loss (3570)
3577 - dynamic ptlrpc request buffer allocation (2102)
3578 - don't allow unlinking open directory if it isn't empty (2904)
3579 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3580 - remove extraneous obd dereference causing LASSERT failure (3334)
3581 - don't use get_cycles() when creating temp. files on the mds (3156)
3582 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3583 - handle EEXIST for set-stripe, set proper directory name (3336)
3585 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3586 - fix ksocknal_fmb_callback() error messages (2918)
3588 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3591 - don't copy lvb into (possibly NULL) reply on error (2983)
3592 - don't deref dentry after dput, don't free lvb on error (2922)
3593 - use the kms to determine writeback rpc length (2947)
3594 - increment oti_logcookies when osc is inactive (2948)
3595 - update client's i_blocks count via lvb messages (2543)
3596 - handle intent open/close of special files properly (1557)
3597 - mount MDS with errors=remount-ro, like obdfilter (2009)
3598 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3599 - don't use cancelling-locks' kms values (2947)
3600 - use highest lock extent for kms, not last one (2925)
3601 - don't dereference ERR_PTR() dentry in error handling path (3107)
3602 - fix thread race in portals_debug_dumplog() (3122)
3603 - create lprocfs device entries at setup instead of at attach (1519)
3604 - common AST error handler, don't evict client on completion race (3145)
3605 - zero nameidata in detach_mnt in 2.6 (3118)
3606 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3607 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3608 - initialize RPC timeout timer earlier for 2.6 (3219)
3609 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3610 - print nal/nid for unknown nid (3258)
3611 - additional checks for oscc recovery before doing precreate (3284)
3612 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3613 - don't crash in mdc_close for bad permissions on open (3285)
3614 - zero i_rdev for non-device files (3147)
3615 - clear page->private before handing to FS, better assertion (3119)
3616 - tune the read pipeline (3236)
3617 - fix incorrect decref of invalidated dentry (2350)
3618 - provide read-ahead stats and refine rpc in flight stats (3328)
3619 - don't hold journal transaction open across create RPC (3313)
3620 - update atime on MDS at close time (3265)
3621 - close LDAP connection when recovering to avoid server load (3315)
3622 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3623 - don't leak open file on MDS after open resend (3325)
3624 - serialize filter_precreate and filter_destroy_precreated (3329)
3625 - loop device shouldn't call sync_dev() for nul device (3092)
3626 - clear page cache after eviction (2766)
3627 - resynchronize MDS->OST in background (2824)
3628 - refuse to mount the same filesystem twice on same mountpoint (3394)
3629 - allow llmount to create routes for mounting behind routers (3320)
3630 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3631 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3632 - fix rare problem with rename on htree directories (3417)
3634 - allow default OST striping configuration per directory (1414)
3635 - fix compilation for qswnal for 2.6 kernels (3125)
3636 - increase maximum number of MDS request buffers for large systems
3637 - change liblustreapi to be useful for external progs like lfsck (3098)
3638 - increase local configuration timeout for slow disks (3353)
3639 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3641 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3644 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3645 - don't overwrite extent policy data in reply if lock was blocked (2901)
3646 - drop filter export grants atomically with removal from device (2663)
3647 - del obd_self_export from work_list in class_disconnect_exports (2908)
3648 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3649 - swab reply message in mdc_close, other PPC fixes (2464)
3650 - fix destroying of named logs (2325)
3651 - overwrite old logs when running lconf --write_conf (2264)
3652 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3653 - fix race in target_handle_connect (2898)
3654 - mds_reint_create() should take same inode create lock (2926)
3655 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3656 - don't close files for self_export to avoid uninitialized obd (2936)
3657 - allow MDS with the same name as client node (2939)
3658 - hold dentry reference for closed log files for unlink (2325)
3659 - reserve space for all logs during transactions (2059)
3660 - don't evict page beyond end of stripe extent (2925)
3661 - don't oops on a deleted current working directory (2399)
3662 - handle hard links to targets without a parent properly (2517)
3663 - don't dereference NULL lock when racing during eviction (2867)
3664 - don't grow lock extents when lots of conflicting locks (2919)
3666 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3669 - account for cache space usage on clients to avoid data loss (974)
3670 - lfsck support in lustre kernel code (2349)
3671 - reduce journal credits needed for BRW writes (2370)
3672 - orphan handling to avoid losing space on client/server crashes
3673 - ptlrpcd can be blocked, stopping ALL progress (2477)
3674 - use lock value blocks to assist in proper KMS, faster stat (1021)
3675 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3676 - recovery for initial connections (2355)
