4 2.6.9-55.0.9.EL (RHEL 4),
5 2.6.16.53-0.8 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: Remove 2.4 kernel support from lustre 1.8.0
18 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
19 quota_interface.c in quota subsystem.
20 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
22 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
23 lvfs_linux.c upcall_cache.c in lvfs subsystem.
24 Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
25 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
27 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
28 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
30 Remove 2.4 kernel check from lproc_osc.c, osc_request.c in osc
32 Remove 2.4 kernel check from filter.c, lproc_obdfilter.c in
34 Remove 2.4 kernel code from lustre/obdecho/echo_client.c.
35 Remove 2.4 kernel check from lustre/mgs/lproc_mgs.c,
37 Remove 2.4 kernel code from lustre/mgc/mgc_request.c.
39 Severity : enhancement
41 Description: Build SLES10 patchless client fails
42 Details : The configure was broken by run ./configure with
43 --with-linux-obj=.... argument for patchless client. When the
44 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
45 can't search header adequately. Use absolute path such as
46 -I($LINUX)/include instead.
50 Description: interrupt oig_wait produce painc on resend.
51 Details : brw_redo_request can be used for resend requests from ptlrpcd and
52 private set, and this produce situation when rq_ptlrpcd_data not
53 copyed to new allocated request and triggered LBUG on assert
54 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
55 ptlrpcd set if request is changed and can be safety changed to use
60 Description: LASSERT_{REQ,REP}SWAB macros are buggy
61 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
62 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
63 these macros and replace them with their difinitions instead.
67 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
68 Details : Update which_patch & target file for SLES10 latest kernel.
71 Description: Remove client patches from SLES 10 kernel.
72 Details : This causes SLES 10 clients to behave as patchless clients
73 even on a Lustre-patched (server) kernel.
75 Severity : enhancement
77 Description: self-adjustable client's lru lists
78 Details : use adaptive algorithm for managing client cached locks lru
79 lists according to current server load, other client's work
80 pattern, memory activities, etc. Both, server and client
81 side namespaces provide number of proc tunables for controlling
84 Severity : enhancement
86 Description: light-weight GSS support
87 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
88 and reduce performance overhead.
90 Severity : enhancement
92 Description: Linux keyring support
93 Details : Support using service of Linux keyring for Lustre GSS internal
94 context refresh/cache mechanism.
98 Description: Fix errors in lfs documentation
99 Details : Fixes man pages
103 Description: don't use GFP_* in generic Lustre code.
104 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
108 Description: obdclass is limited by single OBD_ALLOC(idarray)
109 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
113 Description: warnings with build patchless client with vanila 2.6.19 and up
114 Details : change old ctl_table style and replace ctl_table/ctl_table_header
115 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
119 Frequency : Only for SLES
120 Description: sanity_quota fail test_1
121 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
122 file, which makes TSTID[2] inunique.
126 Frequency : Only for SLES9
127 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
128 Details : Update target/ChangeLog/which_patch .
132 Frequency : Only for test_10 in sanity-quota.sh
133 Description: a bug in quota test code
134 Details : checking a test flag in a wrong place
138 Frequency : Only for test_13 in sanity-quota.sh
139 Description: a bug in quota test script
140 Details : shouldn't check overfull allocated quota
144 Description: jbd statistics
145 Details : Port older jbd statistics patch for sles10
147 Severity : enhancement
149 Description: Add printing inode info into message about error in writepage.
154 Description: reply_lock_interpret crash due to race with it and lock cancel.
155 Details : Do not replay locks that are being cancelled. Do not reference
156 locks by their address during replay, just by their handle.
160 Frequency : When flocks are used.
161 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
162 Details : Flock locks might destroy just granted lock if it could be merged
163 with another existing flock, this is done in completion handler,
164 so teach ldlm_cli_enquque_fini that this is a valid case for
170 Description: Oops in read and write path when failing to allocate lock.
171 Details : Check if lock allocation failed and return error back.
175 Description: lstripe command fails for valid OST index
176 Details : The stripe offset is compared to lov->desc.ld_tgt_count
177 instead of lov->desc.ld_active_tgt_count.
181 Description: sanity.sh failed test 103
182 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
187 Description: df doesn't work properly if diskfs blocksize != 4K
188 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
192 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
193 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
197 Description: obd_health_check_timeout too short
198 Details : set obd_health_check_timeout as 1.5x of obd_timeout
202 Description: llapi_file_create() does not allow some changes
203 Details : add llapi_file_open() that allows specifying the mode and
204 open flags, and also returns an open file handle.
208 Description: merge and cleanup kernel patches.
209 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
213 Description: Add journal checksum support.(Kernel part)
214 Details : The journal checksum feature adds two new flags i.e
215 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
216 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
217 indicates that the commit block contains the checksum for
218 the blocks described by the descriptor blocks. Now commit
219 record can be sent to disk without waiting for descriptor
220 blocks to be written to disk. This behavior is controlled
221 using JBD2_FEATURE_ASYNC_COMMIT flag.
225 Description: OSS needs mutliple precreate threads
226 Details : Add ability to start more than one create thread per OSS.
230 Description: Sanity.sh test_65k failed: setstripe should have succeeded
231 Details : execute lfs setstripe on client
235 Description: mds_obd_create error creating tmp object
236 Details : When the user sets quota on root, llog will be affected and can't
237 create files and write files.
240 Frequency : Always on ia64 patchless client, and possibly others.
242 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
243 Details : This allows the patchless client to be loaded on architectures
248 Description: RedHat Update kernel for RHEL5
249 Details : Add the kernel config file for RHEL5.
253 Description: RedHat Update kernel for RHEL5
254 Details : Modify the kernel config file more closer RHEL5.
258 Description: Build failure against Centos5 (RHEL5)
259 Details : Define PAGE_SIZE when it isn't present.
263 Description: test_7 on sanity-quota.sh can't be used on separate machines
264 Details : add facet to handle it
268 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
269 Details : with NFS, the anon dentry's parent was set to itself in
270 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
271 corresponding dentry other than use rec->ur_name.
273 Severity : enhancement
275 Description: lfs setstripe enhancement
276 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
280 Description: testing performance impact of enabling checksumming
281 Details : enable checksum by default, allow --disable-checksum
282 configure option and "-o nochecksum" mount option
286 Description: lustre support for RHEL5
287 Details : Add support for RHEL5.
291 Description: Client eviction due to failover config
292 Details : after a connection loss, the lustre client should attempt to
293 reconnect to the last active server first before trying the
294 other potential connections.
298 Description: when mds and osts use different quota unit(32bit and 64bit),
299 quota will be released repeatly.
300 Details : void sending multiple quota reqs to mds, which will keep the
301 status between the reqs.
305 Description: osts not allocated evenly to files
306 Details : change the condition to increase offset_idx
309 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
311 Description: Data corruption for OSTs that are formatted larger than 2TB
313 Details : When generating the bio request for lustre file writes the
314 sector number would overflow a temporary variable before being
315 used for the IO. The data reads correctly from Lustre (which
316 will overflow in a similar manner) but other file data or
317 filesystem metadata may be corrupted in some cases.
321 Description: Tune the kernel for good SCSI performance.
322 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
323 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
328 Description: rewrite ext2-derived code in obdclass/uuid.c
329 Details : rewrite inherited code (uuid parsing code from ext2 utils)
330 from scratch preserving functionality.
334 Description: "lfs find -obd UUID" prints directories
335 Details : "lfs find -obd UUID" will return all directory names instead
336 of just file names. It is incorrect because the directories
337 do not reside on the OSTs.
