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 2.6.18-8.1.15.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-cfs4
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Frequency : rare, at shutdown
18 Description: access already free / zero obd_namespace.
19 Details : if client_disconnect_export was called without force flag set,
20 and exist connect request in flight, this can produce access to
21 NULL pointer (or already free pointer) when connect_interpret
22 store ocd flags in obd_namespace.
26 Frequency : only at startup
27 Description: not alloc memory with spinlock held.
28 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
31 Severity : enhancement
33 Description: make lustre randomly fail allocating memory
34 Details : Make lustre randomly failed allocating memory for testing purpose.
36 Severity : enhancement
38 Description: lost problems with lov objid file
39 Details : Fixes some scability and access to not inited memory problems
40 in work with lov objdid file.
45 Description: lfs find does not continue on file error
46 Details : Continue other files processing when a file/dir is absent.
50 Description: Inconsistent usage of lustre_pack_reply()
51 Details : Standardize the usage of lustre_pack_reply() such that it
52 always generate a CERROR on failure.
55 Frequency : occasional
57 Description: MDS hang or stay in waiting lock
58 Details : If client receive lock with CBPENDING flag ldlm need send lock
59 cancel as separate rpc, to avoid situation when cancel request
60 can't processed due all i/o threads stay in wait lock.
64 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
68 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
72 Description: Update to SLES9 kernel-2.6.5-7.287.3.
76 Description: Update to RHEL4 latest kernel.
78 Severity : enhancement
80 Description: Build SLES10 patchless client fails
81 Details : The configure was broken by run ./configure with
82 --with-linux-obj=.... argument for patchless client. When the
83 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
84 can't search header adequately. Use absolute path such as
85 -I($LINUX)/include instead.
89 Description: interrupt oig_wait produce painc on resend.
90 Details : brw_redo_request can be used for resend requests from ptlrpcd and
91 private set, and this produce situation when rq_ptlrpcd_data not
92 copyed to new allocated request and triggered LBUG on assert
93 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
94 ptlrpcd set if request is changed and can be safety changed to use
99 Description: LASSERT_{REQ,REP}SWAB macros are buggy
100 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
101 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
102 these macros and replace them with their difinitions instead.
106 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
107 Details : Update which_patch & target file for SLES10 latest kernel.
110 Description: Remove client patches from SLES 10 kernel.
111 Details : This causes SLES 10 clients to behave as patchless clients
112 even on a Lustre-patched (server) kernel.
114 Severity : enhancement
116 Description: self-adjustable client's lru lists
117 Details : use adaptive algorithm for managing client cached locks lru
118 lists according to current server load, other client's work
119 pattern, memory activities, etc. Both, server and client
120 side namespaces provide number of proc tunables for controlling
123 Severity : enhancement
125 Description: light-weight GSS support
126 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
127 and reduce performance overhead.
129 Severity : enhancement
131 Description: Linux keyring support
132 Details : Support using service of Linux keyring for Lustre GSS internal
133 context refresh/cache mechanism.
137 Description: Fix errors in lfs documentation
138 Details : Fixes man pages
142 Description: don't use GFP_* in generic Lustre code.
143 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
147 Description: obdclass is limited by single OBD_ALLOC(idarray)
148 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
152 Description: warnings with build patchless client with vanila 2.6.19 and up
153 Details : change old ctl_table style and replace ctl_table/ctl_table_header
154 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
158 Frequency : Only for SLES
159 Description: sanity_quota fail test_1
160 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
161 file, which makes TSTID[2] inunique.
165 Frequency : Only for SLES9
166 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
167 Details : Update target/ChangeLog/which_patch .
171 Frequency : Only for test_10 in sanity-quota.sh
172 Description: a bug in quota test code
173 Details : checking a test flag in a wrong place
177 Frequency : Only for test_13 in sanity-quota.sh
178 Description: a bug in quota test script
179 Details : shouldn't check overfull allocated quota
183 Description: jbd statistics
184 Details : Port older jbd statistics patch for sles10
186 Severity : enhancement
188 Description: Add printing inode info into message about error in writepage.
193 Description: reply_lock_interpret crash due to race with it and lock cancel.
194 Details : Do not replay locks that are being cancelled. Do not reference
195 locks by their address during replay, just by their handle.
199 Frequency : When flocks are used.
200 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
201 Details : Flock locks might destroy just granted lock if it could be merged
202 with another existing flock, this is done in completion handler,
203 so teach ldlm_cli_enquque_fini that this is a valid case for
209 Description: Oops in read and write path when failing to allocate lock.
210 Details : Check if lock allocation failed and return error back.
214 Description: lstripe command fails for valid OST index
215 Details : The stripe offset is compared to lov->desc.ld_tgt_count
216 instead of lov->desc.ld_active_tgt_count.
220 Description: sanity.sh failed test 103
221 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
226 Description: df doesn't work properly if diskfs blocksize != 4K
227 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
231 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
232 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
236 Description: obd_health_check_timeout too short
237 Details : set obd_health_check_timeout as 1.5x of obd_timeout
241 Description: llapi_file_create() does not allow some changes
242 Details : add llapi_file_open() that allows specifying the mode and
243 open flags, and also returns an open file handle.
247 Description: merge and cleanup kernel patches.
248 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
252 Description: Add journal checksum support.(Kernel part)
253 Details : The journal checksum feature adds two new flags i.e
254 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
255 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
256 indicates that the commit block contains the checksum for
257 the blocks described by the descriptor blocks. Now commit
258 record can be sent to disk without waiting for descriptor
259 blocks to be written to disk. This behavior is controlled
260 using JBD2_FEATURE_ASYNC_COMMIT flag.
264 Description: OSS needs mutliple precreate threads
265 Details : Add ability to start more than one create thread per OSS.
269 Description: Sanity.sh test_65k failed: setstripe should have succeeded
270 Details : execute lfs setstripe on client
274 Description: mds_obd_create error creating tmp object
275 Details : When the user sets quota on root, llog will be affected and can't
276 create files and write files.
279 Frequency : Always on ia64 patchless client, and possibly others.
281 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
282 Details : This allows the patchless client to be loaded on architectures
287 Description: RedHat Update kernel for RHEL5
288 Details : Add the kernel config file for RHEL5.
292 Description: RedHat Update kernel for RHEL5
293 Details : Modify the kernel config file more closer RHEL5.
297 Description: Build failure against Centos5 (RHEL5)
298 Details : Define PAGE_SIZE when it isn't present.
302 Description: test_7 on sanity-quota.sh can't be used on separate machines
303 Details : add facet to handle it
307 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
308 Details : with NFS, the anon dentry's parent was set to itself in
309 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
310 corresponding dentry other than use rec->ur_name.
312 Severity : enhancement
314 Description: lfs setstripe enhancement
315 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
319 Description: enable data checksumming by default
320 Details : enable checksum by default, allow --disable-checksum
321 configure option and "-o nochecksum" mount option. Checksums
322 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
326 Description: lustre support for RHEL5
327 Details : Add support for RHEL5.
331 Description: Client eviction due to failover config
332 Details : after a connection loss, the lustre client should attempt to
333 reconnect to the last active server first before trying the
334 other potential connections.
338 Description: when mds and osts use different quota unit(32bit and 64bit),
339 quota will be released repeatly.
340 Details : void sending multiple quota reqs to mds, which will keep the
341 status between the reqs.
345 Description: osts not allocated evenly to files
346 Details : change the condition to increase offset_idx
349 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
351 Description: Data corruption for OSTs that are formatted larger than 2TB
353 Details : When generating the bio request for lustre file writes the
354 sector number would overflow a temporary variable before being
355 used for the IO. The data reads correctly from Lustre (which
356 will overflow in a similar manner) but other file data or
357 filesystem metadata may be corrupted in some cases.
361 Description: Tune the kernel for good SCSI performance.
362 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
363 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
368 Description: rewrite ext2-derived code in obdclass/uuid.c
369 Details : rewrite inherited code (uuid parsing code from ext2 utils)
370 from scratch preserving functionality.
374 Description: "lfs find -obd UUID" prints directories
375 Details : "lfs find -obd UUID" will return all directory names instead
376 of just file names. It is incorrect because the directories
377 do not reside on the OSTs.
381 Description: use i_size_read and i_size_write in 2.6 port
382 Details : replace inode->i_size access with i_size_read/write()
385 Frequency : when removing large files
387 Description: scheduling issue during removal of large Lustre files
388 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
389 It causes scheduling issues when removing large files (17TB in the
393 Frequency : only with liblustre clients on XT3
395 Description: evictions taking too long
396 Details : allow llrd to evict clients directly on OSTs
399 Frequency : only on ppc
401 Description: /proc/fs/lustre/devices broken on ppc
402 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
403 We should make sure the type of 'index' is loff_t and then cast
404 explicitly as needed below. Do not assign an explicitly cast
408 Frequency : only for rhel5
410 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
411 Details : Modify the target file & which_kernel.
413 Severity : enhancement
415 Description: omit set fsid for export NFS
416 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
417 as FS_REQUIRES_DEV for avoid problems with generate fsid.
421 Frequency : Always, for kernels after 2.6.16
422 Description: Fix warning idr_remove called for id=.. which is not allocated.
423 Details : Last kernels save old s_dev before kill super and not allow
424 to restore from callback - restore it before call kill_anon_super.
428 Description: buffer overruns could theoretically occur
429 Details : llapi_semantic_traverse() modifies the "path" argument by
430 appending values to the end of the origin string, and a
431 overrun may occur. Adding buffer overrun check in liblustreapi.
435 Description: Fix error on 'ls .' at the top of the Lustre mount.
436 Details : Don't revalidate dentry if it is a root dentry.
440 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
441 Details : Modify vm-tunables-rhel4.patch.
445 Description: Kernel config for 2.6.18-vanilla.
446 Details : Modify targets/2.6-vanilla.target.in.
447 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
448 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
449 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
450 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
454 Description: improve handling recoverable errors
455 Details : if request processig with error which can be recoverable on server
456 request should be resend, otherwise page released from cache and
461 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
462 Details : Modify target file & which_patch.
463 A flaw was found in the IA32 system call emulation provided
464 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
465 value could be stored in the %RAX register, which could trigger an
466 out-of-bounds system call table access. An untrusted local user
467 could exploit this flaw to run code in the kernel
468 (ie a root privilege escalation). (CVE-2007-4573).
472 Description: O_DIRECT bypasses client statistics.
473 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
474 read_ahead_stats were updated. copied the stats section from
475 osc_send_oap_rpc() into async_internal().
479 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
483 Description: change order of libsysio includes
484 Details : '#include sysio.h' should always come before '#include xtio.h'
488 Description: handle "serious error: objid * already exists" more gracefully
489 Details : If LAST_ID value on disk is smaller than the objects existing in
490 the O/0/d* directories, it indicates disk corruption and causes an
491 LBUG(). If the object is 0-length, then we should use the existing
492 object. This will help to avoid a full fsck in most cases.
494 Severity : enhancement
496 Description: adapt the lustre_config script to support the upgrade case
497 Details : Add "-u" option for lustre_config script to support upgrading 1.4
498 server targets to 1.6 in parallel.
502 Description: To avoid grant space > avaible space when the disk is almost
503 full. Without this patch you might see the error "grant XXXX >
504 available" or some LBUG about grant, when the disk is almost
506 Details : In filter_check_grant, for non_grant cache write, we should
507 check the left space by if (*left > ungranted + bytes), instead
508 of (*left > ungranted), because only we are sure the left space
509 is enough for another "bytes", then the ungrant space should be
510 increase. In client, we should update cl_avail_grant only there
511 is OBD_MD_FLGRANT in the reply.
515 Description: Update RHEL 4 kernel to fix local root privilege escalation.
516 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
517 described in CVE-2007-4573. This problem could allow untrusted
518 local users to gain root access.
521 Frequency : when using O_DIRECT and quotas
523 Description: Incorrect file ownership on O_DIRECT output files
524 Details : block usage reported by 'lfs quota' does not take into account
525 files that have been written with O_DIRECT.
530 Description: touch file failed when fs is not full
531 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
532 otherwise we can get ENOSP while fs is not full.
536 Description: parallel lock callbacks
537 Details : Instead of sending blocking and completion callbacks as separated
538 requests, adding them to a set and sending in parallel.