3677 - fixes for mds_cleanup_orphans (1934)
3678 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3679 - block all file creations until orphan recovery completes (1901)
3680 - client remove rq_connection from request struct (2423)
3681 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3682 - recovery timer race (2670)
3683 - mdc_close recovey bug (2532)
3684 - ptlrpc cleanup bug (2710)
3685 - mds timeout on local locks (2588)
3686 - namespace lock held during RPCs (2431)
3687 - handle interrupted sync write properly (2503)
3688 - don't try to handle a message that hasn't been replied to (2699)
3689 - client assert failure during cleanup after abort recovery (2701)
3690 - leak mdc device after failed mount (2712)
3691 - ptlrpc_check_set allows timedout requests to complete (2714)
3692 - wait for inflight reqs when ptlrpcd finishes (2710)
3693 - make sure unregistered services are removed from the srv_list
3694 - reset bulk XID's when resending them (caught by 1138 test)
3695 - unregister_bulk after timeout
3696 - fix lconf error (2694)
3697 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3698 - readahead locks pages, leaves pending causing memory pressure (2673)
3699 - increase OST request buffers to 4096 on large machines (2729)
3700 - fix up permission of existing directories in simple_mkdir (2661)
3701 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3702 - don't assign transno to errored transactions (2742)
3703 - don't delete objects on OST if given a bogus objid from MDS (2751)
3704 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3705 - if rq_no_resend, then timeout request after recovery (2432)
3706 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3707 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3708 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3709 - MDS recovery shouldn't delete objects that it has given out (2730)
3710 - if enqueue arrives after completion, don't clobber LVB (2819)
3711 - don't unlock pages twice when trigger_group_io returns error (2814)
3712 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3713 - don't write pages to disk if there was an error (1450)
3714 - don't ping imports that have recovery disabled (2676)
3715 - take buffered bytes into account when balancing socknal conn (2817)
3716 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3717 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3718 - remove little-endian swabbing of llog records (1987)
3719 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3720 - retry reposting request buffers if they fail (1191)
3721 - grow extent at grant time to avoid granting a revoked lock (2809)
3722 - lock revoke doesn't evict page if covered by a second lock (2765)
3723 - disable VM readahead to avoid reading outside lock extents (2805)
3725 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3726 - updated kernel patches for hp-2.4.20 kernel (2681)
3728 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3731 - fix truncated write corruption (2366)
3732 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3734 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3735 - ensure len(uuid) < 37 in lmc (1171)
3736 - fix ia64 OOPS in llog_test (2255)
3737 - zero end of page at obdfilter for partial page writes (2648)
3738 - don't leave stale dentries around after renames (bug 2428)
3739 - fix timeouts when evicting a client with a single lock held (2642)
3740 - set deadline for the initial HELLO message to drain (2634)
3741 - print out dotted-quad IP addresses in the socknal (2302)
3743 - additional debugging for MDS client eviction problem (2443)
3744 - fix mkfsoptions support for osts (2603, 2604)
3746 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3749 - add series for the vanilla 2.6.0 kernel
3750 - add series for the vanilla 2.4.24 kernel
3751 - add series for a cray x86/64 UL kernel drop
3752 - fix xattr patches for the vanilla 2.4.19 series
3754 - generate true UUIDs in lmc (1171)
3755 - have portals stack dumping break in UML (2466)
3756 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3757 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3758 - wake callback waiting threads on client eviction (2460)
3759 - Add --ptldebug and --subsystem to lmc (1719)
3760 - update assertion to allow safe interrupt allocation
3761 - set rq_no_resend for cancel requests (2432)
3762 - recalculate ptlrpcd timeout after resend (2494)
3763 - call vfs_rmdir when removing pending directories (2368)
3764 - fix renaming a file to itself (2429)
3765 - lmc creates a default one-stripe lov (2454)
3766 - expand procfs space to handle large clusters (2326)
3767 - increase UML stack to avoid overflow
3768 - update lconf's list of debug and subsystem masks
3769 - fix lfs find --obd (2510)
3770 - /proc tunable for disabling filter read caching (2591)
3771 - stop rpm packages from altering slapd.conf (2301)
3772 - disable nagle in the socknal under 0conf (2578)
3773 - choose mds inode size based on stripe count (2572)
3774 - fix kernel-source rpm problems (2516)
3776 - add --disable-doc to avoid pdf generation (2421)
3777 - update documentation, tests, type-os, comments
3778 - avoid format warnings on ia64
3779 - remove the TOE NAL
3780 - tiny code cleanups by removing unused fields
3782 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3785 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3786 - remove the most blatant lies from BUILDING file (2371)
3787 - change default debug level to reasonable production setting