341 Description: use i_size_read and i_size_write in 2.6 port
342 Details : replace inode->i_size access with i_size_read/write()
345 Frequency : when removing large files
347 Description: scheduling issue during removal of large Lustre files
348 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
349 It causes scheduling issues when removing large files (17TB in the
353 Frequency : only with liblustre clients on XT3
355 Description: evictions taking too long
356 Details : allow llrd to evict clients directly on OSTs
359 Frequency : only on ppc
361 Description: /proc/fs/lustre/devices broken on ppc
362 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
363 We should make sure the type of 'index' is loff_t and then cast
364 explicitly as needed below. Do not assign an explicitly cast
368 Frequency : only for rhel5
370 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
371 Details : Modify the target file & which_kernel.
373 Severity : enhancement
375 Description: omit set fsid for export NFS
376 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
377 as FS_REQUIRES_DEV for avoid problems with generate fsid.
381 Frequency : Always, for kernels after 2.6.16
382 Description: Fix warning idr_remove called for id=.. which is not
384 Details : Last kernels save old s_dev before kill super and not allow
385 to restore from callback - restore it before call kill_anon_super.
389 Description: buffer overruns could theoretically occur
390 Details : llapi_semantic_traverse() modifies the "path" argument by
391 appending values to the end of the origin string, and a
392 overrun may occur. Adding buffer overrun check in liblustreapi.
396 Description: Fix error on 'ls .' at the top of the Lustre mount.
397 Details : Don't revalidate dentry if it is a root dentry.
401 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
402 Details : Modify vm-tunables-rhel4.patch.
406 Description: Kernel config for 2.6.18-vanilla.
407 Details : Modify targets/2.6-vanilla.target.in.
408 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
409 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
410 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
411 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
415 Description: improve handling recoverable errors
416 Details : if request processig with error which can be recoverable on server
417 request should be resend, otherwise page released from cache and
422 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
423 Details : Modify target file & which_patch.
424 A flaw was found in the IA32 system call emulation provided
425 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
426 value could be stored in the %RAX register, which could trigger an
427 out-of-bounds system call table access. An untrusted local user
428 could exploit this flaw to run code in the kernel
429 (ie a root privilege escalation). (CVE-2007-4573).
433 Description: O_DIRECT bypasses client statistics.
434 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
435 read_ahead_stats were updated. copied the stats section from
436 osc_send_oap_rpc() into async_internal().
440 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
444 Description: change order of libsysio includes
445 Details : '#include sysio.h' should always come before '#include xtio.h'
449 Description: handle "serious error: objid * already exists" more gracefully
450 Details : If LAST_ID value on disk is smaller than the objects existing in
451 the O/0/d* directories, it indicates disk corruption and causes an
452 LBUG(). If the object is 0-length, then we should use the existing
453 object. This will help to avoid a full fsck in most cases.
455 Severity : enhancement
457 Description: adapt the lustre_config script to support the upgrade case
458 Details : Add "-u" option for lustre_config script to support upgrading 1.4
459 server targets to 1.6 in parallel.
463 Description: To avoid grant space > avaible space when the disk is almost
464 full. Without this patch you might see the error "grant XXXX >
465 available" or some LBUG about grant, when the disk is almost
467 Details : In filter_check_grant, for non_grant cache write, we should
468 check the left space by if (*left > ungranted + bytes), instead
469 of (*left > ungranted), because only we are sure the left space
470 is enough for another "bytes", then the ungrant space should be
471 increase. In client, we should update cl_avail_grant only there
472 is OBD_MD_FLGRANT in the reply.
476 Description: Update RHEL 4 kernel to fix local root privilege escalation.
477 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
478 described in CVE-2007-4573. This problem could allow untrusted
479 local users to gain root access.
483 Frequency : when using O_DIRECT and quotas
485 Description: Incorrect file ownership on O_DIRECT output files
486 Details : block usage reported by 'lfs quota' does not take into account
487 files that have been written with O_DIRECT.
489 --------------------------------------------------------------------------------
491 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
493 * Support for kernels:
494 2.6.5-7.283 (SLES 9),
495 2.6.9-55.EL (RHEL 4),
496 2.6.16.46-0.14 (SLES 10),
497 2.6.18.8 vanilla (kernel.org)
498 * Client support for unpatched kernels:
499 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
500 2.6.16 - 2.6.22 vanilla (kernel.org)
501 * Due to recently discovered recovery problems, we do not recommend
502 using patchless RHEL 4 clients with this or any earlier release.
503 * Recommended e2fsprogs version: 1.39.cfs8
504 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
505 * Starting with this release, the ldiskfs backing filesystem required
506 by Lustre is now in its own package, lustre-ldiskfs. This package
507 should be installed. It is versioned separately from Lustre and
508 may be released separately in future.
512 Description: block reactivating mgc import until all deactivates complete
513 Details : Fix race when failing back MDT/MGS to itself (testing)
516 Frequency : at statup only
518 Description: mds_lov_synchronize race leads to various problems
519 Details : simultaneous MDT->OST connections at startup can cause the
520 sync to abort, leaving the OSC in a bad state.
522 Severity : enhancement
524 Description: add optional extra BUILD_VERSION info
525 Details : add a new environment variable (namely LUSTRE_VERS) which allows
526 to override the lustre version.
529 Frequency : 2.6.18 servers only
531 Description: ll_kern_mount() doesn't release the module reference
532 Details : The ldiskfs module reference count never drops down to 0
533 because ll_kern_mount() doesn't release the module reference.
538 Description: server LBUG when using old ost_num_threads parameter
539 Details : Accept the old ost_num_threads parameter but warn that it
540 is deprecated, and fix an off-by-one error that caused an LBUG.
545 Description: Transient SCSI error results in persistent IO issue
546 Details : iobuf->dr_error is not reinitialized to 0 between two
550 Frequency : sometimes when underlying device returns I/O errors
552 Description: OSTs not going read-only during write failures
553 Details : OSTs are not remounted read-only when the journal commit threads
554 get I/O errors because fsfilt_ext3 calls journal_start/stop()
555 instead of the ext3 wrappers.
559 Description: poor connect scaling with increasing client count
560 Details : Don't run filter_grant_sanity_check for more than 100 exports
561 to improve scaling for large numbers of clients.
564 Frequency : SLES10 only
566 Description: sanity-quota.sh quotacheck failed: rc = -22
567 Details : Quotas cannot be enabled on SLES10.
570 Frequency : liblustre clients only
572 Description: getdirentries does not give error when run on compute nodes
573 Details : getdirentries does not fail when the size specified as an argument
574 is too small to contain at least one entry
576 Severity : enhancement
578 Description: Add LNET router traceability for debug purposes
579 Details : If a checksum failure occurs with a router as part of the
580 IO path, the NID of the last router that forwarded the bulk data
581 is printed so it can be identified.
586 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
587 Details : Due to a race condition, liblustre clients were occasionally
590 Severity : enhancement
592 Description: lfs setstripe use optional parameters instead of postional
595 Severity : enhancement
597 Description: Nanosecond timestamp support for ldiskfs
598 Details : The on-disk ldiskfs filesystem has added support for nanosecond
599 resolution timestamps. There is not yet support for this at
600 the Lustre filesystem level.
603 Frequency : during server recovery
605 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
606 Details : request with rq_no_resend flag not awake l_wait_event if they get a
610 Frequency : nfs export on patchless client
612 Description: connectathon hang when test nfs export over patchless client
613 Details : Disconnected dentry cannot be found with lookup, so we do not need
614 to unhash it or make it invalid
617 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
618 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
619 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
624 Description: open req refcounting wrong on reconnect
625 Details : If reconnect happened between getting open reply from server and
626 call to mdc_set_replay_data in ll_file_open, we will schedule
627 replay for unreferenced request that we are about to free.