541 Frequency : only for Cray XT3
542 Bugzilla : 12829/13455
543 Description: Changing primary group doesn't change the group lustre assigns to
545 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
546 overridden with the primary group provided by the group upcall,
547 whereas the supplied fsgid can be trusted if it is in the list of
548 supplementary groups returned by the group upcall.
550 --------------------------------------------------------------------------------
552 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
554 * Support for kernels:
555 2.6.5-7.283 (SLES 9),
556 2.6.9-55.EL (RHEL 4),
557 2.6.16.46-0.14 (SLES 10),
558 2.6.18.8 vanilla (kernel.org)
559 * Client support for unpatched kernels:
560 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
561 2.6.16 - 2.6.22 vanilla (kernel.org)
562 * Due to recently discovered recovery problems, we do not recommend
563 using patchless RHEL 4 clients with this or any earlier release.
564 * Recommended e2fsprogs version: 1.39.cfs8
565 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
566 * Starting with this release, the ldiskfs backing filesystem required
567 by Lustre is now in its own package, lustre-ldiskfs. This package
568 should be installed. It is versioned separately from Lustre and
569 may be released separately in future.
573 Description: block reactivating mgc import until all deactivates complete
574 Details : Fix race when failing back MDT/MGS to itself (testing)
577 Frequency : at statup only
579 Description: mds_lov_synchronize race leads to various problems
580 Details : simultaneous MDT->OST connections at startup can cause the
581 sync to abort, leaving the OSC in a bad state.
583 Severity : enhancement
585 Description: add optional extra BUILD_VERSION info
586 Details : add a new environment variable (namely LUSTRE_VERS) which allows
587 to override the lustre version.
590 Frequency : 2.6.18 servers only
592 Description: ll_kern_mount() doesn't release the module reference
593 Details : The ldiskfs module reference count never drops down to 0
594 because ll_kern_mount() doesn't release the module reference.
599 Description: server LBUG when using old ost_num_threads parameter
600 Details : Accept the old ost_num_threads parameter but warn that it
601 is deprecated, and fix an off-by-one error that caused an LBUG.
606 Description: Transient SCSI error results in persistent IO issue
607 Details : iobuf->dr_error is not reinitialized to 0 between two
611 Frequency : sometimes when underlying device returns I/O errors
613 Description: OSTs not going read-only during write failures
614 Details : OSTs are not remounted read-only when the journal commit threads
615 get I/O errors because fsfilt_ext3 calls journal_start/stop()
616 instead of the ext3 wrappers.
620 Description: poor connect scaling with increasing client count
621 Details : Don't run filter_grant_sanity_check for more than 100 exports
622 to improve scaling for large numbers of clients.
625 Frequency : SLES10 only
627 Description: sanity-quota.sh quotacheck failed: rc = -22
628 Details : Quotas cannot be enabled on SLES10.
631 Frequency : liblustre clients only
633 Description: getdirentries does not give error when run on compute nodes
634 Details : getdirentries does not fail when the size specified as an argument
635 is too small to contain at least one entry
637 Severity : enhancement
639 Description: Add LNET router traceability for debug purposes
640 Details : If a checksum failure occurs with a router as part of the
641 IO path, the NID of the last router that forwarded the bulk data
642 is printed so it can be identified.
647 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
648 Details : Due to a race condition, liblustre clients were occasionally
651 Severity : enhancement
653 Description: lfs setstripe use optional parameters instead of postional
656 Severity : enhancement
658 Description: Nanosecond timestamp support for ldiskfs
659 Details : The on-disk ldiskfs filesystem has added support for nanosecond
660 resolution timestamps. There is not yet support for this at
661 the Lustre filesystem level.
664 Frequency : during server recovery
666 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
667 Details : request with rq_no_resend flag not awake l_wait_event if they get a
671 Frequency : nfs export on patchless client
673 Description: connectathon hang when test nfs export over patchless client
674 Details : Disconnected dentry cannot be found with lookup, so we do not need
675 to unhash it or make it invalid
678 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
679 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
680 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
685 Description: open req refcounting wrong on reconnect
686 Details : If reconnect happened between getting open reply from server and
687 call to mdc_set_replay_data in ll_file_open, we will schedule
688 replay for unreferenced request that we are about to free.
689 Subsequent close will crash in variety of ways.
690 Check that request is still eligible for replay in
691 mdc_set_replay_data().
696 Description: disable writes to filesystem when reading health_check file
697 Details : the default for reading the health_check proc file has changed
698 to NOT do a journal transaction and write to disk, because this
699 can cause reads of the /proc file to hang and block HA state
700 checking on a healthy but otherwise heavily loaded system. It
701 is possible to return to the previous behaviour during configure
702 with --enable-health-write.
704 Severity : enhancement
706 Description: 64-bit inode version
707 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
708 to the inode. This will be required for version-based recovery.
713 Description: MDS fails to start if a duplicate client export is detected
714 Details : in some rare cases it was possible for a client to connect to
715 an MDS multiple times. Upon recovery the MDS would detect this
716 and fail during startup. Handle this more gracefully.
718 Severity : enhancement
720 Description: Add -o localflock option to simulate old noflock
722 Details : This will achieve local-only flock/fcntl locks
728 Description: log_commit_thread vs filter_destroy race leads to crash
729 Details : Take import reference before releasing llog record semaphore
734 Description: Wrong request locking in request set processing
735 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
736 imp_delayed_list, in this place should be used imp_lock.
739 Frequency : when reconnection
741 Description: Grant Leak when osc reconnect to OST
742 Details : When osc reconnect ost, OST(filter) should check whether it
743 should grant more space to client by comparing fed_grant and
744 cl_avail_grant, and return the granted space to client instead
745 of "new granted" space, because client will call osc_init_grant
746 to update the client grant space info.
749 Frequency : when client reconnect to OST
751 Description: Grant Leak when osc do resend and replay bulk write
752 Details : When osc reconnect to OST, OST(filter)should clear grant info of
753 bulk write request, because the grant info while be sync between
754 OSC and OST when reconnect, and we should ignore the grant info
755 these of resend/replay write req.
760 Description: Grant space more than avaiable left space sometimes.
761 Details : When then OST is about to be full, if two bulk writing from
762 different clients came to OST. Accord the avaliable space of the
763 OST, the first req should be permitted, and the second one
764 should be denied by ENOSPC. But if the seconde arrived before
765 the first one is commited. The OST might wrongly permit second
766 writing, which will cause grant space > avaiable space.
769 Frequency : when client is evicted
771 Description: Grant might be wrongly erased when osc is evicted by OST
772 Details : when the import is evicted by server, it will fork another
773 thread ptlrpc_invalidate_import_thread to invalidate the
774 import, where the grant will be set to 0. While the original
775 thread will update the grant it got when connecting. So if
776 the former happened latter, the grant will be wrongly errased
777 because of this race.
782 Description: Checking Stale with correct fid
783 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
784 to check whether it is stale, because sometimes, we want the
785 enqueue happened anyway, and op_data.fid2 will not be initialized.
787 Severity : enhancement
789 Description: update patchless client
790 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
793 Frequency : only with 2.4 kernel
795 Description: random memory corruption
796 Details : size of struct ll_inode_info is to big for union inode.u and this
797 can be cause of random memory corruption.
802 Description: Memory leak in recovery
803 Details : Lov_mds_md was not free in an error handler in mds_create_object.
804 It should also check obd_fail before fsfilt_start, otherwise if
805 fsfilt_start return -EROFS,(failover mds during mds recovery).
806 then the req will return with repmsg->transno = 0 and rc = EROFS.
807 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
808 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
809 be freed no matter whether fsfilt_commit success or not.
812 Frequency : only with huge count clients
814 Description: Prevents from taking the superblock lock in llap_from_page for
816 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
817 for llap_from_page call in ll_removepage prevents from taking the
818 superblock lock for a soon died page.
823 Description: Not check open intent error before release open handle
824 Details : in some rare cases, the open intent error is not checked before
825 release open handle, which may cause
826 ASSERTION(open_req->rq_transno != 0), because it tries to release
827 the failed open handle.
832 Description: Set cat log bitmap only after create log success.
833 Details : in some rare cases, the cat log bitmap is set too early. and it
834 should be set only after create log success.
838 Description: Accessing a block bevice can re-enable I/O when Lustre is
839 tearing down a device.
840 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
845 Description: service threads may hog cpus when there are a lot of requests
846 Details : Insert cond_resched to give other threads a chance to use some CPU
851 Description: the cat log was not initialized in recovery
852 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
853 unlink log on mds will not be initialized until mds post
854 recovery. And also in mds post recovery, the unlink log will
855 initialization will be done asynchronausly, so there will be race
856 between add unlink log and unlink log initialization.
860 Description: brw_stats were being printed incorrectly
861 Details : brw_stats were being printed as log2 but all of them were not
862 recorded as log2. Also remove some code duplication arising from
863 filter_tally_{read,write}.
867 Frequency : rare, only in recovery.
868 Description: ASSERTION(req->rq_type != LI_POISON) failed
869 Details : imp_lock should be held while iterating over imp_sending_list for
870 prevent destroy request after get timeout in ptlrpc_queue_wait.
874 Description: replay-single.sh test 52 fails
875 Details : A lock's skiplist need to be cleanup when it being unlinked
876 from its resource list.
880 Description: Short directio read returns full requested size rather than
882 Details : Direct I/O operations should return actual amount of bytes
883 transferred rather than requested size.
885 Severity : enhancement
887 Description: metadata RPC reduction (e.g. for rm performance)
888 Details : decrease the amount of synchronous RPC between clients and servers
889 by canceling conflicing lock before the operation on the client
890 and packing thier handles into the main operation RPC to server.
892 Severity : enhancement
894 Description: Async OSC create to avoid the blocking unnecessarily.
895 Details : If a OST has no remain object, system will block on the creating
896 when need to create a new object on this OST. Now, ways use
897 pre-created objects when available, instead of blocking on an
898 empty osc while others are not empty. If we must block, we block
899 for the shortest possible period of time.
901 Severity : enhancement
903 Description: refine locking for avoid write wrong info into lov_objid file
904 Details : fix possible races with add new target and write/update data in
907 --------------------------------------------------------------------------------
909 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
914 Frequency : on some architectures
916 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
917 Details : Uninitialized flags sometimes cause configuration commands to
921 Frequency : patchless clients only
923 Description: missing __iget() symbol export
924 Details : The __iget() symbol export is missing. To avoid the need for
925 this on patchless clients the deathrow inode reaper is turned
926 off, and we depend on the VM to clean up old inodes. This
927 dependency was during via the fix for bug 12181.
929 --------------------------------------------------------------------------------
931 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
933 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
934 INTEROPERATE with older versions automatically. In many cases a
935 special upgrade step is needed. Please read the
936 user documentation before upgrading any part of a live system.
937 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
938 version will not interoperate with 1.6 betas before beta5 (1.5.95).
939 * WARNING: Lustre configuration and startup changes are required with
940 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
946 Severity : enhancement
948 Description: MountConf
949 Details : Lustre configuration is now managed via mkfs and mount
950 commands instead of lmc and lconf. New obd types (MGS, MGC)
951 are added for dynamic configuration management. See
952 https://mail.clusterfs.com/wikis/lustre/MountConf for
955 Severity : enhancement
957 Description: dynamic OST addition
958 Details : OSTs can now be added to a live filesystem
960 Severity : enhancement
962 Description: startup order invariance
963 Details : MDTs and OSTs can be started in any order. Clients only
964 require the MDT to complete startup.
966 Severity : enhancement
968 Description: parallel, asynchronous orphan cleanup
969 Details : orphan cleanup is now performed in separate threads for each
970 OST, allowing parallel non-blocking operation.
972 Severity : enhancement
974 Description: optimized stripe assignment
975 Details : stripe assignments are now made based on ost space available,
976 ost previous usage, and OSS previous usage, in order to try
977 to optimize storage space and networking resources.
979 Severity : enhancement
981 Description: Permanently set tunables
982 Details : All writable /proc/fs/lustre tunables can now be permanently
983 set on a per-server basis, at mkfs time or on a live system.
985 Severity : enhancement
987 Description: Lustre message v2
988 Details : Add lustre message format v2.
990 Severity : enhancement
992 Description: client OST exclusion list
993 Details : Clients can be started with a list of OSTs that should be
994 declared "inactive" for known non-responsive OSTs.
998 Description: SPEC SFS validation failure on NFS v2 over lustre.
999 Details : Changes the blocksize for regular files to be 2x RPC size,
1000 and not depend on stripe size.