3788 - reduce client side cache size to reduce cache flush time
3789 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3790 - make TCP zerocopy and pinger support enabled by default (2476)
3791 - sync writes completed after process exits caused crashes (2319)
3792 - maintain correct mount count on the MDS (2356)
3793 - backout 1557, because 2316 wasn't really fixed
3794 - better file I/O statistics gathering in /proc
3795 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3796 - another kernel patch to fix zero-copy TCP function export
3797 - don't take duplicate lock when processing re-sent getattr (2420)
3798 - lctl uses obd_self_export instead of creating new conn (2353)
3799 - MDS/OST recovery case which requires object creation asserted (2425)
3800 - move lfs from /usr/sbin to /usr/bin in packages
3801 - fix race between mds_client_add and mds_client_free (2417)
3802 - use kmalloc instead of slabs in portals (2430)
3803 - don't create duplicate records when a failover MDS is present (2442)
3804 - remove unnecessary mount age check (2332)
3805 - don't remove directory inodes from locks prematurely (2451)
3806 - don't break if MDS service name is the same as hostname (2103)
3807 - fix races in client write RPC generation when cache full (2482)
3809 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3812 - remove now-unused request->rq_obd (278)
3813 - if an allocation fails, print out how much memory we've used (1933)
3814 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3815 - add the "configurable stack size" patch to most series files (1256)
3816 - ability to write large log records, for 100+ OST configs (2306)
3817 - fix NULL deref when filter_prep fails (2314)
3818 - fix operator precedence error in filter_sync
3819 - dynamic allocation of socknal TX descriptors (2315)
3820 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3821 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3822 - ensure that CWARN messages in Portals always get to the syslog
3823 - __init/__exit are not for prototype decls (ldlm_init/exit)
3824 - x86-64 compile warning fixes
3825 - fix gateway LMC keyword conflict (2318)
3826 - fix MDS lock inversions in getattr/reint paths (1844)
3827 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3828 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3829 - DLM race condition prevented some lock evictions (2328)
3830 - ENOMEM detection and retry on socknal sends (2230)
3831 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3832 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3833 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3834 - better error messages when a client is rejected during recovery (1505)
3835 - avoid cancelling locks which were never granted, after failure (2330)
3836 - fix i_sem/journal inversion in mds_client_add (2333)
3837 - fix truncate/getattr lock cycle deadlock (2334)
3838 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3839 - fix two rare exit paths which could leak an l_lock() ref (2321)
3840 - fencepost error in MDS/OST orphan recovery (2226)
3841 - make log record alignment 8 bytes (1988)
3842 - lstripe now fails when requested offset > ost_count (2237)
3843 - ensure that all kernel series have a complete list.h (1607)
3844 - fix crashes in special-file operations (2316)
3845 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3847 - allow configurable automake binary, for testing new versions
3848 - small update to the lfs documentation
3850 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3852 * fix negative export reference count in fsfilt_sync (2312)
3854 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3855 * release candidate 0.9.1
3857 - orphans are moved into the PENDING directory for possible recovery
3858 - replayed opens now open by fid for orphan/rename safety (1042)
3859 - last close of an orphan inode generates a transno (683)
3860 - chdir() and mount() now pin the directory entry (1020)
3861 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3862 - discard very old requests without processing them (1502)
3863 - remove some common, well-understood CERRORs (1505)
3864 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3865 - clear "grant" flags in OST replies until OST grant code lands (1644)
3866 - fix read performance by not clobbering i_blksize on client (1598)
3867 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3868 - make LRU size a /proc tunable, clears locks when reduced (707)
3869 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3870 - server locks take references on exports now (1558)
3871 - build fixes for 2.4.20-rh trees (1663)
3872 - return an error from lov_create if all OSCs are inactive (1751)
3873 - fix import levels when a reconnect happens without a timeout (1597)
3874 - exit early from mds_open if we get a lookup error (1749)
3875 - partial page read at EOF wouldn't wait for disk before sending (1642)
3876 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3877 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3878 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3879 - write/truncate lock inversion (1639)
3880 - Don't auto-load obdclass, portals modules during cleanup (1495)
3881 - fix timestamps from jumping to "now" (1763)
3882 - extra journal assertions (1648)
3883 - add an extra multiunlink test (1771)
3884 - fix read_record/write_record API (1776)
3885 - fix leak of offset_extent, possible incorrect i_size later (1772)
3886 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3887 - Debugging for the kqswnal_get_idle_tx problems (1820)