628 Subsequent close will crash in variety of ways.
629 Check that request is still eligible for replay in
630 mdc_set_replay_data().
635 Description: disable writes to filesystem when reading health_check file
636 Details : the default for reading the health_check proc file has changed
637 to NOT do a journal transaction and write to disk, because this
638 can cause reads of the /proc file to hang and block HA state
639 checking on a healthy but otherwise heavily loaded system. It
640 is possible to return to the previous behaviour during configure
641 with --enable-health-write.
643 Severity : enhancement
645 Description: 64-bit inode version
646 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
647 to the inode. This will be required for version-based recovery.
652 Description: MDS fails to start if a duplicate client export is detected
653 Details : in some rare cases it was possible for a client to connect to
654 an MDS multiple times. Upon recovery the MDS would detect this
655 and fail during startup. Handle this more gracefully.
657 Severity : enhancement
659 Description: Add -o localflock option to simulate old noflock
661 Details : This will achieve local-only flock/fcntl locks
667 Description: log_commit_thread vs filter_destroy race leads to crash
668 Details : Take import reference before releasing llog record semaphore
673 Description: Wrong request locking in request set processing
674 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
675 imp_delayed_list, in this place should be used imp_lock.
678 Frequency : when reconnection
680 Description: Grant Leak when osc reconnect to OST
681 Details : When osc reconnect ost, OST(filter) should check whether it
682 should grant more space to client by comparing fed_grant and
683 cl_avail_grant, and return the granted space to client instead
684 of "new granted" space, because client will call osc_init_grant
685 to update the client grant space info.
688 Frequency : when client reconnect to OST
690 Description: Grant Leak when osc do resend and replay bulk write
691 Details : When osc reconnect to OST, OST(filter)should clear grant info of
692 bulk write request, because the grant info while be sync between
693 OSC and OST when reconnect, and we should ignore the grant info
694 these of resend/replay write req.
699 Description: Grant space more than avaiable left space sometimes.
700 Details : When then OST is about to be full, if two bulk writing from
701 different clients came to OST. Accord the avaliable space of the
702 OST, the first req should be permitted, and the second one
703 should be denied by ENOSPC. But if the seconde arrived before
704 the first one is commited. The OST might wrongly permit second
705 writing, which will cause grant space > avaiable space.
708 Frequency : when client is evicted
710 Description: Grant might be wrongly erased when osc is evicted by OST
711 Details : when the import is evicted by server, it will fork another
712 thread ptlrpc_invalidate_import_thread to invalidate the
713 import, where the grant will be set to 0. While the original
714 thread will update the grant it got when connecting. So if
715 the former happened latter, the grant will be wrongly errased
716 because of this race.
721 Description: Checking Stale with correct fid
722 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
723 to check whether it is stale, because sometimes, we want the
724 enqueue happened anyway, and op_data.fid2 will not be initialized.
726 Severity : enhancement
728 Description: update patchless client
729 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
732 Frequency : only with 2.4 kernel
734 Description: random memory corruption
735 Details : size of struct ll_inode_info is to big for union inode.u and this
736 can be cause of random memory corruption.
741 Description: Memory leak in recovery
742 Details : Lov_mds_md was not free in an error handler in mds_create_object.
743 It should also check obd_fail before fsfilt_start, otherwise if
744 fsfilt_start return -EROFS,(failover mds during mds recovery).
745 then the req will return with repmsg->transno = 0 and rc = EROFS.
746 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
747 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
748 be freed no matter whether fsfilt_commit success or not.
751 Frequency : only with huge count clients
753 Description: Prevents from taking the superblock lock in llap_from_page for
755 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
756 for llap_from_page call in ll_removepage prevents from taking the
757 superblock lock for a soon died page.
762 Description: Not check open intent error before release open handle
763 Details : in some rare cases, the open intent error is not checked before
764 release open handle, which may cause
765 ASSERTION(open_req->rq_transno != 0), because it tries to release
766 the failed open handle.
771 Description: Set cat log bitmap only after create log success.
772 Details : in some rare cases, the cat log bitmap is set too early. and it
773 should be set only after create log success.
777 Description: Accessing a block bevice can re-enable I/O when Lustre is
778 tearing down a device.
779 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
784 Description: service threads may hog cpus when there are a lot of requests
786 Details : Insert cond_resched to give other threads a chance to use some of
792 Description: the cat log was not initialized in recovery
793 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
794 unlink log on mds will not be initialized until mds post
795 recovery. And also in mds post recovery, the unlink log will
796 initialization will be done asynchronausly, so there will be race
797 between add unlink log and unlink log initialization.
801 Description: brw_stats were being printed incorrectly
802 Details : brw_stats were being printed as log2 but all of them were not
803 recorded as log2. Also remove some code duplication arising from
804 filter_tally_{read,write}.
808 Frequency : rare, only in recovery.
809 Description: ASSERTION(req->rq_type != LI_POISON) failed
810 Details : imp_lock should be held while iterating over imp_sending_list for
811 prevent destroy request after get timeout in ptlrpc_queue_wait.
815 Description: replay-single.sh test 52 fails
816 Details : A lock's skiplist need to be cleanup when it being unlinked
817 from its resource list.
821 Description: Short directio read returns full requested size rather than
823 Details : Direct I/O operations should return actual amount of bytes
824 transferred rather than requested size.
826 Severity : enhancement
828 Description: metadata RPC reduction (e.g. for rm performance)
829 Details : decrease the amount of synchronous RPC between clients and servers
830 by canceling conflicing lock before the operation on the client side
831 and packing thier handles into the main operation RPC to server.
833 Severity : enhancement
835 Description: Async OSC create to avoid the blocking unnecessarily.
836 Details : If a OST has no remain object, system will block on the creating
837 when need to create a new object on this OST. Now, ways use
838 pre-created objects when available, instead of blocking on an
839 empty osc while others are not empty. If we must block, we block
840 for the shortest possible period of time.
842 --------------------------------------------------------------------------------
844 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
849 Frequency : on some architectures
851 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
852 Details : Uninitialized flags sometimes cause configuration commands to
856 Frequency : patchless clients only
858 Description: missing __iget() symbol export
859 Details : The __iget() symbol export is missing. To avoid the need for
860 this on patchless clients the deathrow inode reaper is turned
861 off, and we depend on the VM to clean up old inodes. This
862 dependency was during via the fix for bug 12181.
864 --------------------------------------------------------------------------------
866 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
868 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
869 INTEROPERATE with older versions automatically. In many cases a
870 special upgrade step is needed. Please read the
871 user documentation before upgrading any part of a live system.
872 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
873 version will not interoperate with 1.6 betas before beta5 (1.5.95).
874 * WARNING: Lustre configuration and startup changes are required with
875 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
881 Severity : enhancement
883 Description: MountConf
884 Details : Lustre configuration is now managed via mkfs and mount
885 commands instead of lmc and lconf. New obd types (MGS, MGC)
886 are added for dynamic configuration management. See
887 https://mail.clusterfs.com/wikis/lustre/MountConf for
890 Severity : enhancement
892 Description: dynamic OST addition
893 Details : OSTs can now be added to a live filesystem
895 Severity : enhancement
897 Description: startup order invariance
898 Details : MDTs and OSTs can be started in any order. Clients only
899 require the MDT to complete startup.
901 Severity : enhancement
903 Description: parallel, asynchronous orphan cleanup
904 Details : orphan cleanup is now performed in separate threads for each
905 OST, allowing parallel non-blocking operation.
907 Severity : enhancement
909 Description: optimized stripe assignment
910 Details : stripe assignments are now made based on ost space available,
911 ost previous usage, and OSS previous usage, in order to try
912 to optimize storage space and networking resources.