1002 Severity : enhancement
1004 Description: Multiple MD RPCs in flight.
1005 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1006 To avoid overly-overloading MDS, introduce a limit on number of
1007 MDS RPCs in flight for a single client and add /proc controls
1008 to adjust this limit.
1010 Severity : enhancement
1012 Description: client read/write statistics
1013 Details : Add client read/write call usage stats for performance
1014 analysis of user processes.
1015 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1016 file access. extents_stats shows chunk size distribution.
1017 extents_stats_per_process show chunk size distribution per
1020 Severity : enhancement
1022 Description: mds statistics
1023 Details : Add detailed mds operations statistics in
1024 /proc/fs/lustre/mds/*/stats.
1028 Description: Failure of copying files with lustre special EAs.
1029 Details : Client side always return success for setxattr call for lustre
1030 special xattr (currently only "trusted.lov").
1035 Description: Refcount LNET uuids
1036 Details : The global LNET uuid list grew linearly with every startup;
1037 refcount repeated list entries instead of always adding to
1040 Severity : enhancement
1042 Description: Dynamic service threads
1043 Details : Within a small range, start extra service threads
1044 automatically when the request queue builds up.
1047 Frequency : mixed-endian client/server environments
1049 Description: mixed-endian crashes
1050 Details : The new msg_v2 system had some failures in mixed-endian
1053 Severity : enhancement
1055 Description: Easy OST removal
1056 Details : OSTs can be permanently deactivated with e.g. 'lctl
1057 conf_param lustre-OST0001.osc.active=0'
1059 Severity : enhancement
1061 Description: MGS proc entries
1062 Details : Added basic proc entries for the MGS showing what filesystems
1065 Severity : enhancement
1067 Description: provide MGS failover
1068 Details : Added config lock reacquisition after MGS server failover.
1070 Severity : enhancement
1072 Description: add Linux 2.4 support
1073 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1077 Description: a large application tries to do I/O to the same resource and dies
1078 in the middle of it.
1079 Details : Check the req->rq_arrival time after the call to
1080 ost_brw_lock_get(), but before we do anything about
1081 processing it & sending the BULK transfer request. This
1082 should help move old stale pending locks off the queue as
1083 quickly as obd_timeout.
1086 Frequency : when an incorrect nid is specified during startup
1088 Description: ptlrpc connect to non-existant node causes kernel crash
1089 Details : LNET can't be re-entered from an event callback, which
1090 happened when we expire a message after the export has been
1091 cleaned up. Instead, hand the zombie cleanup off to another
1094 Severity : enhancement
1096 Description: plain/inodebits lock performance improvement
1097 Details : Grouping plain/inodebits in granted list by their request modes
1098 and bits policy, thus improving the performance of search through
1102 Frequency : only if OST filesystem is corrupted
1104 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1105 Details : for a short time RPCs with bulk IO are in the replay list,
1106 but replay of bulk IOs is unimplemented. If the OST filesystem
1107 is corrupted due to disk cache incoherency and then replay is
1108 started it is possible to trip an assertion. Avoid putting
1109 committed RPCs into the replay list at all to avoid this issue.
1112 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1114 Description: System hang on startup
1115 Details : This bug allowed the liblustre (e.g. catamount) client to
1116 return to the app before handling all startup RPCs. This
1117 could leave the node unresponsive to lustre network traffic
1118 and manifested as a server ptllnd timeout.
1120 Severity : enhancement
1122 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1123 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1124 boolean to control whether to print peer debug info when a
1125 client's RPC times out.
1128 Frequency : only for kernels with patches from Lustre below 1.4.3
1130 Description: Remove old rdonly API
1131 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1134 Frequency : only for devices with external journals
1136 Description: Set external device read-only also
1137 Details : During a commanded failover stop, we set the disk device
1138 read-only while the server shuts down. We now also set any
1139 external journal device read-only at the same time.
1142 Frequency : when upgrading from 1.4 while trying to change parameters
1144 Description: The wrong (new) MDC name was used when setting parameters for
1145 upgraded MDT's. Also allows changing of OSC (and MDC)
1146 parameters if --writeconf is specified at tunefs upgrade time.
1149 Frequency : when setting specific ost indicies
1151 Description: QOS code breaks on skipped indicies
1152 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1153 created with --index need not be sequential.
1155 Severity : enhancement
1157 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1158 Details : The uninit_groups feature works in conjunction with the kernel
1159 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1160 up the pass1 processing of e2fsck. This is a read-only feature
1161 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1162 mount filesystems that have had this feature enabled.
1164 Severity : enhancement
1166 Description: Improve multi-block allocation algorithm to avoid fragmentation
1167 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1168 allocation locality and avoid filesystem fragmentation.
1170 ------------------------------------------------------------------------------
1172 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1174 * Support for kernels:
1175 2.4.21-47.0.1.EL (RHEL 3)
1176 2.6.5-7.283 (SLES 9)
1177 2.6.9-42.0.10.EL (RHEL 4)
1178 2.6.12.6 vanilla (kernel.org)
1179 2.6.16.27-0.9 (SLES 10)
1180 * Recommended e2fsprogs version: 1.39.cfs5
1182 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1186 Frequency : occasional, depends on client load and configuration
1187 Bugzilla : 12181, 12203
1188 Description: data loss for recently-modified files
1190 Details : In some cases it is possible that recently written or created
1191 files may not be written to disk in a timely manner (this should
1192 normally be within 30s unless client IO load is very high).
1193 The problem appears as zero-length files or files that are a
1194 multiple of 1MB in size after a client crash or client eviction
1195 that are missing data at the end of the file.
1197 This problem is more likely to be hit on clients where files are
1198 repeatedly created and unlinked in the same directory, clients
1199 have a large amount of RAM, have many CPUs, the filesystem has
1200 many OSTs, the clients are rebooted frequently, and/or the files
1201 are not accessed by other nodes after being written.
1203 The presence of the problem can be detected by looking at
1204 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1205 smaller than the second (nr_unused) then dirty files will not
1206 be flushed automatically to disk. "sync; sleep 10" should be
1207 run several times on the node before unmounting it to update
1208 Lustre (this is also safe to run on nodes without this problem).
1210 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1211 that can cause this same problem, so customers using that kernel
1212 also need to update the kernel in addition to Lustre. In order
1213 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1215 It is normal that files written just before a client crash (less
1216 than 30s) may not yet have been flushed to disk, even for local
1220 Frequency : frequent on thin XT3 nodes
1222 Description: UUID collision on thin XT3 Linux nodes
1223 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1224 randomly, since we relied on an insufficiently-seeded PRNG.
1229 Description: OSS hangs after "All ost request buffers busy"
1230 Details : A deadlock between quota and journal operations caused OSS
1231 hangs after printing "All ost request buffers busy."
1234 Frequency : always on liblustre builds
1236 Description: Cleanup compiler warnings on liblustre
1239 Frequency : always on liblustre builds on XT3
1241 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1245 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1246 > 32000 subdirectories
1247 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1248 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1249 subdirectory count crosses 32000. This will aid e2fsck to
1250 correctly handle more than 32000 subdirectories.
1253 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1255 Description: System hang on startup
1256 Details : This bug allowed the liblustre (e.g. catamount) client to
1257 return to the app before handling all startup RPCs. This
1258 could leave the node unresponsive to lustre network traffic
1259 and manifested as a server ptllnd timeout.
1261 Severity : enhancement
1263 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1264 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1265 boolean to control whether to print peer debug info when a
1266 client's RPC times out.
1271 Description: make O_SYNC working on 2.6 kernels
1272 Details : 2.6 kernels use different method for mark pages for write,
1273 so need add a code to lustre for O_SYNC work.
1278 Description: Failure to close file and release space on NFS
1279 Details : Put inode details into lock acquired in ll_intent_file_open.
1280 Use mdc_intent_lock in ll_intent_open to properly
1281 detect all kind of errors unhandled by mdc_enqueue.
1286 Description: proc file read during shutdown sometimes raced obd removal,
1288 Details : Add lock to prevent obd access after proc file removal.
1291 Frequency : Only for files larger than 4GB on 32-bit clients.
1293 Description: improperly doing page alignment of locks
1294 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1295 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1300 Description: RPCs being resent when they shouldn't be
1301 Details : Some RPCs that should not be resent are being resent. This
1302 can cause inconsistencies in the RPC state machine. Do not
1303 resend such requests.
1306 Frequency : rare, only with NFS export
1308 Description: Crash on NFS re-export node
1309 Details : under very unusual load conditions an assertion is hit in
1310 ll_intent_file_open()
1313 Frequency : only if OST filesystem is corrupted
1315 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1316 Details : for a short time RPCs with bulk IO are in the replay list,
1317 but replay of bulk IOs is unimplemented. If the OST filesystem
1318 is corrupted due to disk cache incoherency and then replay is
1319 started it is possible to trip an assertion. Avoid putting
1320 committed RPCs into the replay list at all to avoid this issue.
1325 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1326 Details : Large single O_DIRECT read and write calls can fail to allocate
1327 a sufficiently large buffer to process the request. In case of
1328 allocation failure the allocation is retried with a smaller
1329 buffer and broken into smaller requests.
1331 Severity : enhancement
1333 Description: Add -o localflock option to simulate old noflock behaviour.
1334 Details : This will achieve local-only flock/fcntl locks coherentness.
1339 Description: versioning check is incomplete
1340 Details : Checking the version difference of client vs. server, report
1341 error if the gap is too big.
1346 Description: add support PG_writeback bit
1347 Details : add support for PG_writeback bit for Lustre, for more carefull
1348 work with page cache in 2.6 kernel. This also fix some deadlocks
1349 and remove hack for work O_SYNC with 2.6 kernel.
1351 Severity : enhancement
1353 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1354 Details : The uninit_groups feature works in conjunction with the kernel
1355 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1356 up the pass1 processing of e2fsck. This is a read-only feature
1357 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1358 mount filesystems that have had this feature enabled.
1360 Severity : enhancement
1362 Description: Improve multi-block allocation algorithm to avoid fragmentation
1363 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1364 allocation locality and avoid filesystem fragmentation.
1366 ------------------------------------------------------------------------------
1368 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1370 * Support for kernels:
1371 2.6.9-42.0.3.EL (RHEL 4)
1372 2.6.5-7.276 (SLES 9)
1373 2.4.21-47.0.1.EL (RHEL 3)
1374 2.6.12.6 vanilla (kernel.org)
1375 2.6.16.21-0.8 (SLES10)
1376 * Recommended e2fsprogs version: 1.39.cfs2-0
1378 * The backwards-compatible /proc/sys/portals symlink has been removed
1379 in this release. Before upgrading, please ensure that you change
1380 any configuration scripts or /etc/sysctl.conf files that access
1381 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1382 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1383 in advance of the upgrade on any system running Lustre 1.4.6 or
1384 newer, since /proc/sys/lnet was added in that version.
1385 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1389 Frequency : only when quota is used
1391 Description: avoid scanning export list for quota master
1392 Details : Change the algorithms to avoid scanning export list in order
1393 to improve the efficiency.
1396 Frequency : MDS failover only, very rarely
1398 Description: "went back in time" messages on mds failover
1399 Details : The greatest transno may be lost when the current operation
1400 finishes with an error (transno==0) and the client's last_rcvd
1401 record is over-written. Save the greatest transno in the
1402 mds_last_transno for this case.
1405 Frequency : always for specific kernels and striping counts
1407 Description: client may get "Matching packet too big" without ACL support
1408 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1409 when trying to access files in certain configurations. The
1410 clients should in fact be denied when mounting because they do
1411 not understand ACLs.
1414 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1416 Description: many clients connecting with IO in progress causes connect timeouts
1417 Details : Avoid synchronous journal commits to avoid delays caused by many
1418 clients connecting/disconnecting when bulk IO is in progress.
1419 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1420 OST_IO_PORTAL to avoid delays behind potentially many pending
1424 Frequency : occasionally with multiple writers to a single file
1426 Description: shared writes to file may result in wrong size reported by stat()
1427 Details : Allow growing of kms when extent lock is cancelled
1430 Frequency : always with random mmap IO to multi-striped file
1432 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1433 Details : Only if the hole is at the end of OST object so that kms is too
1434 small. Fix is to increase kms accordingly in ll_nopage.