3888 - Allow recovery to be attempted multiple times (1536)
3889 - Write out MDS last_rcvd file after it is first created (1600)
3890 - Fix tx_descriptor leak in failed transmit situations (1827)
3891 - ext3 journaling fixes for assertion failure after IO error (1871)
3892 - class_export_put() on freed export after completion AST error (1896)
3893 - Fix revalidate looping in VFS (1322)
3894 - Don't access a freed export during MDS_REINT timeout (1521)
3895 - Add open-unlink recovery support on the MDS (1673,1764)
3896 - Return an error if no MDS data was read from last_rcvd (1946)
3897 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3898 - Don't setup a disk device that doesn't match exported UUID (317)
3899 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3900 - avoid committing NULL handle in force close
3901 - local.sh is now a one-stripe LOV configuration
3902 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3903 - fix ext3 htree duplicate directory entry corruption (1516)
3904 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3905 - update to new LOV EA format (2097)
3906 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3908 2003-06-15 Phil Schwan <phil@clusterfs.com>
3911 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3912 - per-import/export recovery handling (958, 931, 959)
3913 - multiple last-rcvd slots, for serving multiple FSes (949)
3914 - connections are again shared between multiple imp/exports (963, 964)
3915 - "umount -f" would hang if any requests needed to be sent (393, 978)
3916 - avoid pinning large req buffer by copying for queued messages (989)
3917 - add "uuid" to "lctl device" command to help upcalls (991)
3918 - "open" RPCs with transnos would confuse recovery counters (1037)
3919 - do proper endian conversion of all wire messages (288, 340, 891)
3920 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3921 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3922 - fix ptlrpc_connection leak in target_handle_connect (1174)
3923 - fix import refcounting bug in OST and MDS cleanup (1134)
3924 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3925 - fix very unlikely obd_types race condition (501)
3926 - remove osc_open hack for echo_client (1187)
3927 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3928 - a failure in read_inode2 leads to deadlock (1139)
3929 - cancel ack-locks as soon as transaction is committed (1072)
3930 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3931 - make sure to commitrw after any preprw to avoid deadlock (1162)
3932 - failing to execute a file in a lustre FS would lock inode (1203)
3933 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3934 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3935 - obd_uuid2tgt would walk off the end of the list (1255)
3936 - on IA64 the osc would give portals incorrect bulk size (1258)
3937 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3938 - fix lock inversion caused by new llite matching code (1282)
3939 - limit the number of dirty pages on a client to 10MB (1286)
3940 - timed out locks were not being corrected cancelled (1289)
3941 - fix O_DIRECT above 4GB on IA-32 (1292)
3942 * major user-visible changes
3943 - fail out/fail over policy now controlled by the upcall (993)
3945 - add OBD_PING to check server availability and failure (954)
3946 - lustre messages are now sent in sending host order (288, 340, 891)
3947 - add eadatalen to MDS getattr reply (340)
3948 - OST read replies may contain second buffer, with per-page status (593)
3950 2003-03-11 Phil Schwan <phil@clusterfs.com>
3953 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3954 - failed open()s could cause deadlock; fixed (867, 869)
3955 - stop cancelling OST locks when files are closed (481)
3956 - overlapping XID spaces caused network corruption (851, 853)
3957 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3958 - setattr_raw added, to do single-RPC, server-side setattrs
3959 - lmc/lconf syntax change for OST UUIDs
3960 - fix crashy race condition between ptlrpc_free_req and osc_close
3961 - don't use request in mdc_enqueue if we hit a timeout (889)
3962 - don't set the inode i_size for regular files from the MDS (896)
3963 - handle out of order completion AST (842)
3964 - don't LBUG if a lock request times out after receiving AST (913)
3965 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3966 - if a bad lock AST arrives, send an error instead of dropping entirely
3967 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3968 - fix leak in bulk IO when only partially completed (899, 900, 926)
3969 - fix O_DIRECT for ia64 (55)
3970 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3971 - C-z after timeout could hang a process forever; fixed (977)
3973 - client-side I/O cache (678, 924, 929, 941, 970)
3975 - READPAGE and SETATTRs which don't take server-side locks get
3978 2003-02-11 Phil Schwan <phil@clusterfs.com>
3981 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3982 - Fix fsfilt_extN_readpage() to read a full page of directory
3983 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3984 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3985 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3986 - Most intent operations were removed, in favour of a new RPC mode
3987 that does a single RPC to the server and bypasses most of the VFS
3988 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3989 - Aggressively cancel local locks on DLM servers
3990 - mds_reint_unlink sends EA to the client if it's the last nlink.