914 Severity : enhancement
916 Description: Permanently set tunables
917 Details : All writable /proc/fs/lustre tunables can now be permanently
918 set on a per-server basis, at mkfs time or on a live
921 Severity : enhancement
923 Description: Lustre message v2
924 Details : Add lustre message format v2.
926 Severity : enhancement
928 Description: client OST exclusion list
929 Details : Clients can be started with a list of OSTs that should be
930 declared "inactive" for known non-responsive OSTs.
934 Description: SPEC SFS validation failure on NFS v2 over lustre.
935 Details : Changes the blocksize for regular files to be 2x RPC size,
936 and not depend on stripe size.
938 Severity : enhancement
940 Description: Multiple MD RPCs in flight.
941 Details : Further unserialise some read-only MDS RPCs - learn about intents.
942 To avoid overly-overloading MDS, introduce a limit on number of
943 MDS RPCs in flight for a single client and add /proc controls
944 to adjust this limit.
946 Severity : enhancement
948 Description: client read/write statistics
949 Details : Add client read/write call usage stats for performance
950 analysis of user processes.
951 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
952 file access. extents_stats shows chunk size distribution.
953 extents_stats_per_process show chunk size distribution per
956 Severity : enhancement
958 Description: mds statistics
959 Details : Add detailed mds operations statistics in
960 /proc/fs/lustre/mds/*/stats.
964 Description: Failure of copying files with lustre special EAs.
965 Details : Client side always return success for setxattr call for lustre
966 special xattr (currently only "trusted.lov").
971 Description: Refcount LNET uuids
972 Details : The global LNET uuid list grew linearly with every startup;
973 refcount repeated list entries instead of always adding to
976 Severity : enhancement
978 Description: Dynamic service threads
979 Details : Within a small range, start extra service threads
980 automatically when the request queue builds up.
983 Frequency : mixed-endian client/server environments
985 Description: mixed-endian crashes
986 Details : The new msg_v2 system had some failures in mixed-endian
989 Severity : enhancement
991 Description: Easy OST removal
992 Details : OSTs can be permanently deactivated with e.g. 'lctl
993 conf_param lustre-OST0001.osc.active=0'
995 Severity : enhancement
997 Description: MGS proc entries
998 Details : Added basic proc entries for the MGS showing what filesystems
1001 Severity : enhancement
1003 Description: provide MGS failover
1004 Details : Added config lock reacquisition after MGS server failover.
1006 Severity : enhancement
1008 Description: add Linux 2.4 support
1009 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1013 Description: a large application tries to do I/O to the same resource and dies
1014 in the middle of it.
1015 Details : Check the req->rq_arrival time after the call to
1016 ost_brw_lock_get(), but before we do anything about
1017 processing it & sending the BULK transfer request. This
1018 should help move old stale pending locks off the queue as
1019 quickly as obd_timeout.
1022 Frequency : when an incorrect nid is specified during startup
1024 Description: ptlrpc connect to non-existant node causes kernel crash
1025 Details : LNET can't be re-entered from an event callback, which
1026 happened when we expire a message after the export has been
1027 cleaned up. Instead, hand the zombie cleanup off to another
1030 Severity : enhancement
1032 Description: plain/inodebits lock performance improvement
1033 Details : Grouping plain/inodebits in granted list by their request modes
1034 and bits policy, thus improving the performance of search through
1038 Frequency : only if OST filesystem is corrupted
1040 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1041 Details : for a short time RPCs with bulk IO are in the replay list,
1042 but replay of bulk IOs is unimplemented. If the OST filesystem
1043 is corrupted due to disk cache incoherency and then replay is
1044 started it is possible to trip an assertion. Avoid putting
1045 committed RPCs into the replay list at all to avoid this issue.
1048 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1050 Description: System hang on startup
1051 Details : This bug allowed the liblustre (e.g. catamount) client to
1052 return to the app before handling all startup RPCs. This
1053 could leave the node unresponsive to lustre network traffic
1054 and manifested as a server ptllnd timeout.
1056 Severity : enhancement
1058 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1059 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1060 boolean to control whether to print peer debug info when a
1061 client's RPC times out.
1064 Frequency : only for kernels with patches from Lustre below 1.4.3
1066 Description: Remove old rdonly API
1067 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1070 Frequency : only for devices with external journals
1072 Description: Set external device read-only also
1073 Details : During a commanded failover stop, we set the disk device
1074 read-only while the server shuts down. We now also set any
1075 external journal device read-only at the same time.
1078 Frequency : when upgrading from 1.4 while trying to change parameters
1080 Description: The wrong (new) MDC name was used when setting parameters for
1081 upgraded MDT's. Also allows changing of OSC (and MDC)
1082 parameters if --writeconf is specified at tunefs upgrade time.
1085 Frequency : when setting specific ost indicies
1087 Description: QOS code breaks on skipped indicies
1088 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1089 created with --index need not be sequential.
1091 Severity : enhancement
1093 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1094 Details : The uninit_groups feature works in conjunction with the kernel
1095 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1096 up the pass1 processing of e2fsck. This is a read-only feature
1097 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1098 mount filesystems that have had this feature enabled.
1100 Severity : enhancement
1102 Description: Improve multi-block allocation algorithm to avoid fragmentation
1103 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1104 allocation locality and avoid filesystem fragmentation.
1106 ------------------------------------------------------------------------------
1108 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1110 * Support for kernels:
1111 2.4.21-47.0.1.EL (RHEL 3)
1112 2.6.5-7.283 (SLES 9)
1113 2.6.9-42.0.10.EL (RHEL 4)
1114 2.6.12.6 vanilla (kernel.org)
1115 2.6.16.27-0.9 (SLES 10)
1116 * Recommended e2fsprogs version: 1.39.cfs5
1118 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1122 Frequency : occasional, depends on client load and configuration
1123 Bugzilla : 12181, 12203
1124 Description: data loss for recently-modified files
1126 Details : In some cases it is possible that recently written or created
1127 files may not be written to disk in a timely manner (this should
1128 normally be within 30s unless client IO load is very high).
1129 The problem appears as zero-length files or files that are a
1130 multiple of 1MB in size after a client crash or client eviction
1131 that are missing data at the end of the file.
1133 This problem is more likely to be hit on clients where files are
1134 repeatedly created and unlinked in the same directory, clients
1135 have a large amount of RAM, have many CPUs, the filesystem has
1136 many OSTs, the clients are rebooted frequently, and/or the files
1137 are not accessed by other nodes after being written.
1139 The presence of the problem can be detected by looking at
1140 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1141 smaller than the second (nr_unused) then dirty files will not
1142 be flushed automatically to disk. "sync; sleep 10" should be
1143 run several times on the node before unmounting it to update
1144 Lustre (this is also safe to run on nodes without this problem).
1146 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1147 that can cause this same problem, so customers using that kernel
1148 also need to update the kernel in addition to Lustre. In order
1149 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1151 It is normal that files written just before a client crash (less
1152 than 30s) may not yet have been flushed to disk, even for local
1156 Frequency : frequent on thin XT3 nodes
1158 Description: UUID collision on thin XT3 Linux nodes
1159 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1160 randomly, since we relied on an insufficiently-seeded PRNG.
1165 Description: OSS hangs after "All ost request buffers busy"
1166 Details : A deadlock between quota and journal operations caused OSS
1167 hangs after printing "All ost request buffers busy."