1437 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1439 Description: writes to a missing object would leak memory on the OST
1440 Details : If there is an inconsistency between the MDS and OST filesystems,
1441 such that the MDS references an object that doesn't exist, writes
1442 to that object will leak memory due to incorrect cleanup in the
1443 error handling path, eventually running out of memory on the OST.
1448 Description: Creating too long symlink causes lustre errors
1449 Details : Check symlink and name lengths before sending requests to MDS.
1452 Frequency : only if flock is enabled (not on by default)
1454 Description: posix locks not released on fd closure on 2.6.9+
1455 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1456 kernels, this caused such locks not to be released on fd close and
1457 then assertions on fs unmount about still used locks.
1460 Frequency : MDS failover only, very rarely
1462 Description: clients may get ASSERTION(granted_lock != NULL)
1463 Details : When request was taking a long time, and a client was resending
1464 a getattr by name lock request. The were multiple lock requests
1465 with the same client lock handle and
1466 mds_getattr_name->fixup_handle_for_resent_request found one of the
1467 lock handles but later failed with ASSERTION(granted_lock != NULL).
1472 Description: handle->h_buffer_credits > 0, assertion failure
1473 Details : h_buffer_credits is zero after truncate, causing assertion
1474 failure. This patch extends the transaction or creates a new
1478 Frequency : NFS re-export or patchless client
1479 Bugzilla : 11179, 10796
1480 Description: Crash on NFS re-export node (__d_move)
1481 Details : We do not want to hash the dentry if we don't have a lock.
1482 But if this dentry is later used in d_move, we'd hit uninitialised
1483 list head d_hash, so we just do this to init d_hash field but
1484 leave dentry unhashed.
1487 Frequency : NFS re-export or patchless client
1489 Description: NFS exports has problem with symbolic link
1490 Details : lustre client didn't properly install dentry when re-exported
1491 to NFS or running patchless client.
1494 Frequency : NFS re-export or patchless client
1496 Description: Various nfs/patchless fixes.
1497 Details : fixes reuse disconected alias for lookup process - this fixes
1498 warning "find_exported_dentry: npd != pd",
1499 fix permission error with open files at nfs.
1500 fix apply umask when do revalidate.
1503 Frequency : occasional
1505 Description: Crash on NFS re-export node
1506 Details : calling clear_page() on the wrong pointer triggered oops in
1507 generic_mapping_read().
1510 Frequency : rarely, using O_DIRECT IO
1512 Description: unaligned directio crashes client with LASSERT
1513 Details : check for unaligned buffers before trying any requests.
1516 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1518 Description: stale data returned from RAID cache
1519 Details : If only a small amount of IO is done to the RAID device before
1520 reading it again it is possible to get stale data from the RAID
1521 cache instead of reading it from disk.
1524 Frequency : always for sles10 kernel
1526 Description: sles10 support
1527 Details : ll_follow_link: compile fixes and using of nd_set_link
1528 under newer kernels.
1531 Frequency : depends on arch, kernel and compiler version, always on sles10
1534 Description: recursive or deep enough symlinks cause stack overflow
1535 Details : getting rid of large stack-allocated variable in
1539 Frequency : depends on hardware
1541 Description: lustre write performance loss in the SLES10 kernel
1542 Details : the performance loss is caused by using of write barriers in the
1543 ext3 code. The SLES10 kernel turns barrier support on by
1544 default. The fix is to undo that change for ldiskfs.
1546 ------------------------------------------------------------------------------
1548 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1550 * Support for kernels:
1551 2.6.9-42.0.3EL (RHEL 4)
1552 2.6.5-7.276 (SLES 9)
1553 2.4.21-47.0.1.EL (RHEL 3)
1554 2.6.12.6 vanilla (kernel.org)
1558 Frequency : quota enabled and large files being deleted
1560 Description: releasing more than 4GB of quota at once hangs OST
1561 Details : If a user deletes more than 4GB of files on a single OST it
1562 will cause the OST to spin in an infinite loop. Release
1563 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1568 Description: statfs data retrieved from /proc may be stale or zero
1569 Details : When reading per-device statfs data from /proc, in the
1570 {kbytes,files}_{total,free,avail} files, it may appear
1571 as zero or be out of date.
1574 Frequency : systems with MD RAID1 external journal devices
1576 Description: lconf's call to blkid is confused by RAID1 journal devices
1577 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1578 of returning all block devices that match the journal UUID.
1581 Frequency : always, for aggregate stripe size over 4GB
1583 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1584 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1585 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1588 Frequency : always if "lfs find" used on a local file/directory
1590 Description: "lfs find" segfaults if used on a local file/directory
1591 Details : The case where a directory component was not specified wasn't
1592 handled correctly. Handle this properly.
1595 Frequency : always on ppc64
1597 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1598 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1599 perform endianness conversion.
1602 Frequency : rarely (truncate to non-zero file size after write under load)
1603 Bugzilla : 10730, 10687
1604 Description: Files padded with zeros to next 4K multiple
1605 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1606 it is possible that files that are truncated to a non-zero size
1607 immediately after being written are filled with zero bytes beyond
1608 the truncated size. No file data is lost.
1610 Severity : enhancement
1612 Description: Allow recovery/failover for liblustre clients.
1613 Details : liblustre clients were unaware of failover configurations until
1616 Severity : enhancement
1618 Description: user file locks should fail when not mounting with flock option
1619 Details : Set up an error-returning stub in ll_file_operations.lock field
1620 to prevent incorrect behaviour when client is mounted without
1621 flock option. Also, set up properly f_op->flock field for
1625 Frequency : always on ia64
1627 Description: "lfs df" loops on printing out MDS statfs information
1628 Details : The obd_ioctl_data was not initialized and in some systems
1629 this caused a failure during the ioctl that did not return
1630 an error. Initialize the struct and return an error on failure.
1633 Frequency : SLES 9 only
1635 Description: Error of copying files with lustre special EAs as root
1636 Details : Client side always return success for setxattr call for lustre
1637 special xattr (currently only "trusted.lov").
1640 Frequency : rarely on clusters with both ia64+i386 clients
1642 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1643 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1644 clients are doing concurrent writes to the same file it is
1645 possible that the ia64 clients may LASSERT because the OST
1646 extent locks are not PAGE_SIZE aligned. Ensure that grown
1647 locks are always aligned on the request boundary.
1650 Frequency : specific use, occasional
1652 Description: Overwriting in use executable truncates on-disk binary image
1653 Details : If one node attempts to overwrite an executable in use by
1654 another node, we now correctly return ETXTBSY instead of
1655 truncating the file.
1657 Severity : enhancement
1659 Description: Async OSC create to avoid the blocking unnecessarily.
1660 Details : If a OST has no remain object, system will block on the creating
1661 when need to create a new object on this OST. Now, ways use
1662 pre-created objects when available, instead of blocking on an
1663 empty osc while others are not empty. If we must block, we block
1664 for the shortest possible period of time.
1669 Description: chmod on Lustre root is propagated to other clients
1670 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1671 invalid by the follow_mount time.
1674 Frequency : liblustre clients only
1676 Description: Race in 'instant cancel' lock handling could lead to such locks
1677 never to be granted in case of SMP MDS
1678 Details : Do not destroy not yet granted but cbpending locks in
1682 Frequency : replay/resend of open
1684 Description: non null lock assetion failure in mds_intent_policy
1685 Details : Trying to replay/resend lockless open requests resulted in
1686 mds_open() returning 0 with no lock. Now it sets a flag if
1687 a lock is going to be returned.
1689 Severity : enhancement
1691 Description: Checksum enhancements
1692 Details : New checksum enhancements allow for resending RPCs that failed
1695 Severity : enhancement
1697 Description: Tunables on number of dirty pages in cacche
1698 Details : Allow to set limit on number of dirty pages cached.
1703 Description: client crash on unmount - lock still has references
1704 Details : In some error handling cases it was possible to leak a lock
1705 reference on a client while accessing a file. This was not
1706 harmful to the client during operation, but would cause the
1707 client to crash when the filesystem is unmounted.
1710 Frequency : specific case, rare
1712 Description: ETXTBSY on mds though file not in use by client
1713 Details : ETXTBSY is no longer incorrectly returned when attempting to
1714 chmod or chown a directory that the user previously tried to
1715 execute or a currently-executing binary.
1718 Frequency : extremely rare except on liblustre-based clients
1720 Description: Lustre space not freed when files are deleted
1721 Details : Clean up open-unlinked files after client eviction. Previously
1722 the unlink was skipped and the files remained as orphans.
1727 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1728 Details : In some cases it was possible to send a blocking callback to
1729 a client doing a glimpse, even though that client didn't get
1730 a lock granted. When the glimpse lock is cancelled on the OST
1731 the freed lock is left on the waiting list and corrupted the list.
1734 Frequency : all core dumps
1736 Description: Broke core dumps to lustre
1737 Details : Negative dentry may be unhashed if parent does not have UPDATE
1738 lock, but some callers, e.g. do_coredump, expect dentry to be
1739 hashed after successful create, hash it in ll_create_it.
1741 ------------------------------------------------------------------------------
1743 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1745 * Support for kernels:
1746 2.6.9-42.0.2.EL (RHEL 4)
1747 2.6.5-7.276 (SLES 9)
1748 2.4.21-40.EL (RHEL 3)
1749 2.6.12.6 vanilla (kernel.org)
1753 Frequency : always on RHEL 3
1755 Description: Number of open files grows over time
1756 Details : The number of open files grows over time, whether or not
1757 Lustre is started. This was due to a filp leak introduced
1758 by one of our kernel patches.
1760 ------------------------------------------------------------------------------
1762 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1764 * Support for kernels:
1765 2.6.9-42.EL (RHEL 4)
1766 2.6.5-7.276 (SLES 9)
1767 2.4.21-40.EL (RHEL 3)
1768 2.6.12.6 vanilla (kernel.org)
1773 Bugzilla : 5719, 9635, 9792, 9684
1774 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1775 Details : If a server is under heavy load and cannot reply to new
1776 connection requests before the client resends the (re)connect,
1777 the connection handling code can behave badly if two service
1778 threads are concurrently handing separate (re)connections from
1779 the same client. Add better locking to the connection handling
1780 code, and ensure that only a single connection will be processed
1781 for a given client UUID, even if the lock is dropped.
1783 Severity : enhancement
1785 Description: add TCP zero-copy support to kernel
1786 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1787 sends if the hardware supports scatter-gather and checksumming.
1788 This allows socklnd to do client-write and server-read more
1789 efficiently and reduce CPU utilization from skbuf copying.
1792 Frequency : only if NFS exporting from client
1794 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1795 Details : If setting attributes on a file created under NFS that had
1796 never been opened it would be possible to oops the client
1797 if the file had no objects.
1800 Frequency : always for liblustre
1802 Description: liblustre client does MDS+OSTs setattr RPC for each write
1803 Details : When doing a write from a liblustre client, the client
1804 incorrectly issued an RPC to the MDS and each OST the file was
1805 striped over in order to update the timestamps. When writing
1806 with small chunks and many clients this could overwhelm the MDS
1807 with RPCs. In all cases it would slow down the write because
1808 these RPCs are unnecessary.
1810 Severity : enhancement
1812 Description: allow number of MDS service threads to be changed at module load
1813 Details : It is now possible to change the number of MDS service threads
1814 running. Adding "options mds mds_num_threads={N}" to the MDS's
1815 /etc/modprobe.conf will set the number of threads for the next
1816 time Lustre is restarted (assuming the "mds" module is also
1817 reloaded at that time). The default number of threads will
1818 stay the same, 32 for most systems.
1823 Description: OST crash if filesystem is unformatted or corrupt
1824 Details : If an OST is started on a device that has never been formatted
1825 or if the filesystem is corrupt and cannot even mount then the
1826 error handling cleanup routines would dereference a NULL pointer.
1831 Description: NULL pointer deref in llap_from_page.
1832 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1833 incorrect) mapping if the page was truncated/reclaimed while it was
1834 searched for. Check for this condition and skip such pages when
1835 doing readahead. Introduce extra check to llap_from_page() to
1836 verify page->mapping->host is non-NULL (so page is not anonymous).