3991 client uses that EA to unlink OST objects.
3992 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3993 - recursive symlinks were fixed (439)
3994 - fixed NULL deref in DEBUG_REQ
3995 - filter_update_lastobjid no longer calls sync, which annoyed extN
3996 - fixed multi-client small-writes to a single file problem (445)
3997 - fixed mtime updates during file writes (607)
3998 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3999 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4000 - fixed Linux OST size reporting problem (444, 656)
4001 - OST now updates object mtime with writes or setattr (607, 619)
4002 - client verifies file size before zeroing page past EOF (445)
4003 - OST now writes last allocated objid to disk with allocation (108)
4004 - LOV on echo now works (409)
4006 - mds_reint_unlink sends a new buffer, with the EA included. this
4007 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4008 set if a regular file was being unlinked, and it was the last link
4009 - use PtlGet from the target for bulk writes (315)
4010 - OST now updates object mtime with writes or setattr (607, 619)
4011 - LDLM now has a grant-time callback to revalidate locked items, if
4013 - Many MDS operations were reorganized to combat race conditions
4015 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4016 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4018 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4021 - Fully reactivate OST imports after reconnection (512, others)
4022 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4023 - More graceful error handling for truncating on dead OST (515)
4024 - Don't error out unless we're actually accessing dead stripes (474)
4025 - Fix garbage sizes when stripes are missing (410)
4026 - LRU counters were broken, causing constant lock purge (433, 432)
4027 - garbage on read from stripes with failed OSTs (441)
4028 - mark OSCs as active before reconnecting during recovery (438)
4029 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4030 - lfind did not preserve OST order in output (443)
4031 - symlinks cause hung clients, incorrect data (439)
4032 - stop dereferencing request after dropping refcount (457)
4033 - don't LASSERT(spin_is_locked) on non-SMP (455)
4034 - fixes for many rename() bugs
4035 - fstat didn't correctly synchronize attributes (399)
4036 - server must handle lock cancellation during blocking AST prep (487)
4037 - bulk descriptors were free()d too soon (511)
4038 - fix paths in lconf, which would load incorrect modules (451, 507)
4039 - fix confusing lconf 'host not found' error message (386)
4040 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4041 - fix race condition in mdc_blocking_ast() for inode access (526)
4042 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4043 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4044 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4045 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4046 - fix lstripe in several ways (kernel side) (527)
4047 - fix request leak in ldlm_cli_enqueue (262)
4048 - incorrect OSC was marked inactive after OST failure
4049 - call mds_fs_cleanup before unmounting filesystem (524)
4050 - fix races between taking ns_lock and ldlm_lock_change_resource
4051 - fix races updating LOV export open file list
4052 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4053 - fix recovery NULL deref in ldlm_cli_cancel_unused
4054 - fix some DLM races by using new hash table for lock handles (419)
4055 - permit the client to specify desired inodes, at replay
4056 - duplicate requests when we queue them for replay reintegration
4057 - fix last_rcvd offset calculation
4058 - sync after each recovered transaction, so we always make progress
4059 - never, not always, ERESTART requests without transnos
4060 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4062 - skip replay if the MDS says that the client is already connected
4063 - don't check for a recovery-enabled export to match lctl's UUID
4064 - don't INC_USE_COUNT for phantom exports
4065 - don't crash when cleaning up phantom exports (567)
4066 - don't double-finish or set replay data for errored mdc_open requests
4067 - abort requests when they time out, so we don't get old replies
4068 - send/receive replies for AST messages again
4069 - if the client says that it doesn't have the lock, cancel it on the
4071 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4072 mark it as destroyed, it will all work out when decref is called