1170 Frequency : always on liblustre builds
1172 Description: Cleanup compiler warnings on liblustre
1175 Frequency : always on liblustre builds on XT3
1177 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1181 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1182 > 32000 subdirectories
1183 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1184 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1185 subdirectory count crosses 32000. This will aid e2fsck to
1186 correctly handle more than 32000 subdirectories.
1189 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1191 Description: System hang on startup
1192 Details : This bug allowed the liblustre (e.g. catamount) client to
1193 return to the app before handling all startup RPCs. This
1194 could leave the node unresponsive to lustre network traffic
1195 and manifested as a server ptllnd timeout.
1197 Severity : enhancement
1199 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1200 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1201 boolean to control whether to print peer debug info when a
1202 client's RPC times out.
1207 Description: make O_SYNC working on 2.6 kernels
1208 Details : 2.6 kernels use different method for mark pages for write,
1209 so need add a code to lustre for O_SYNC work.
1214 Description: Failure to close file and release space on NFS
1215 Details : Put inode details into lock acquired in ll_intent_file_open.
1216 Use mdc_intent_lock in ll_intent_open to properly
1217 detect all kind of errors unhandled by mdc_enqueue.
1222 Description: proc file read during shutdown sometimes raced obd removal,
1224 Details : Add lock to prevent obd access after proc file removal.
1227 Frequency : Only for files larger than 4GB on 32-bit clients.
1229 Description: improperly doing page alignment of locks
1230 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1231 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1236 Description: RPCs being resent when they shouldn't be
1237 Details : Some RPCs that should not be resent are being resent. This
1238 can cause inconsistencies in the RPC state machine. Do not
1239 resend such requests.
1242 Frequency : rare, only with NFS export
1244 Description: Crash on NFS re-export node
1245 Details : under very unusual load conditions an assertion is hit in
1246 ll_intent_file_open()
1249 Frequency : only if OST filesystem is corrupted
1251 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1252 Details : for a short time RPCs with bulk IO are in the replay list,
1253 but replay of bulk IOs is unimplemented. If the OST filesystem
1254 is corrupted due to disk cache incoherency and then replay is
1255 started it is possible to trip an assertion. Avoid putting
1256 committed RPCs into the replay list at all to avoid this issue.
1261 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1262 Details : Large single O_DIRECT read and write calls can fail to allocate
1263 a sufficiently large buffer to process the request. In case of
1264 allocation failure the allocation is retried with a smaller
1265 buffer and broken into smaller requests.
1267 Severity : enhancement
1269 Description: Add -o localflock option to simulate old noflock behaviour.
1270 Details : This will achieve local-only flock/fcntl locks coherentness.
1275 Description: versioning check is incomplete
1276 Details : Checking the version difference of client vs. server, report
1277 error if the gap is too big.
1282 Description: add support PG_writeback bit
1283 Details : add support for PG_writeback bit for Lustre, for more carefull
1284 work with page cache in 2.6 kernel. This also fix some deadlocks
1285 and remove hack for work O_SYNC with 2.6 kernel.
1287 Severity : enhancement
1289 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1290 Details : The uninit_groups feature works in conjunction with the kernel
1291 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1292 up the pass1 processing of e2fsck. This is a read-only feature
1293 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1294 mount filesystems that have had this feature enabled.
1296 Severity : enhancement
1298 Description: Improve multi-block allocation algorithm to avoid fragmentation
1299 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1300 allocation locality and avoid filesystem fragmentation.
1302 ------------------------------------------------------------------------------
1304 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1306 * Support for kernels:
1307 2.6.9-42.0.3.EL (RHEL 4)
1308 2.6.5-7.276 (SLES 9)
1309 2.4.21-47.0.1.EL (RHEL 3)
1310 2.6.12.6 vanilla (kernel.org)
1311 2.6.16.21-0.8 (SLES10)
1312 * Recommended e2fsprogs version: 1.39.cfs2-0
1314 * The backwards-compatible /proc/sys/portals symlink has been removed
1315 in this release. Before upgrading, please ensure that you change
1316 any configuration scripts or /etc/sysctl.conf files that access
1317 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1318 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1319 in advance of the upgrade on any system running Lustre 1.4.6 or
1320 newer, since /proc/sys/lnet was added in that version.
1321 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1325 Frequency : only when quota is used
1327 Description: avoid scanning export list for quota master
1328 Details : Change the algorithms to avoid scanning export list in order
1329 to improve the efficiency.
1332 Frequency : MDS failover only, very rarely
1334 Description: "went back in time" messages on mds failover
1335 Details : The greatest transno may be lost when the current operation
1336 finishes with an error (transno==0) and the client's last_rcvd
1337 record is over-written. Save the greatest transno in the
1338 mds_last_transno for this case.
1341 Frequency : always for specific kernels and striping counts
1343 Description: client may get "Matching packet too big" without ACL support
1344 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1345 when trying to access files in certain configurations. The
1346 clients should in fact be denied when mounting because they do
1347 not understand ACLs.
1350 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1352 Description: many clients connecting with IO in progress causes connect timeouts
1353 Details : Avoid synchronous journal commits to avoid delays caused by many
1354 clients connecting/disconnecting when bulk IO is in progress.
1355 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1356 OST_IO_PORTAL to avoid delays behind potentially many pending
1360 Frequency : occasionally with multiple writers to a single file
1362 Description: shared writes to file may result in wrong size reported by stat()
1363 Details : Allow growing of kms when extent lock is cancelled
1366 Frequency : always with random mmap IO to multi-striped file
1368 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1369 Details : Only if the hole is at the end of OST object so that kms is too
1370 small. Fix is to increase kms accordingly in ll_nopage.
1373 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1375 Description: writes to a missing object would leak memory on the OST
1376 Details : If there is an inconsistency between the MDS and OST filesystems,
1377 such that the MDS references an object that doesn't exist, writes
1378 to that object will leak memory due to incorrect cleanup in the
1379 error handling path, eventually running out of memory on the OST.
1384 Description: Creating too long symlink causes lustre errors
1385 Details : Check symlink and name lengths before sending requests to MDS.
1388 Frequency : only if flock is enabled (not on by default)
1390 Description: posix locks not released on fd closure on 2.6.9+
1391 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1392 kernels, this caused such locks not to be released on fd close and
1393 then assertions on fs unmount about still used locks.
1396 Frequency : MDS failover only, very rarely
1398 Description: clients may get ASSERTION(granted_lock != NULL)
1399 Details : When request was taking a long time, and a client was resending
1400 a getattr by name lock request. The were multiple lock requests
1401 with the same client lock handle and
1402 mds_getattr_name->fixup_handle_for_resent_request found one of the
1403 lock handles but later failed with ASSERTION(granted_lock != NULL).
1408 Description: handle->h_buffer_credits > 0, assertion failure
1409 Details : h_buffer_credits is zero after truncate, causing assertion
1410 failure. This patch extends the transaction or creates a new
1414 Frequency : NFS re-export or patchless client
1415 Bugzilla : 11179, 10796
1416 Description: Crash on NFS re-export node (__d_move)
1417 Details : We do not want to hash the dentry if we don't have a lock.
1418 But if this dentry is later used in d_move, we'd hit uninitialised
1419 list head d_hash, so we just do this to init d_hash field but
1420 leave dentry unhashed.
1423 Frequency : NFS re-export or patchless client
1425 Description: NFS exports has problem with symbolic link
1426 Details : lustre client didn't properly install dentry when re-exported
1427 to NFS or running patchless client.
1430 Frequency : NFS re-export or patchless client
1432 Description: Various nfs/patchless fixes.
1433 Details : fixes reuse disconected alias for lookup process - this fixes
1434 warning "find_exported_dentry: npd != pd",
1435 fix permission error with open files at nfs.
1436 fix apply umask when do revalidate.