1839 Frequency : Sometimes when using sys_sendfile
1841 Description: "page not covered by a lock" warnings from ll_readpage
1842 Details : sendfile called ll_readpage without right page locks present.
1843 Now we introduced ll_file_sendfile that does necessary locking
1844 around call to generic_file_sendfile() much like we do in
1848 Frequency : with certain MDS communication failures at client mount time
1850 Description: NULL pointer deref after failed client mount
1851 Details : a client connection request may delayed by the network layer
1852 and not be sent until after the PTLRPC layer has timed out the
1853 request. If the client fails the mount immediately it will try
1854 to clean up before the network times out the request. Add a
1855 reference from the request import to the obd device and delay
1856 the cleanup until the network drops the request.
1859 Frequency : occasionally during client (re)connect
1861 Description: assertion failure during client (re)connect
1862 Details : processing a client connection request may be delayed by the
1863 client or server longer than the client connect timeout. This
1864 causes the client to resend the connection request. If the
1865 original connection request is replied in this interval, the
1866 client may trip an assertion failure in ptlrpc_connect_interpret()
1867 which thought it would be the only running connect process.
1870 Frequency : only with obd_echo servers and clients that are rebooted
1872 Description: kernel BUG accessing uninitialized data structure
1873 Details : When running an obd_echo server it did not start the ping_evictor
1874 thread, and when a client was evicted an uninitialized data
1875 structure was accessed. Start the ping_evictor in the RPC
1876 service startup instead of the OBD startup.
1878 Severity : enhancement
1879 Bugzilla : 10193 (patchless)
1880 Description: Remove dependency on various unexported kernel interfaces.
1881 Details : No longer need reparent_to_init, exit_mm, exit_files,
1882 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1885 Frequency : rare (only users of deprecated and unsupported LDAP config)
1887 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1888 Details : LDAP apparently contains 'lustreName' attributes instead of
1889 'name'. A simple remapping of the name is sufficient.
1892 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1894 Description: waiting_locks_callback trips kernel BUG if client is evicted
1895 Details : Running with the dump_on_timeout debug flag turned on makes
1896 it possible that the waiting_locks_callback() can try to dump
1897 the Lustre kernel debug logs from an interrupt handler. Defer
1898 this log dumping to the expired_lock_main() thread.
1900 Severity : enhancement
1902 Description: Support NFS exporting on 2.6 kernels.
1903 Details : Implement non-rawops metadata methods for NFS server to use without
1904 changing NFS server code.
1907 Frequency : very rare (synthetic metadata workload only)
1909 Description: two racing renames might cause an MDS thread to deadlock
1910 Details : Running the "racer" program may cause one MDS thread to rename
1911 a file from being the source of a rename to being the target of
1912 a rename at exactly the same time that another thread is doing
1913 so, and the second thread has already enqueued these locks after
1914 doing a lookup of the target and is trying to relock them in
1915 order. Ensure that we don't try to re-lock the same resource.
1918 Frequency : only very large systems with liblustre clients
1920 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1921 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1922 all OSTs in parallel. This allows the eviction of stale liblustre
1923 clients to proceed much faster than if they were done in series,
1924 and also offers similar improvements for other set_info RPCs.
1929 Description: excessive CPU usage during initial read phase on client
1930 Details : During the initial read phase on a client, it would agressively
1931 retry readahead on the file, consuming too much CPU and impacting
1932 performance (since 1.4.5.8). Improve the readahead algorithm
1933 to avoid this, and also improve some other common cases (read
1934 of small files in particular, where "small" is files smaller than
1935 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1940 Description: MDS crash when receiving packet with unknown intent.
1941 Details : Do not LBUG in unknown intent case, just return -EFAULT
1943 Severity : enhancement
1944 Bugzilla : 9293, 9385
1945 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1946 Details : Do not serialize getattr (non-intent version) and statfs.
1949 Frequency : occasional, when OST network is overloaded/intermittent
1951 Description: client evicted by OST after bulk IO timeout
1952 Details : If a client sends a bulk IO request (read or write) the OST
1953 may evict the client if it is unresposive to its data GET/PUT
1954 request. This is incorrect if the network is overloaded (takes
1955 too long to transfer the RPC data) or dropped the OST GET/PUT
1956 request. There is no need to evict the client at all, since
1957 the pinger and/or lock callbacks will handle this, and the
1958 client can restart the bulk request.
1961 Frequency : Always when mmapping file with no objects
1963 Description: client crashes when mmapping file with no objects
1964 Details : Check that we actually have objects in a file before doing any
1965 operations on objects in ll_vm_open, ll_vm_close and
1971 Description: Request leak when working with deleted CWD
1972 Details : Introduce advanced request refcount tracking for requests
1973 referenced from lustre intent.
1975 Severity : Enhancement
1977 Description: Cache open file handles on client.
1978 Details : MDS now will return special lock along with openhandle, if
1979 requested and client is allowed to hold openhandle, even if unused,
1980 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1981 closing files for every read/write openration.
1983 Severity : Enhancement
1985 Description: Cache open negative dentries on client when possible.
1986 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1987 negative dentries on lock revocation.
1992 Description: Remounting a client read-only wasn't possible with a zconf mount
1993 Details : It wasn't possible to remount a client read-only with llmount.
1995 Severity : enhancement
1996 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1997 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1998 MPICH is included to add Lustre-specific ADIO interfaces.
1999 This is based closely on the UFS ADIO layer and only differs
2000 in file creation, in order to allow the OST striping to be set.
2001 This is user-contributed code and not supported by CFS.
2006 Description: extended inode attributes (immutable, append-only) work improperly
2007 when 2.4 and 2.6 kernels are used on client/server or vice versa
2008 Details : Introduce kernel-independent values for these flags.
2010 Severity : enhancement
2013 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2014 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2015 granularity. Now, Fractional MB granularity is be supported,
2016 this is very useful for low memory system.
2018 Severity : enhancement
2020 Description: Getattr by fid
2021 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2022 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2023 have locks and avoids getattr rpc requests for it on every stat.
2026 Frequency : Always, for filesystems larger than 2TB
2028 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2029 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2030 filesystems larger than 2TB to be mounted with mballoc enabled.
2033 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2035 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2036 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2037 filesystems larger than 2TB is mounted then the kernel will
2038 silently corrupt the start of the filesystem. CONFIG_LBD is
2039 enabled for all CFS-supported kernels, but the possibility of
2040 this happening with a modified kernel config exists.
2042 Severity : enhancement
2044 Description: add client O_DIRECT support for 2.6 kernels
2045 Details : It is now possible to do O_DIRECT reads and writes to files
2046 in the Lustre client mountpoint on 2.6 kernel clients.
2048 Severity : enhancement
2050 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2051 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2052 parallel, not waiting for response from every OST before sending
2053 a rpc to the next OST.
2058 Description: setattr vs write race when updating file timestamps
2059 Details : Client processes that update a file timestamp into the past
2060 right after writing to the file (e.g. tar) it is possible that
2061 the updated file modification time can be reset to the current
2062 time due to a race between processing the setattr and write RPC.
2064 Severity : enhancement
2066 Description: Bring 'lfs find' closer in line with regular Linux find.
2067 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2068 -print0 options and obtains all the needed info through the lustre
2071 Severity : enhancement
2073 Description: support up to 1024 configured devices on one node
2074 Details : change obd_dev array from statically allocated to dynamically
2075 allocated structs as they are first used to reduce memory usage
2080 Description: Flush dirty partially truncated pages during truncate
2081 Details : Immediatelly flush partially truncated pages in filter_setattr,
2082 this way we completely avoid having any pages in page cache on OST
2083 and can retire ugly workarounds during writes to flush such pages.
2088 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2089 Details : protect inode from truncation within vfs_unlink() context
2090 just take a reference before calling vfs_unlink() and release it
2091 when parent's i_sem is free.
2096 Description: last_id value checked outside lock on OST caused LASSERT failure
2097 Details : If there were multiple MDS->OST object precreate requests in
2098 flight, it was possible that the OST's last object id was checked
2099 outside a lock and incorrectly tripped an assertion. Move checks
2100 inside locks, and discard old precreate requests.
2103 Frequency : always, if extents are used on OSTs
2105 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2106 Details : OSTs using the extents format would not zero the high 16 bits of
2107 the index physical block number. This is not a problem for any
2108 OST filesystems smaller than 16TB, and no kernels support ext3
2109 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2110 new/modified files) and can be fixed for existing filesystems
2111 with e2fsprogs-1.39-cfs1.
2116 Description: import connection selection may be incorrect if timer wraps
2117 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2118 import connections to be ignored if the 32-bit jiffies counter
2119 wraps. Use a 64-bit jiffies counter.
2122 Frequency : very large clusters immediately after boot
2124 Description: LNET request buffers exhausted under heavy short-term load
2125 Details : If a large number of client requests are generated on a service
2126 that has previously never seen so many requests it is possible
2127 that the request buffer growth cannot keep up with the spike in
2128 demand. Instead of dropping incoming requests, they are held in
2129 the LND until the RPC service can accept more requests.
2132 Frequency : Sometimes during replay
2134 Description: Assertion failure in ll_local_open after replay.
2135 Details : If replay happened on an open request reply before we were able
2136 to set replay handler, reply will become not swabbed tripping the
2137 assertion in ll_local_open. Now we set the handler right after
2138 recognising of open request
2141 Frequency : very rare
2143 Description: kernel reports "badness in vsnprintf"
2144 Details : Reading from the "recovery_status" /proc file in small chunks
2145 may cause a negative length in lprocfs_obd_rd_recovery_status()
2146 call to vsnprintf() (which is otherwise harmless). Exit early
2147 if there is no more space in the output buffer.
2149 Severity : enhancement
2151 Description: clear OBD RPC statistics by writing to them
2152 Details : It is now possible to clear the OBD RPC statistics by writing
2153 to the "stats" file.
2158 Description: Client mtime is not the same on different clients after utimes
2159 Details : In some cases, the client was using the utimes() syscall on
2160 a file cached on another node. The clients now validate the
2161 ctime from the MDS + OSTs to determine which one is right.
2166 Description: Inability to activate failout mode
2167 Details : lconf script incorrectly assumed that in pythong string's numeric
2168 value is used in comparisons.
2171 Frequency : always with multiple stripes per file
2173 Description: Inefficient object allocation for mutli-stripe files
2174 Details : When selecting which OSTs to stripe files over, for files with
2175 a stripe count that divides evenly into the number of OSTs,
2176 the MDS is always picking the same starting OST for each file.
2177 Return the OST selection heuristic to the original design.
2182 Description: mount failures may take full timeout to return an error
2183 Details : Under some heavy load conditions it is possible that a
2184 failed mount can wait for the full obd_timeout interval,
2185 possibly several minutes, before reporting an error.
2186 Instead return an error as soon as the status is known.
2188 Frequency : quota enabled and large files being deleted
2190 Description: releasing more than 4GB of quota at once hangs OST
2191 Details : If a user deletes more than 4GB of files on a single OST it
2192 will cause the OST to spin in an infinite loop. Release
2193 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2198 Description: statfs data retrieved from /proc may be stale or zero
2199 Details : When reading per-device statfs data from /proc, in the
2200 {kbytes,files}_{total,free,avail} files, it may appear
2201 as zero or be out of date.
2204 Frequency : systems with MD RAID1 external journal devices
2206 Description: lconf's call to blkid is confused by RAID1 journal devices
2207 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2208 of returning all block devices that match the journal UUID.
2211 Frequency : always, for aggregate stripe size over 4GB
2213 Description: assertion fails when trying to use 4GB stripe size
2214 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2215 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2218 Frequency : always on ppc64
2220 Description: the first write on an ext3 filesystem with mballoc got stuck
2221 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2222 perform endianness conversion.
2224 ------------------------------------------------------------------------------
2226 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2228 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2229 INTEROPERATE with older versions automatically. Please read the
2230 user documentation before upgrading any part of a live system.
2231 * WARNING: Lustre networking configuration changes are required with
2232 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2235 * Support for kernels:
2236 2.6.9-22.0.2.EL (RHEL 4)
2237 2.6.5-7.244 (SLES 9)
2238 2.6.12.6 vanilla (kernel.org)
2241 Severity : enhancement
2242 Bugzilla : 7981/8208
2243 Description: Introduced Lustre Networking (LNET)
2244 Details : LNET is new networking infrastructure for Lustre, it includes
2245 a reorganized network configuration mode (see the user
2246 documentation for full details) as well as support for routing
2247 between different network fabrics. Lustre Networking Devices
2248 (LNDs) for the supported network fabrics have also been
2249 created for this new infrastructure.