4073 - fix module use counts (22, 581)
4075 - ASTs now expect a reply (server cancels lock on error reply)
4077 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4080 - fix many simultaneous client startup (392)
4081 - fix dentry->d_it clobbering
4082 - credentials weren't being shipped for readdir/getattr operations
4083 - remove invalid assertions triggered during some concurrent MD
4085 - proper Lustre versions added (336, 389)
4086 - fix memory leak for create error case (398)
4087 - fix LOV locking bug that would get cli/srv out of sync
4088 - fix echo client over LOV (409)
4089 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4090 - fix double-O_EXCL intent crash (424)
4091 - avoid sending multiple lock CANCELs (352)
4093 - MDS can do multi-client recovery (modulo bugs in new code)
4095 - many updates, edits, cleanups
4097 2002-11-18 Phil Schwan <phil@clusterfs.com>
4100 - fix null d_it dereference (346)
4101 - fix full OST/dbench hang (333)
4102 - fix permission problem with file removal (286)
4103 - fix removal of OSCs from LOV when they fail
4104 - fix NULL deref during bulk timeout (214)
4105 - fix problems related to multiple filesystems on one MDS (241)
4106 - fixed serious subtle metadata locking bugs
4107 - free locks on clients when inodes are removed due to memory
4109 - fix inode pointer in lock data (285)
4110 - partial support for multiple MDS on a single host (241)
4111 - data locks weren't cancelled at clear_inode time (290, 311)
4112 - intent locks could lead to unbounded lock growth (205)
4113 - added a maximum lock count, an LRU list, and a flusher
4114 - fix multiple rename (365)
4115 - properly abstracted the echo client
4116 - OSC locked 1 byte too many; fixed
4117 - rewrote brw callback code:
4118 - fixed recovery bugs related to LOVs (306)
4119 - fixed too-many-pages-in-one-write crash (191)
4120 - fixed (again) crash in sync_io_timeout (214)
4121 - probably fixed callback-related race (385)
4123 - Add capability to MDS protocol
4124 - LDLM cancellations and callbacks on different portals
4126 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4129 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4130 - defer open object destruction to close time (601981, 138)
4131 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4132 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4133 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4134 - don't call obd_setattr in ll_file_release for destroyed objects
4135 * protocol change to lustre_msg: move |version| and add |flags|
4136 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4137 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4138 * added replay of create, unlink, link and rename operations during
4139 MDS failover; recovery should be much more robust now
4140 * remove failed OSCs from LOVs (only lov_create uses this so far)
4141 * the lustre-HOWTO was brought (more) up to date (582544)
4143 2002-10-23 Phil Schwan <phil@clusterfs.com>
4146 - in-use dentries weren't being reused properly (617851)
4147 - prevent multiple LDLM setup (599178)
4148 - fix LOV size calculations for truncate (617853)
4149 - fix client handling of MDS intent errors (POSIX)
4150 - fix permission bug in lovstripe.c test (624321)
4151 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4152 - truncate past end of file could corrupt data
4153 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4154 - a race in recovery could return ETIMEDOUT to apps (623947)
4155 - building outside the source directory was fixed
4156 * the lustre-HOWTO was brought (more) up to date (582544)
4157 * major progress was made on recovery functionality
4159 2002-10-10 Phil Schwan <phil@clusterfs.com>
4162 - recovery deadlock fix
4163 - rm -rf causes LBUG fix (617817)
4164 - file open by multiple tasks fix (618962)
4165 - directory permissions bugs (602707 and 620007)
4166 - journal_stop fixed with locking (611313)
4167 - O_APPEND failures resolved (618273, perhaps 614459)
4168 - lconf PATH fix (619770)
4169 - IA64 build fix (621450)
4170 - RPC buffer sizes scale with amount of memory
4172 2002-10-01 Phil Schwan <phil@clusterfs.com>
4175 - locks would be cancelled without throwing away data pages,
4176 resulting in inconsistent data (605627)
4177 - inode attributes were not always being refreshed (605627, 612449)
4178 - lconf now continues to cleanup after lctl reports an error
4179 - MDS now enforces user permissions (602707)
4180 - lprocfs cleanup fixed, but not yet enabled (614157)