1439 Frequency : occasional
1441 Description: Crash on NFS re-export node
1442 Details : calling clear_page() on the wrong pointer triggered oops in
1443 generic_mapping_read().
1446 Frequency : rarely, using O_DIRECT IO
1448 Description: unaligned directio crashes client with LASSERT
1449 Details : check for unaligned buffers before trying any requests.
1452 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1454 Description: stale data returned from RAID cache
1455 Details : If only a small amount of IO is done to the RAID device before
1456 reading it again it is possible to get stale data from the RAID
1457 cache instead of reading it from disk.
1460 Frequency : always for sles10 kernel
1462 Description: sles10 support
1463 Details : ll_follow_link: compile fixes and using of nd_set_link
1464 under newer kernels.
1467 Frequency : depends on arch, kernel and compiler version, always on sles10
1470 Description: recursive or deep enough symlinks cause stack overflow
1471 Details : getting rid of large stack-allocated variable in
1475 Frequency : depends on hardware
1477 Description: lustre write performance loss in the SLES10 kernel
1478 Details : the performance loss is caused by using of write barriers in the
1479 ext3 code. The SLES10 kernel turns barrier support on by
1480 default. The fix is to undo that change for ldiskfs.
1482 ------------------------------------------------------------------------------
1484 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1486 * Support for kernels:
1487 2.6.9-42.0.3EL (RHEL 4)
1488 2.6.5-7.276 (SLES 9)
1489 2.4.21-47.0.1.EL (RHEL 3)
1490 2.6.12.6 vanilla (kernel.org)
1494 Frequency : quota enabled and large files being deleted
1496 Description: releasing more than 4GB of quota at once hangs OST
1497 Details : If a user deletes more than 4GB of files on a single OST it
1498 will cause the OST to spin in an infinite loop. Release
1499 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1504 Description: statfs data retrieved from /proc may be stale or zero
1505 Details : When reading per-device statfs data from /proc, in the
1506 {kbytes,files}_{total,free,avail} files, it may appear
1507 as zero or be out of date.
1510 Frequency : systems with MD RAID1 external journal devices
1512 Description: lconf's call to blkid is confused by RAID1 journal devices
1513 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1514 of returning all block devices that match the journal UUID.
1517 Frequency : always, for aggregate stripe size over 4GB
1519 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1520 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1521 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1524 Frequency : always if "lfs find" used on a local file/directory
1526 Description: "lfs find" segfaults if used on a local file/directory
1527 Details : The case where a directory component was not specified wasn't
1528 handled correctly. Handle this properly.
1531 Frequency : always on ppc64
1533 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1534 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1535 perform endianness conversion.
1538 Frequency : rarely (truncate to non-zero file size after write under load)
1539 Bugzilla : 10730, 10687
1540 Description: Files padded with zeros to next 4K multiple
1541 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1542 it is possible that files that are truncated to a non-zero size
1543 immediately after being written are filled with zero bytes beyond
1544 the truncated size. No file data is lost.
1546 Severity : enhancement
1548 Description: Allow recovery/failover for liblustre clients.
1549 Details : liblustre clients were unaware of failover configurations until
1552 Severity : enhancement
1554 Description: user file locks should fail when not mounting with flock option
1555 Details : Set up an error-returning stub in ll_file_operations.lock field
1556 to prevent incorrect behaviour when client is mounted without
1557 flock option. Also, set up properly f_op->flock field for
1561 Frequency : always on ia64
1563 Description: "lfs df" loops on printing out MDS statfs information
1564 Details : The obd_ioctl_data was not initialized and in some systems
1565 this caused a failure during the ioctl that did not return
1566 an error. Initialize the struct and return an error on failure.
1569 Frequency : SLES 9 only
1571 Description: Error of copying files with lustre special EAs as root
1572 Details : Client side always return success for setxattr call for lustre
1573 special xattr (currently only "trusted.lov").
1576 Frequency : rarely on clusters with both ia64+i386 clients
1578 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1579 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1580 clients are doing concurrent writes to the same file it is
1581 possible that the ia64 clients may LASSERT because the OST
1582 extent locks are not PAGE_SIZE aligned. Ensure that grown
1583 locks are always aligned on the request boundary.
1586 Frequency : specific use, occasional
1588 Description: Overwriting in use executable truncates on-disk binary image
1589 Details : If one node attempts to overwrite an executable in use by
1590 another node, we now correctly return ETXTBSY instead of
1591 truncating the file.
1593 Severity : enhancement
1595 Description: Async OSC create to avoid the blocking unnecessarily.
1596 Details : If a OST has no remain object, system will block on the creating
1597 when need to create a new object on this OST. Now, ways use
1598 pre-created objects when available, instead of blocking on an
1599 empty osc while others are not empty. If we must block, we block
1600 for the shortest possible period of time.
1605 Description: chmod on Lustre root is propagated to other clients
1606 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1607 invalid by the follow_mount time.
1610 Frequency : liblustre clients only
1612 Description: Race in 'instant cancel' lock handling could lead to such locks
1613 never to be granted in case of SMP MDS
1614 Details : Do not destroy not yet granted but cbpending locks in
1618 Frequency : replay/resend of open
1620 Description: non null lock assetion failure in mds_intent_policy
1621 Details : Trying to replay/resend lockless open requests resulted in
1622 mds_open() returning 0 with no lock. Now it sets a flag if
1623 a lock is going to be returned.
1625 Severity : enhancement
1627 Description: Checksum enhancements
1628 Details : New checksum enhancements allow for resending RPCs that failed
1631 Severity : enhancement
1633 Description: Tunables on number of dirty pages in cacche
1634 Details : Allow to set limit on number of dirty pages cached.
1639 Description: client crash on unmount - lock still has references
1640 Details : In some error handling cases it was possible to leak a lock
1641 reference on a client while accessing a file. This was not
1642 harmful to the client during operation, but would cause the
1643 client to crash when the filesystem is unmounted.
1646 Frequency : specific case, rare
1648 Description: ETXTBSY on mds though file not in use by client
1649 Details : ETXTBSY is no longer incorrectly returned when attempting to
1650 chmod or chown a directory that the user previously tried to
1651 execute or a currently-executing binary.
1654 Frequency : extremely rare except on liblustre-based clients
1656 Description: Lustre space not freed when files are deleted
1657 Details : Clean up open-unlinked files after client eviction. Previously
1658 the unlink was skipped and the files remained as orphans.
1663 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1664 Details : In some cases it was possible to send a blocking callback to
1665 a client doing a glimpse, even though that client didn't get
1666 a lock granted. When the glimpse lock is cancelled on the OST
1667 the freed lock is left on the waiting list and corrupted the list.
1670 Frequency : all core dumps
1672 Description: Broke core dumps to lustre
1673 Details : Negative dentry may be unhashed if parent does not have UPDATE
1674 lock, but some callers, e.g. do_coredump, expect dentry to be
1675 hashed after successful create, hash it in ll_create_it.
1677 ------------------------------------------------------------------------------
1679 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1681 * Support for kernels:
1682 2.6.9-42.0.2.EL (RHEL 4)
1683 2.6.5-7.276 (SLES 9)
1684 2.4.21-40.EL (RHEL 3)
1685 2.6.12.6 vanilla (kernel.org)
1689 Frequency : always on RHEL 3
1691 Description: Number of open files grows over time
1692 Details : The number of open files grows over time, whether or not
1693 Lustre is started. This was due to a filp leak introduced
1694 by one of our kernel patches.