2251 Severity : enhancement
2252 Description: Introduced Access control lists
2253 Details : clients can set ACLs on files and directories in order to have
2254 more fine-grained permissions than the standard Unix UGO+RWX.
2255 The MDS must be started with the "-o acl" mount option.
2257 Severity : enhancement
2258 Description: Introduced filesystem quotas
2259 Details : Administrators may now establish per-user quotas on the
2262 Severity : enhancement
2264 Description: Configuration change for the XT3
2265 The PTLLND is now used to run Lustre over Portals on the XT3
2266 The configure option(s) --with-cray-portals are no longer used.
2267 Rather --with-portals=<path-to-portals-includes> is used to
2268 enable building on the XT3. In addition to enable XT3 specific
2269 features the option --enable-cray-xt3 must be used.
2274 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2275 Details : A race between allocating a new llcd and re-getting the llcd_lock
2276 allowed another thread to grab newly-allocated llcd.
2278 Severity : enhancement
2280 Description: 2.6 OST async journal commit and locking fix to improve performance
2281 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2282 2.6 kernels are now async as they already were in 2.4 kernels so
2283 that they can commit concurrently with the network bulk transfer.
2284 For block-allocated files the filter allocation semaphore is held
2285 to avoid filesystem fragmentation during allocation. BKL lock
2286 removed for 2.6 xattr operations where it is no longer needed.
2291 Description: lconf incorrectly determined whether two IP networks could talk
2292 Details : In some more complicated routing and multiple-network
2293 configurations, lconf will avoid trying to make a network
2294 connection to a disjoint part of the IP space. It was doing the
2295 math incorrectly for one set of cases.
2300 Description: Fix for potential infinite loop processing records in an llog.
2301 Details : If an llog record is corrupted/zeroed, it is possible to loop
2302 forever in llog_process(). Validate the llog record length
2303 and skip the remainder of the block on error.
2306 Frequency : occasional (liblustre only)
2308 Description: liblustre could not open files whose last component is a symlink
2309 Details : sysio_path_walk() would incorrectly pass the open intent to
2310 intermediate path components.
2313 Frequency : rare (liblustre only with non-standard tuning)
2314 Bugzilla : 7201 (7350)
2315 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2316 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2317 after it has been released, only do this for referenced locks
2319 Severity : enhancement
2321 Description: specifying an (invalid) directory default stripe_size of -1
2322 would reset the directory default striping
2323 Details : stripe_size -1 was used internally to signal directory stripe
2324 removal, now use "all default" to signal dir stripe removal
2325 as a directory striping of "all default" is not useful
2328 Frequency : common for large clusters running liblustre clients
2330 Description: doing an ls when liblustre clients are running is slow
2331 Details : sending a glimpse AST to a liblustre client waits for every AST
2332 to time out, as liblustre clients will not respond. Since they
2333 cannot cache data we refresh the OST lock LVB from disk instead.
2335 Severity : enhancement
2337 Description: doing an ls at the same time as file IO can be slow
2338 Details : enqueue and other "small" requests can be blocked behind many
2339 large IO requests. Create a new OST IO portal for non-IO
2340 requests so they can be processed faster.
2343 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2345 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2346 created file with mode 000
2347 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2348 allow an ftruncate() as a non-root user to a file with mode 000.
2349 Lustre now respects this flag to disable mode checks when
2350 truncating a file owned by the user
2353 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2356 Description: Revoking locks from clients that went dead or catatonic might take
2358 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2359 cancellation of such locks instant on servers without waiting for
2360 any reply from clients. Clients drops these locks when cancel
2361 notification from server is received without replying.
2364 Frequency : liblustre-only, when liblustre client dies or becomes busy
2366 Description: Doing ls on Linux clients can take a long time with active
2368 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2369 granting such locks to it in the first place if possible. Locks
2370 are taken by proxy on the OST during the read or write and
2371 dropped immediately afterward. Add connect flags handling, do
2372 not grant locks to liblustre clients for glimpse ASTs.
2374 Severity : enhancement
2376 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2377 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2378 sequential reads and behaves badly with applications doing random
2379 reads. Improve it by reading ahead at least read region, and
2380 avoiding excessive large RPC for small reads.
2382 Severity : enhancement
2384 Description: Creating more than 1000 files for a single job may cause a load
2385 imbalance on the OSTs if there are also a large number of OSTs.
2386 Details : qos_prep_create() uses an OST index reseed value that is an
2387 even multiple of the number of available OSTs so that if the
2388 reseed happens in the middle of the object allocation it will
2389 still utilize the OSTs as uniformly as possible.
2394 Description: OST or MDS may oops in ping_evictor_main()
2395 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2396 but doesn't restart at beginning of obd_exports_timed list
2399 Severity : enhancement
2401 Description: improve by-nid export eviction on the MDS and OST
2402 Details : allow multiple exports with the same NID to be evicted at one
2403 time without re-searching the exports list.
2406 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2408 Description: MDS may oops in groups_free()
2409 Details : in rare race conditions a newly allocated group_info struct is
2410 freed again, and this can be NULL. The 2.4 compatibility code
2411 for groups_free() checked for a NULL pointer, but 2.6 did not.
2414 Frequency : common for liblustre clients doing little filesystem IO
2415 Bugzilla : 9352, 7313
2416 Description: server may evict liblustre clients accessing contended locks
2417 Details : if a client is granted a lock or receives a completion AST
2418 with a blocking AST already set it would not reply to the AST
2419 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2422 Frequency : lfs setstripe, only systems with more than 160 OSTs
2424 Description: unable to set striping with a starting offset beyond OST 160
2425 Details : llapi_create_file() incorrectly limited the starting stripe
2426 index to the maximum single-file stripe count.
2429 Frequency : LDAP users only
2431 Description: lconf did not handle in-kernel recovery with LDAP properly
2432 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2434 Severity : enhancement
2436 Description: bind OST threads to NUMA nodes to improve performance
2437 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2438 node and do their allocations there to localize memory access
2440 Severity : enhancement
2442 Description: llmount can determine client NID directly from Myrinet (GM)
2443 Details : the client NID code from gmnalnid was moved directly into
2444 llmount, removing the need to use this or specifying the
2445 client NID explicitly when mounting GM clients with zeroconf
2448 Frequency : if client is started with down MDS
2450 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2451 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2452 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2453 ptlrpc_import_delayed_req() because we didn't check if the
2454 request was interrupted, and we also didn't break out of the
2455 event loop if there was no timeout
2460 Description: data loss during non-page-aligned writes to a single file from
2461 both multiple nodes and multiple threads on one node at same time
2462 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2463 inconsistency led to false short-reads, that were cached and later
2464 used by ->prepare_write() to fill in partially written page,
2465 leading to data loss.
2468 Frequency : always, if lconf --abort_recovery used
2470 Description: lconf --abort_recovery fails with 'Operation not supported'
2471 Details : lconf was attempting to abort recovery on the MDT device and not
2474 Severity : enhancement
2476 Description: remove cleanup logs
2477 Details : replace lconf-generated cleanup logs with lustre internal
2478 cleanup routines. Eliminates the need for client-cleanup and
2481 Severity : enhancement
2483 Description: add support for EAs (user and system) on lustre filesystems
2484 Details : it is now possible to store extended attributes in the Lustre
2485 client filesystem, and with the user_xattr mount option it
2486 is possible to allow users to store EAs on their files also
2488 Severity : enhancement
2490 Description: Add possibility (config option) to show minimal available OST free
2492 Details : When compiled with --enable-mindf configure option, statfs(2)
2493 (and so, df) will return least minimal free space available from
2494 all OSTs as amount of free space on FS, instead of summary of
2495 free spaces of all OSTs.
2497 Severity : enhancement
2499 Description: do not expand extent locks acquired on OST-side
2500 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2501 by server: they are not cached anyway.
2504 Frequency : when mmap is used/binaries executed from Lustre
2506 Description: Unmmap pages before throwing them away from read cache.
2507 Details : llap_shrink cache now attempts to unmap pages before discarding
2508 them (if unmapping failed - do not discard). SLES9 kernel has
2509 extra checks that trigger if this unmapping is not done first.
2514 Description: lconf didn't resolve symlinks before checking to see whether a
2515 given mountpoint was already in use
2518 Frequency : when migrating failover services
2519 Bugzilla : 6395, 9514
2520 Description: When migrating a subset of services from a node (e.g. failback
2521 from a failover service node) the remaining services would
2522 time out and evict clients.
2523 Details : lconf --force (implied by --failover) sets the global obd_timeout
2524 to 5 seconds in order to quickly disconnect, but this caused
2525 other RPCs to time out too quickly. Do not change the global
2526 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2528 Severity : enhancement
2529 Frequency : if MDS is started with down OST
2530 Bugzilla : 9439,5706
2531 Description: Allow startup/shutdown of an MDS without depending on the
2532 availability of the OSTs.
2533 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2534 Add appropriate locking and lov-osc refcounts for safe
2535 cleaning. Add osc abort_inflight calls in case the
2536 synchronize never started.
2539 Frequency : occasional (Cray XT3 only)
2541 Description: root not authorized to access files in CRAY_PORTALS environment
2542 Details : The client process capabilities were not honoured on the MDS in
2543 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2544 been accessed by an authorized user then root was able to access
2545 the file on the local client also. The root user capabilities
2546 are now allowed on the MDS, as this environment has secure UID.
2549 Frequency : occasional
2551 Description: ldiskfs "too long searching" message happens too often
2552 Details : A debugging message (otherwise harmless) prints too often on
2553 the OST console. This has been reduced to only happen when
2554 there are fragmentation problems on the filesystem.
2559 Description: Division by zero in statfs when all OSCs are inactive
2560 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2561 lov_statfs divides by value returned by lov_get_stripecnt().
2565 Bugzilla : 9489, 3273
2566 Description: First write from each client to each OST was only 4kB in size,
2567 to initialize client writeback cache, which caused sub-optimal
2568 RPCs and poor layout on disk for the first writen file.
2569 Details : Clients now request an initial cache grant at (re)connect time
2570 and so that they can start streaming writes to the cache right
2571 away and always do full-sized RPCs if there is enough data.
2572 If the OST is rebooted the client also re-establishes its grant
2573 so that client cached writes will be honoured under the grant.
2578 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2579 Details : Now I/O RPCs go to different portal number and (presumably) fast
2580 lock requests (and glimses) and other RPCs get their own service
2581 threads pool that should be able to service those RPCs
2584 Severity : enhancement
2586 Description: Ability to exchange lustre version between client and servers and
2587 issue warnings at client side if client is too old. Also for
2588 liblustre clients there is ability to refuse connection of too old
2590 Details : New 'version' field is added to connect data structure that is
2591 filled with version info. That info is later checked by server and
2595 Frequency : rare, liblustre only.
2596 Bugzilla : 9296, 9581
2597 Description: Two simultaneous writes from liblustre at offset within same page
2598 might proceed at the same time overwriting eachother with stale
2600 Details : I/O lock withing llu_file_prwv was released too early, before data
2601 actually was hitting the wire. Extended lock-holding time until
2602 server acknowledges receiving data.
2605 Frequency : extremely rare. Never observed in practice.
2607 Description: avoid generating lustre_handle cookie of 0.
2608 Details : class_handle_hash() generates handle cookies by incrementing
2609 global counter, and can hit 0 occasionaly (this is unlikely, but
2610 not impossible, because initial value of cookie counter is
2611 selected randonly). Value of 0 is used as a sentinel meaning
2612 "unassigned handle" --- avoid it. Also coalesce two critical
2613 sections in this function into one.
2615 Severity : enhancement
2617 Description: allow liblustre clients to delegate truncate locking to OST
2618 Details : To avoid overhead of locking, liblustre client instructs OST to
2619 take extent lock in ost_punch() on client's behalf. New connection
2620 flag is added to handle backward compatibility.
2622 Severity : enhancement
2623 Bugzilla : 4928, 7341, 9758
2624 Description: allow number of OST service threads to be specified
2625 Details : a module parameter allows the number of OST service threads
2626 to be specified via "options ost ost_num_threads={N}" in the
2627 OSS's /etc/modules.conf or /etc/modprobe.conf.
2631 Bugzilla : 6146, 9635, 9895
2632 Description: servers crash with bad pointer in target_handle_connect()
2633 Details : In rare cases when a client is reconnecting it was possible that
2634 the connection request was the last reference for that export.