4181 - fixed infinite server hang, should a client not respond to an AST
4182 - avoid going into recovery if user calls readlink() with a buffer
4183 that's too small (613941)
4184 - AST RPCs no longer require replies (614867) -- this may be changed
4185 - don't crash server if client sends an IOV that's too big (611336)
4186 - fixed lock conversion deadlock (611892)
4187 - fixed the following of symlinks (614622)
4188 * recovery: the server can remove locks from a client that dies, other
4189 clients can make progress
4190 * more extN patch fixes
4191 * compile-time configurable ptlrpc buffer allocations
4193 - collaborative read cache document
4194 - Lustre Lite Performance CDR document-in-progress
4196 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4199 - fix typo in patch-2.4.18
4201 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4204 - clear ptlrpc request each time in handle_incoming_request()
4205 - unlink of files now destroys the object on the OST
4207 2002-09-19 Peter Braam <braam@clusterfs.com>
4209 * add hard link support
4210 * change obdfile creation method
4211 * kernel patch changed
4213 2002-09-19 Peter Braam <braam@clusterfs.com>
4216 - stack overflow bug in extN fixed
4218 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4220 * documentation updates
4221 - add man pages for config tools
4222 - update tests/README to describe testing with new config tools
4223 - finish metadata API descriptions
4224 * bug fixes and cleanups
4225 - statfs workaround for 16TB limit
4226 - LOV stripe allocation improved, can stripe on subset of OSTs
4227 - LOV file size/IO offset was wrong for files > 4GB in size
4228 - object EA data was being dropped, caused files to be unreadable
4229 - memory overflow with non-LOV OST caused memory corruption
4230 - fixed regression tests to work with new config tools, obdfilter
4231 - fixed bug when directory size became larger than 1 block
4232 - fixed bug (for single client case) when PWD was deleted
4233 - invalidate local directory pages when doing intent-based ops
4234 - avoid LDLM oops when lock callback contained bad data
4236 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4238 * documentation updates
4239 * bug fixes and cleanups
4240 - configuration tools
4243 - 64-bit compile warnings
4244 - 64-bit internal statfs data
4246 * test_brw on persistent OST devices
4248 * lprocfs (disabled)
4250 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4252 * documentation updates
4253 * bug fixes and cleanups
4254 * configuration tools
4256 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4259 * many small fixes to 0_5_4
4260 * io/network handling
4261 * thinkos in MDS operations
4263 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4266 * crucial basic fixes to 0.5.3
4267 * IOR, Iozone work over Elan
4270 2002-08-07 Phil Schwan <phil@clusterfs.com>
4271 * version 0_5_3, our first alpha
4272 * we use the new Portals iovs
4273 * documentation updates
4274 * bug fixes and cleanups
4275 * small changes in the DLM wire protocol
4277 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4278 * version 0_5_1 with some initial stability,
4279 * locking on MD and file I/O.
4280 * documentation updates
4281 * several bug fixes since 0.5.0
4282 * small changes in wire protocol
4284 2002-07-18 Phil Schwan <phil@clusterfs.com>
4286 * delivered as Lustre Light Alpha
4287 * fixed a crash after handling invalid MDS requests
4288 * fixed directory pages for architectures with non-4k pages sizes
4290 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4291 * release version v0_4_4
4292 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4294 * Fixes a number of interruption problems with OST operations.
4295 * Update documentation for portals header changes
4296 * Move all wire protocol structs/defines to lustre_idl.h
4297 * Fixes symlink length bug.
4298 * Add tcpdump to repository.
4300 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4301 * release version v0_4_3
4302 * Fixes statfs for inodes on extN.
4303 * Fixes bug in runtests which would delete /etc/hosts.
4304 * Use 64-bit object IDs wherever possible (not into VFS though)
4305 Remove ost_get_info, which is unused by lustre, and out of date.
4307 2002-07-03 Peter Braam <braam@clusterfs.com>
4308 * release version v0_4_2 Fixes a lookup error (type not passed)
4309 * move forward to head of Portals
4310 * move forward to latest Lustre kernel
4312 2002-06-25 Peter Braam <braam@clusterfs.com>
4313 * release version v0_4_1. Hopefully stable on single node use.