1696 ------------------------------------------------------------------------------
1698 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1700 * Support for kernels:
1701 2.6.9-42.EL (RHEL 4)
1702 2.6.5-7.276 (SLES 9)
1703 2.4.21-40.EL (RHEL 3)
1704 2.6.12.6 vanilla (kernel.org)
1709 Bugzilla : 5719, 9635, 9792, 9684
1710 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1711 Details : If a server is under heavy load and cannot reply to new
1712 connection requests before the client resends the (re)connect,
1713 the connection handling code can behave badly if two service
1714 threads are concurrently handing separate (re)connections from
1715 the same client. Add better locking to the connection handling
1716 code, and ensure that only a single connection will be processed
1717 for a given client UUID, even if the lock is dropped.
1719 Severity : enhancement
1721 Description: add TCP zero-copy support to kernel
1722 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1723 sends if the hardware supports scatter-gather and checksumming.
1724 This allows socklnd to do client-write and server-read more
1725 efficiently and reduce CPU utilization from skbuf copying.
1728 Frequency : only if NFS exporting from client
1730 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1731 Details : If setting attributes on a file created under NFS that had
1732 never been opened it would be possible to oops the client
1733 if the file had no objects.
1736 Frequency : always for liblustre
1738 Description: liblustre client does MDS+OSTs setattr RPC for each write
1739 Details : When doing a write from a liblustre client, the client
1740 incorrectly issued an RPC to the MDS and each OST the file was
1741 striped over in order to update the timestamps. When writing
1742 with small chunks and many clients this could overwhelm the MDS
1743 with RPCs. In all cases it would slow down the write because
1744 these RPCs are unnecessary.
1746 Severity : enhancement
1748 Description: allow number of MDS service threads to be changed at module load
1749 Details : It is now possible to change the number of MDS service threads
1750 running. Adding "options mds mds_num_threads={N}" to the MDS's
1751 /etc/modprobe.conf will set the number of threads for the next
1752 time Lustre is restarted (assuming the "mds" module is also
1753 reloaded at that time). The default number of threads will
1754 stay the same, 32 for most systems.
1759 Description: OST crash if filesystem is unformatted or corrupt
1760 Details : If an OST is started on a device that has never been formatted
1761 or if the filesystem is corrupt and cannot even mount then the
1762 error handling cleanup routines would dereference a NULL pointer.
1767 Description: NULL pointer deref in llap_from_page.
1768 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1769 incorrect) mapping if the page was truncated/reclaimed while it was
1770 searched for. Check for this condition and skip such pages when
1771 doing readahead. Introduce extra check to llap_from_page() to
1772 verify page->mapping->host is non-NULL (so page is not anonymous).
1775 Frequency : Sometimes when using sys_sendfile
1777 Description: "page not covered by a lock" warnings from ll_readpage
1778 Details : sendfile called ll_readpage without right page locks present.
1779 Now we introduced ll_file_sendfile that does necessary locking
1780 around call to generic_file_sendfile() much like we do in
1784 Frequency : with certain MDS communication failures at client mount time
1786 Description: NULL pointer deref after failed client mount
1787 Details : a client connection request may delayed by the network layer
1788 and not be sent until after the PTLRPC layer has timed out the
1789 request. If the client fails the mount immediately it will try
1790 to clean up before the network times out the request. Add a
1791 reference from the request import to the obd device and delay
1792 the cleanup until the network drops the request.
1795 Frequency : occasionally during client (re)connect
1797 Description: assertion failure during client (re)connect
1798 Details : processing a client connection request may be delayed by the
1799 client or server longer than the client connect timeout. This
1800 causes the client to resend the connection request. If the
1801 original connection request is replied in this interval, the
1802 client may trip an assertion failure in ptlrpc_connect_interpret()
1803 which thought it would be the only running connect process.
1806 Frequency : only with obd_echo servers and clients that are rebooted
1808 Description: kernel BUG accessing uninitialized data structure
1809 Details : When running an obd_echo server it did not start the ping_evictor
1810 thread, and when a client was evicted an uninitialized data
1811 structure was accessed. Start the ping_evictor in the RPC
1812 service startup instead of the OBD startup.
1814 Severity : enhancement
1815 Bugzilla : 10193 (patchless)
1816 Description: Remove dependency on various unexported kernel interfaces.
1817 Details : No longer need reparent_to_init, exit_mm, exit_files,
1818 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1821 Frequency : rare (only users of deprecated and unsupported LDAP config)
1823 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1824 Details : LDAP apparently contains 'lustreName' attributes instead of
1825 'name'. A simple remapping of the name is sufficient.
1828 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1830 Description: waiting_locks_callback trips kernel BUG if client is evicted
1831 Details : Running with the dump_on_timeout debug flag turned on makes
1832 it possible that the waiting_locks_callback() can try to dump
1833 the Lustre kernel debug logs from an interrupt handler. Defer
1834 this log dumping to the expired_lock_main() thread.
1836 Severity : enhancement
1838 Description: Support NFS exporting on 2.6 kernels.
1839 Details : Implement non-rawops metadata methods for NFS server to use without
1840 changing NFS server code.
1843 Frequency : very rare (synthetic metadata workload only)
1845 Description: two racing renames might cause an MDS thread to deadlock
1846 Details : Running the "racer" program may cause one MDS thread to rename
1847 a file from being the source of a rename to being the target of
1848 a rename at exactly the same time that another thread is doing
1849 so, and the second thread has already enqueued these locks after
1850 doing a lookup of the target and is trying to relock them in
1851 order. Ensure that we don't try to re-lock the same resource.
1854 Frequency : only very large systems with liblustre clients
1856 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1857 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1858 all OSTs in parallel. This allows the eviction of stale liblustre
1859 clients to proceed much faster than if they were done in series,
1860 and also offers similar improvements for other set_info RPCs.
1865 Description: excessive CPU usage during initial read phase on client
1866 Details : During the initial read phase on a client, it would agressively
1867 retry readahead on the file, consuming too much CPU and impacting
1868 performance (since 1.4.5.8). Improve the readahead algorithm
1869 to avoid this, and also improve some other common cases (read
1870 of small files in particular, where "small" is files smaller than
1871 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1876 Description: MDS crash when receiving packet with unknown intent.
1877 Details : Do not LBUG in unknown intent case, just return -EFAULT
1879 Severity : enhancement
1880 Bugzilla : 9293, 9385
1881 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1882 Details : Do not serialize getattr (non-intent version) and statfs.
1885 Frequency : occasional, when OST network is overloaded/intermittent
1887 Description: client evicted by OST after bulk IO timeout
1888 Details : If a client sends a bulk IO request (read or write) the OST
1889 may evict the client if it is unresposive to its data GET/PUT
1890 request. This is incorrect if the network is overloaded (takes
1891 too long to transfer the RPC data) or dropped the OST GET/PUT
1892 request. There is no need to evict the client at all, since
1893 the pinger and/or lock callbacks will handle this, and the
1894 client can restart the bulk request.
1897 Frequency : Always when mmapping file with no objects
1899 Description: client crashes when mmapping file with no objects
1900 Details : Check that we actually have objects in a file before doing any
1901 operations on objects in ll_vm_open, ll_vm_close and
1907 Description: Request leak when working with deleted CWD
1908 Details : Introduce advanced request refcount tracking for requests
1909 referenced from lustre intent.
1911 Severity : Enhancement
1913 Description: Cache open file handles on client.
1914 Details : MDS now will return special lock along with openhandle, if
1915 requested and client is allowed to hold openhandle, even if unused,
1916 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1917 closing files for every read/write openration.
1919 Severity : Enhancement
1921 Description: Cache open negative dentries on client when possible.
1922 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1923 negative dentries on lock revocation.