2635 We would temporarily drop the export reference and get a new
2636 one, but this may have been the last reference and the export
2637 was just destroyed. Get new reference before dropping old one.
2639 Severity : enhancement
2640 Frequency : if client is started with failover MDS
2642 Description: Allow multiple MDS hostnames in the mount command
2643 Details : Try to read the configuration from all specified MDS
2644 hostnames during a client mount in case the "primary"
2647 Severity : enhancement
2649 Description: Stop sending data to evicted clients as soon as possible.
2650 Details : Check if the client we are about to send or are sending data to
2651 was evicted already. (Check is done every second of waiting,
2652 for which l_wait_event interface was extended to allow checking
2653 of exit condition at specified intervals).
2656 Frequency : rare, normally only when NFS exporting is done from client
2658 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2660 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2663 Frequency : always, if srand() is called before liblustre initialization
2665 Description: Liblustre uses system PRNG disturbing its usage by user application
2666 Details : Introduce internal to lustre fast and high-quality PRNG for
2667 lustre usage and make liblustre and some other places in generic
2668 lustre code to use it.
2670 Severity : enhancement
2671 Bugzilla : 9477, 9557, 9870
2672 Description: Verify that the MDS configuration logs are updated when xml is
2673 Details : Check if the .xml configuration logs are newer than the config
2674 logs stored on the MDS and report an error if this is the case.
2675 Request --write-conf, or allow starting with --old_conf.
2677 Severity : enhancement
2679 Description: Handle symlinks in the path when checking if Lustre is mounted.
2680 Details : Resolve intermediate symlinks when checking if a client has
2681 mounted a filesystem to avoid duplicate client mounts.
2686 Description: lconf can hit an error exception but still return success.
2687 Details : The lconf command catches the Command error exception at the top
2688 level script context and will exit with the associated exit
2689 status, but doesn't ensure that this exit status is non-zero.
2694 Description: failure of ptlrpc thread startup can cause oops
2695 Details : Starting a ptlrpc service thread can fail if there are a large
2696 number of threads or the server memory is very fragmented.
2697 Handle this without oopsing.
2700 Frequency : always, only if liblustre and non-default acceptor port was used
2702 Description: liblustre cannot connect to servers with non-default acceptor port
2703 Details : tcpnal_set_default_params() was not called and was therefore
2704 ignoring the environment varaible TCPNAL_PORT, as well as other
2705 TCPNAL_ environment variables
2710 Description: two objects could be created on the same OST for a single file
2711 Details : If an OST is down, in some cases it was possible to create two
2712 objects on a single OST for a single file. No problems other
2713 than potential performance impact and spurious error messages.
2717 Bugzilla : 5681, 9562
2718 Description: Client may oops in ll_unhash_aliases
2719 Details : Client dcache may become inconsistent in race condition.
2720 In some cases "getcwd" can fail if the current directory is
2726 Description: Inode refcounting problems in NFS export code
2727 Details : link_raw functions used to call d_instantiate without obtaining
2728 extra inode reference first.
2732 Bugzilla : 9942, 9903
2733 Description: Referencing freed requests leading to crash, memleaks with NFS.
2734 Details : We used to require that call to ll_revalidate_it was always
2735 followed by ll_lookup_it. Also with revalidate_special() it is
2736 possible to call ll_revalidate_it() twice for the same dentry
2737 even if first occurence returned success. This fix changes semantic
2738 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2739 reference on a request referred from the intent.
2740 ll_intent_release() then releases such a request.
2743 Frequency : rare, normally benchmark loads only
2745 Description: unlinked inodes were kept in memory on the client
2746 Details : If a client is repeatedly creating and unlinking files it
2747 can accumulate a lot of stale inodes in the inode slab cache.
2748 If there is no other client load running this can cause the
2749 client node to run out of memory. Instead flush old inodes
2750 from client cache that have the same inode number as a new inode.
2753 Frequency : SLES9 2.6.5 kernel and long filenames only
2754 Bugzilla : 9969, 10379
2755 Description: utime reports stale NFS file handle
2756 Details : SLES9 uses out-of-dentry names in some cases, which confused
2757 the lustre dentry revalidation. Change it to always use the
2761 Frequency : rare, unless heavy write-truncate concurrency is continuous
2762 Bugzilla : 4180, 6984, 7171, 9963, 9331
2763 Description: OST becomes very slow and/or deadlocked during object unlink
2764 Details : filter_destroy() was holding onto the parent directory lock
2765 while truncating+unlinking objects. For very large objects this
2766 may block other threads for a long time and slow overall OST
2767 responsiveness. It may also be possible to get a lock ordering
2768 deadlock in this case, or run out of journal credits because of
2769 the combined truncate+unlink. Solution is to do object truncate
2770 first in one transaction without parent lock, and then do the
2771 final unlink in a new transaction with the parent lock. This
2772 reduces the lock hold time dramatically.
2775 Frequency : rare, 2.4 kernels only
2777 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2778 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2779 before unmounting their filesystem in order to avoid deadlock.
2780 The kernel_locked() function in 2.4 kernels only checks whether
2781 the kernel lock is held, not whether it is this process that is
2782 holding it as 2.6 kernels do.
2787 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2788 Details : The client ptlrpc code may be trying to reconnect to a down
2789 server before a previous connection attempt has timed out.
2790 Increase the reconnect interval to be longer than the connection
2791 timeout interval to avoid sending duplicate connections to
2795 Frequency : echo_client brw_test command
2797 Description: fix echo_client to work with OST preallocated code
2798 Details : OST preallocation code (5137) didn't take echo_client IO path
2799 into account: echo_client calls filter methods outside of any
2800 OST thread and, hence, there is no per-thread preallocated
2801 pages and buffers to use. Solution: hijack pga pages for IO. As
2802 a byproduct, this avoids unnecessary data copying.
2806 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2807 Description: Client can oops in mdc_commit_close() after open replay
2808 Details : It was possible for the MDS to return an open request with no
2809 transaction number in mds_finish_transno() if the client was
2810 evicted, but without actually returning an error. Clients
2811 would later try to replay that open and may trip an assertion
2812 Simplify the client close codepath, and always return an error
2813 from the MDS in case the open is not successful.
2816 Frequency : rare, 2.6 OSTs only
2818 Description: OST may deadlock under high load on fragmented files
2819 Details : If there was a heavy load and highly-fragmented OST filesystems
2820 it was possible to have all the OST threads deadlock waiting on
2821 allocation of biovecs, because the biovecs were not released
2822 until the entire RPC IO was completed. Instead, release biovecs
2823 as soon as they are complete to ensure forward IO progress.
2825 Severity : enhancement
2827 Description: Support for specifying external journal device at mount
2828 Details : If an OST or MDS device is formatted with an external journal
2829 device, this device major/minor is stored in the ext3 superblock
2830 and may not be valid for failover. Allow detecting and
2831 specifying the external journal at mount time.
2836 Description: Mounting an MDS with pending unlinked files may cause oops
2837 Details : target_finish_recovery() calls mds_postrecov() which returned
2838 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2839 considers this an error and immediately begins cleaning up the
2840 lov, just after starting the mds_lov process
2842 Severity : enhancement
2844 Description: Implement 'lfs df' to report actual free space on per-OST basis
2845 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2846 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2847 usage of objects; '-h' to report in human readable format.
2849 ------------------------------------------------------------------------------
2851 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2858 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2859 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2860 at startup during orphan unlinking. Instead, skip buddy update
2861 before setup, as it will be regenerated after recovery is complete.
2866 Description: If an OST is inactive, its locks might reference stale inodes.
2867 Details : lov_change_cbdata() must iterate over all namespaces, even if
2868 they are inactive to clear inode references from the lock.
2870 Severity : enhancement
2871 Frequency : occasional, if non-standard max_dirty_mb used
2873 Description: Client will block write RPCs if not enough grant
2874 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2875 then the client will block writes while waiting for another RPC
2876 to complete instead of consuming its dirty limit. With change
2877 we get improved performance when max_dirty_mb is small.
2879 Severity : enhancement
2880 Bugzilla : 3389, 6253
2881 Description: Add support for supplementary groups on the MDS.
2882 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2883 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2884 lookups for user supplementary groups into a cache.
2888 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2889 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2890 for liblustre O_EXCL is enforced by MDS.
2894 Description: readdir never returns NULL in liblustre.
2895 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2896 dentry in struct dirent.
2900 Frequency : liblustre only, depends on application IO pattern
2901 Description: liblustre clients evicted if not contacting servers
2902 Details : Don't put liblustre clients into the ping_evictor list, so
2903 they will not be evicted by the pinger ever.
2905 Severity : enhancement
2907 Description: Add ability to evict clients by NID from MDS.
2908 Details : By echoing "nid:$NID" string into
2909 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2910 $NID would be instantly evicted from this MDS and from all active
2911 OSTs connected to it.
2915 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2916 Details : lookup_it_finish() used to query file size from OSTs that was not
2921 Description: service threads change working directory to that of init
2922 Details : Starting lustre service threads may pin the working directory
2923 of the parent thread, making that filesystem busy. Threads
2924 now change to the working directory of init to avoid this.
2928 Frequency : during shutdown only
2929 Description: shutdown with a failed MDS or OST can cause unmount to hang
2930 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2935 Frequency : 2.6 only
2936 Description: chmod/chown may include an extra supplementary group
2937 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2938 supplementary group and if none is specified this is used.
2941 Bugzilla : 5479 (6816)
2943 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2944 Details : If lookup is in progress on a file that is unlinked we might try
2945 to revalidate the inode and fail in revalidate after lookup is
2946 complete and ll_file_open() enqueues the open again but
2947 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2950 Frequency : always, if lconf --abort_recovery used
2952 Description: lconf --abort_recovery fails with 'Operation not supported'
2953 Details : lconf was attempting to abort recovery on the MDT device and not
2956 ------------------------------------------------------------------------------
2958 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2963 Frequency : rare (only unsupported configurations with a node running as an
2965 Bugzilla : 6514, 5137
2966 Description: Mounting a Lustre file system on a node running as an OST could
2968 Details : OSTs now preallocates memory needed to write out data at
2969 startup, instead of when needed, to avoid having to
2970 allocate memory in possibly low memory situations.
2971 Specifically, if the file system is mounted on on OST,
2972 memory pressure could force it to try to write out data,
2973 which it needed to allocate memory to do. Due to the low
2974 memory, it would be unable to do so and the node would
2975 become unresponsive.
2977 Severity : enhancement
2979 Description: Addition of lconf --service command line option
2980 Details : lconf now accepts a '--service <arg>' option, which is
2981 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2983 Severity : enhancement
2985 Description: Failover mode is now the default for OSTs.
2986 Details : By default, OSTs will now run in failover mode. To return to
2987 the old behaviour, add '--failout' to the lmc line for OSTs.
2989 Severity : enhancement
2991 Description: Health checks are now provided for MDS and OSTs
2992 Details : Additional detailed health check information on MSD and OSTs
2993 is now provided through the procfs health_check value.
2996 Frequency : occasional, depends on IO load
2998 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2999 after numerous create/delete cycles
3000 Details : The ext3 inode allocation policy would not allocate new inodes
3001 very well on the OSTs because there are no new directories
3002 being created. Instead we look for groups with free space if
3003 the parent directories are nearly full.
3008 Description: Network or server problems during mount may cause partially
3009 mounted clients instead of returning an error.
3010 Details : The config llog parsing code may overwrite the error return
3011 code during mount error handling, returning success instead
3017 Description: MDS can fail to allocate large reply buffers
3018 Details : After long uptimes the MDS can fail to allocate large reply
3019 buffers (e.g. zconf client mount config records) due to memory
3020 fragmentation or consumption by the buffer cache. Preallocate
3021 some large reply buffers so that these replies can be sent even
3022 under memory pressure.
3026 Frequency : rare (liblustre)
3027 Description: fsx running with liblustre complained that using truncate() to
3028 extend the file doesn't work. This patch corrects that issue.
3029 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3030 ATTR_SIZE and lsm use in llu_setattr_raw.
3034 Frequency : rare, only 2.6 kernels
3035 Description: Unusual file access patterns on the MDS may result in inode
3036 data being lost in very rare circumstances.
3037 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3038 code in the 2.6 kernel caused the inode and/or EA data not to
3039 be read from disk, causing single-file corruption.
3043 Frequency : rare, only 2.6 filesystems using extents
3044 Description: Heavy concurrent write and delete load may cause data corruption.
3045 Details : It was possible under high-load situations to have an extent
3046 metadata block in the block device cache from a just-unlinked
3047 file overwrite a newly-allocated data block. We now unmap any
3048 metadata buffers that alias just-allocated data blocks.
3052 Frequency : filesystems with default stripe_count larger than 77
3053 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3054 Details : lconf specifies an inode size of 4096 bytes when the default
3055 stripe_count is larger than 77. This conflicts with the default
3056 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3057 this case to avoid pinning too much memory for large EAs.
3059 ------------------------------------------------------------------------------
3061 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3066 Frequency : rare (extremely heavy IO load with hundreds of clients)
3068 Description: Client is evicted, gets IO error writing to file
3069 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3070 caused clients to be evicted for AST timeouts. The fixes in
3071 bug 5192 mean we no longer need to have such short AST timeouts
3072 so ldlm_timeout has been increased.
3075 Frequency : occasional during --force or --failover shutdown under load
3076 Bugzilla : 5949, 4834
3077 Description: Server oops/LBUG if stopped with --force or --failover under load
3078 Details : a collection of import/export refcount and cleanup ordering
3079 issues fixed for safer force cleanup
3082 Frequency : only filesystems larger than 120 OSTs
3083 Bugzilla : 5990, 6223
3084 Description: lfs getstripe would oops on a very large filesystem
3085 Details : lov_getconfig used kfree on vmalloc'd memory
3088 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3089 Bugzilla : 6242, 6243
3090 Description: reading from files that had been truncated to a non-zero size
3091 but never opened returned no data
3092 Details : ll_file_read() reads zeros from no-object files to EOF
3097 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3098 Details : There's little harm in aborting MDS/OSS recovery and letting it
3099 try again, so I removed the LASSERT and return an error instead.
3101 Severity : enhancement
3103 Description: New debugging infrastructure for tracking down data corruption
3104 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3105 (b) cover more of the client-side code path, and (c) try to narrow
3106 down where problems occurred
3110 Bugzilla : 3819, 4364, 4397, 6313
3111 Description: Racing close and eviction MDS could cause assertion in mds_close
3112 Details : It was possible to get multiple mfd references during close and
3113 client eviction, leading to one thread referencing a freed mfd.
3115 Severity: : enhancement
3116 Bugzilla : 3262, 6359
3117 Description: Attempts to reconnect to servers are now more aggressive.
3118 Details : This builds on the enhanced upcall-less recovery that was added
3119 in 1.4.2. When trying to reconnect to servers, clients will
3120 now try each server in the failover group every 10 seconds. By
3121 default, clients would previously try one server every 25 seconds.
3126 Description: After recovery, certain operations trigger a failed
3127 assertion on a client.
3128 Details : Failing over an mds, using lconf -d --failover, while a
3129 client was doing a readdir() call would cause the client to
3130 LBUG after recovery completed and the readdir() was resent.
3132 Severity : enhancement
3134 Description: Default groups are now added by lconf
3135 Details : You can now run lconf --group <servicename> without having to
3136 manually add groups with lmc.
3139 Frequency : occasional
3141 Description: Nodes with an elan id of 0 trigger a failed assertion
3144 Frequency : always when accessing e.g. tty/console device nodes
3146 Description: tty and some other devices nodes cannot be used on lustre
3147 Details : file's private_data field is used by device data and lustre
3148 values in there got lost. New field was added to struct file to
3149 store fs-specific private data.
3152 Frequency : when exporting Lustre via NFS
3154 Description: NFSD failed occasionally when looking up a path component
3155 Details : NFSD is looking up ".." which was broken in ext3 directories
3156 that had grown large enough to become hashed.
3159 Frequency : Clusters with multiple interfaces not on the same subnet
3161 Description: Nodes will repeatedly try to reconnect to an interface which it
3162 cannot reach and report an error to the log.
3163 Details : Extra peer list entries will be created by lconf with some peers
3164 unreachable. lconf now validates the peer before adding it.
3167 Frequency : Only if a default stripe is set on the filesystem root.
3169 Description: Setting a default stripe on the filesystem root prevented the
3170 filesystem from being remounted.
3171 Details : The client was sending extra request flags in the root getattr
3172 request and did not allocate a reply buffer for the dir EA.
3175 Frequency : occasional, higher if lots of files are accessed by one client
3176 Bugzilla : 6159, 6097
3177 Description: Client trips assertion regarding lsm mismatch/magic
3178 Details : While revalidating inodes the VFS looks up inodes with ifind()
3179 and in rare cases can find an inode that is being freed.
3180 The ll_test_inode() code will free the lsm during ifind()
3181 when it finds an existing inode and then the VFS later attaches
3182 this free lsm to a new inode.
3186 Bugzilla : 6422, 7030
3187 Description: MDS deadlock between mkdir and client eviction
3188 Details : Creating a new file via mkdir or mknod (starting a transaction
3189 and getting the ns lock) can deadlock with client eviction
3190 (gets ns lock and trying to finish a synchronous transaction).
3193 Frequency : occasional
3194 Description: While starting a server, the fsfilt_ext3 module could not be
3196 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3197 kernels. Previously, lconf would still use the ext3 name
3198 when trying to load modules. Now, it will correctly use
3199 ext3 on 2.4 and ldiskfs on 2.6.
3201 Severity : enhancement
3202 Description: The default stripe count has been changed to 1
3203 Details : The interpretation of the default stripe count (0, to lfs
3204 or lmc) has been changed to mean striping across a single
3205 OST, rather than all available. For general usage we have
3206 found a stripe count of 1 or 2 works best.
3208 Severity : enhancement
3209 Description: Add support for compiling against Cray portals.
3210 Details : Conditional compiling for some areas that are different
3214 Frequency : occasional
3215 Bugzilla : 6409, 6834
3216 Description: Creating files with an explicit stripe count may lead to
3217 a failed assertion on the MDS
3218 Details : If some OSTs are full or unavailable, creating files may
3219 trigger a failed assertion on the MDS. Now, Lustre will
3220 try to use other servers or return an error to the
3224 Frequency : occasional
3226 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3227 caused lock timeout during readahead (since 1.4.2).
3228 Details : Processes doing readahead might match a lock that hasn't been
3229 granted yet if there are overlapping and conflicting lock
3230 requests. The readahead process waits on ungranted lock
3231 (original lock is CBPENDING), while OST waits for that process
3232 to cancel CBPENDING read lock and eventually evicts client.
3234 Severity : enhancement
3236 Description: Initial enabling of flock support for clients
3237 Details : Implements fcntl advisory locking and file status functions.
3238 This feature is provided as an optional mount flag (default
3239 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3240 locking are implemented yet, and those that are are not guaranteed
3241 to be completely correct in production environments.
3242 mount -t lustre -o [flock|noflock] ...
3245 Frequency : occasional
3247 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3248 assertion in the ext3 JBD (journaling) layer.
3249 Details : The b_committed_data struct is protected by the big kernel lock
3250 in 2.4 kernels, serializing journal_commit_transaction() and
3251 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3252 access to this struct. In 2.6 kernels there is finer grained
3253 locking to improve SMP performance of the JBD layer.
3257 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3262 Description: merge and cleanup kernel patches.
3265 -----------------------------------------------------------------------------
3267 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3269 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3270 versions of Lustre. Please update all servers and clients to
3271 version 1.4.2 or later at the same time. You must also run
3272 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3273 to update the configuration logs.
3275 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3276 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3277 - swab llog records of type '0' so we get proper header size/idx (5861)
3278 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3279 - fix rename of one directory over another leaking an inode (5953)
3280 - avoid SetPageDirty on 2.6 (5981)
3281 - don't re-add just-being-destroyed locks to the waiting list (5653)
3282 - when creating new directories, inherit the parent's custom
3283 striping settings if present parent (3048)
3284 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3285 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3286 - don't hold client locks on temporary worklist from l_lru (5666)
3287 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3288 - automatically evict dead clients (5921)
3289 - Update file size properly in create+truncate+fstat case (6196)
3290 - Do not unhash mountpoint dentries, do not allow removal of
3292 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3293 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3294 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3295 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3296 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3297 - Correctly handle reads of files with no objects (6243)
3298 - lctl recover will also mark a device active if deactivate used (5933)
3300 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3301 - allow --write-conf on an MDS with different nettype than client (5619)
3302 - don't write config llogs to MDS for mounts not from that MDS (5617)
3303 - lconf should create multiple TCP connections from a client (5201)
3304 - init scripts are now turned off by default; run chkconfig --on
3305 lustre and chkconfig --on lustrefs to use them
3306 - upcalls are no longer needed for clients to recover to failover
3308 - add --abort-recovery option to lconf to abort recovery on device
3310 - add support for an arbitrary number of OSTs (3026)
3311 - Quota support protocol changes.
3312 - forward compatibility changes to wire structs (6007)
3313 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3314 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3315 - improved "lustre status" script
3316 - initialize blocksize for non-regular files (6062)
3317 - added --disable-server and --disable-client configure options (5782)
3318 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3319 - Vanilla 2.4.29 support
3320 - increase maximum number of obd devices to 520 (6242)
3321 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3322 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3323 SLES 9 2.6.5 kernels
3324 - limit stripes per file to 160 (the maximum EA size) (6093)
3326 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3329 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3330 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3331 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3332 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3333 - small ext3 extents cleanups and fixes (5733)
3334 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3335 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3336 - avoid needless client->OST connect, fix handle mismatch (5317)
3337 - fix DLM error path that led to out-of-sync client, long delays (5779)
3338 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3339 - fix several locking issues related to i_size (5492,5624,5654,5672)
3340 - don't move pending lock onto export if it is already evicted (5683)
3341 - fix kernel oops when creating .foo in unlinked directory (5548)
3342 - fix deadlock in obdfilter statistics vs. object create (5811)
3343 - use time_{before,after} to avoid timer jiffies wrap (5882)
3344 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3345 - Do not leak request if server was not able to process it (5154)
3346 - If mds_open unable to find parent dir, make that negative lookup(5154)
3347 - don't create new directories with extent-mapping (5909, 5936)
3349 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3350 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3351 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3352 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3353 - support for RHEL4 kernel on i686 (5773)
3354 - provide error messages when incompatible logs are encountered (5898)
3356 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3357 * version 1.4.0.10 (1.4.1 release candidate 1)
3359 - don't keep a lock reference when lock is not granted (4238)
3360 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3361 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3362 - drop import inflight refcount on signal_completed_replay error (5255)
3363 - unlock page after async write error during send (3677)
3364 - handle missing objects in filter_preprw_read properly (5265)
3365 - no transno return for symlink open, don't save no-trasno open (3440)
3366 - don't try to complete elan receive that already failed (4012)
3367 - free RPC server reply state on error (5406)
3368 - clean up thread from ptlrpc_start_thread() on error (5160)
3369 - readahead could read extra page into cache that wasn't ejected (5388)
3370 - prevent races in class_attach/setup/cleanup/detach (5260)
3371 - don't dereference de->d_inode after l_dput of de (5458)
3372 - use "int" for stripe value returned from lock_to_stripe (5544)
3373 - mballoc allocation and error-checking fixes in 2.6 (5504)
3374 - block device patches to fix I/O request sizes in 2.6 (5482)
3375 - look up hostnames for IB nals (5602)
3376 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3377 - don't start multiple acceptors for the same port (5277)
3378 - fix incorrect LASSERT in mds_getattr_name (5635)
3379 - export a proc file for general "ping" checking (5628)
3380 - fix "lfs check" to not block when the MDS is down (5628)
3382 - service request history (4965)
3383 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3384 - create an "evict_client" /proc entry on OSTs, like the MDS has
3385 - fix mount usage message, return errors per mount(8) (5168)
3386 - change grep [] to grep "[]" in tests so they work in more UMLs
3387 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3388 - remove ancient LOV_MAGIC_V0 EA support (5047)
3389 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3390 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3391 - allow readahead allocations to fail when low on memory (5383)
3392 - mmap locking landed again, after considerable improvement (2828)
3393 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3395 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3398 - send OST transaction number in read/write reply to free req (4966)
3399 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3400 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3401 - the watchdog thread now runs as interruptible (5246)
3402 - flock/lockf fixes (but it's still disabled, pending 5135)
3403 - don't use EXT3 constants in llite code (5094)
3404 - memory shortage at startup could cause assertion (5176)
git://git.whamcloud.com / fs / lustre-release.git / blob