1928 Description: Remounting a client read-only wasn't possible with a zconf mount
1929 Details : It wasn't possible to remount a client read-only with llmount.
1931 Severity : enhancement
1932 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1933 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1934 MPICH is included to add Lustre-specific ADIO interfaces.
1935 This is based closely on the UFS ADIO layer and only differs
1936 in file creation, in order to allow the OST striping to be set.
1937 This is user-contributed code and not supported by CFS.
1942 Description: extended inode attributes (immutable, append-only) work improperly
1943 when 2.4 and 2.6 kernels are used on client/server or vice versa
1944 Details : Introduce kernel-independent values for these flags.
1946 Severity : enhancement
1949 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1950 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1951 granularity. Now, Fractional MB granularity is be supported,
1952 this is very useful for low memory system.
1954 Severity : enhancement
1956 Description: Getattr by fid
1957 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1958 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1959 have locks and avoids getattr rpc requests for it on every stat.
1962 Frequency : Always, for filesystems larger than 2TB
1964 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1965 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1966 filesystems larger than 2TB to be mounted with mballoc enabled.
1969 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1971 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1972 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1973 filesystems larger than 2TB is mounted then the kernel will
1974 silently corrupt the start of the filesystem. CONFIG_LBD is
1975 enabled for all CFS-supported kernels, but the possibility of
1976 this happening with a modified kernel config exists.
1978 Severity : enhancement
1980 Description: add client O_DIRECT support for 2.6 kernels
1981 Details : It is now possible to do O_DIRECT reads and writes to files
1982 in the Lustre client mountpoint on 2.6 kernel clients.
1984 Severity : enhancement
1986 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1987 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1988 parallel, not waiting for response from every OST before sending
1989 a rpc to the next OST.
1994 Description: setattr vs write race when updating file timestamps
1995 Details : Client processes that update a file timestamp into the past
1996 right after writing to the file (e.g. tar) it is possible that
1997 the updated file modification time can be reset to the current
1998 time due to a race between processing the setattr and write RPC.
2000 Severity : enhancement
2002 Description: Bring 'lfs find' closer in line with regular Linux find.
2003 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2004 -print0 options and obtains all the needed info through the lustre
2007 Severity : enhancement
2009 Description: support up to 1024 configured devices on one node
2010 Details : change obd_dev array from statically allocated to dynamically
2011 allocated structs as they are first used to reduce memory usage
2016 Description: Flush dirty partially truncated pages during truncate
2017 Details : Immediatelly flush partially truncated pages in filter_setattr,
2018 this way we completely avoid having any pages in page cache on OST
2019 and can retire ugly workarounds during writes to flush such pages.
2024 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2025 Details : protect inode from truncation within vfs_unlink() context
2026 just take a reference before calling vfs_unlink() and release it
2027 when parent's i_sem is free.
2032 Description: last_id value checked outside lock on OST caused LASSERT failure
2033 Details : If there were multiple MDS->OST object precreate requests in
2034 flight, it was possible that the OST's last object id was checked
2035 outside a lock and incorrectly tripped an assertion. Move checks
2036 inside locks, and discard old precreate requests.
2039 Frequency : always, if extents are used on OSTs
2041 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2042 Details : OSTs using the extents format would not zero the high 16 bits of
2043 the index physical block number. This is not a problem for any
2044 OST filesystems smaller than 16TB, and no kernels support ext3
2045 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2046 new/modified files) and can be fixed for existing filesystems
2047 with e2fsprogs-1.39-cfs1.
2052 Description: import connection selection may be incorrect if timer wraps
2053 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2054 import connections to be ignored if the 32-bit jiffies counter
2055 wraps. Use a 64-bit jiffies counter.
2058 Frequency : very large clusters immediately after boot
2060 Description: LNET request buffers exhausted under heavy short-term load
2061 Details : If a large number of client requests are generated on a service
2062 that has previously never seen so many requests it is possible
2063 that the request buffer growth cannot keep up with the spike in
2064 demand. Instead of dropping incoming requests, they are held in
2065 the LND until the RPC service can accept more requests.
2068 Frequency : Sometimes during replay
2070 Description: Assertion failure in ll_local_open after replay.
2071 Details : If replay happened on an open request reply before we were able
2072 to set replay handler, reply will become not swabbed tripping the
2073 assertion in ll_local_open. Now we set the handler right after
2074 recognising of open request
2077 Frequency : very rare
2079 Description: kernel reports "badness in vsnprintf"
2080 Details : Reading from the "recovery_status" /proc file in small chunks
2081 may cause a negative length in lprocfs_obd_rd_recovery_status()
2082 call to vsnprintf() (which is otherwise harmless). Exit early
2083 if there is no more space in the output buffer.
2085 Severity : enhancement
2087 Description: clear OBD RPC statistics by writing to them
2088 Details : It is now possible to clear the OBD RPC statistics by writing
2089 to the "stats" file.
2094 Description: Client mtime is not the same on different clients after utimes
2095 Details : In some cases, the client was using the utimes() syscall on
2096 a file cached on another node. The clients now validate the
2097 ctime from the MDS + OSTs to determine which one is right.
2102 Description: Inability to activate failout mode
2103 Details : lconf script incorrectly assumed that in pythong string's numeric
2104 value is used in comparisons.
2107 Frequency : always with multiple stripes per file
2109 Description: Inefficient object allocation for mutli-stripe files
2110 Details : When selecting which OSTs to stripe files over, for files with
2111 a stripe count that divides evenly into the number of OSTs,
2112 the MDS is always picking the same starting OST for each file.
2113 Return the OST selection heuristic to the original design.
2118 Description: mount failures may take full timeout to return an error
2119 Details : Under some heavy load conditions it is possible that a
2120 failed mount can wait for the full obd_timeout interval,
2121 possibly several minutes, before reporting an error.
2122 Instead return an error as soon as the status is known.
2124 Frequency : quota enabled and large files being deleted
2126 Description: releasing more than 4GB of quota at once hangs OST
2127 Details : If a user deletes more than 4GB of files on a single OST it
2128 will cause the OST to spin in an infinite loop. Release
2129 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2134 Description: statfs data retrieved from /proc may be stale or zero
2135 Details : When reading per-device statfs data from /proc, in the
2136 {kbytes,files}_{total,free,avail} files, it may appear
2137 as zero or be out of date.
2140 Frequency : systems with MD RAID1 external journal devices
2142 Description: lconf's call to blkid is confused by RAID1 journal devices
2143 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2144 of returning all block devices that match the journal UUID.
2147 Frequency : always, for aggregate stripe size over 4GB
2149 Description: assertion fails when trying to use 4GB stripe size
2150 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2151 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2154 Frequency : always on ppc64
2156 Description: the first write on an ext3 filesystem with mballoc got stuck
2157 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2158 perform endianness conversion.
2160 ------------------------------------------------------------------------------
2162 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2164 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2165 INTEROPERATE with older versions automatically. Please read the
2166 user documentation before upgrading any part of a live system.
2167 * WARNING: Lustre networking configuration changes are required with
2168 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2171 * Support for kernels:
2172 2.6.9-22.0.2.EL (RHEL 4)
2173 2.6.5-7.244 (SLES 9)
2174 2.6.12.6 vanilla (kernel.org)
2177 Severity : enhancement
2178 Bugzilla : 7981/8208
2179 Description: Introduced Lustre Networking (LNET)
2180 Details : LNET is new networking infrastructure for Lustre, it includes
2181 a reorganized network configuration mode (see the user
2182 documentation for full details) as well as support for routing
2183 between different network fabrics. Lustre Networking Devices
2184 (LNDs) for the supported network fabrics have also been
2185 created for this new infrastructure.
2187 Severity : enhancement
2188 Description: Introduced Access control lists
2189 Details : clients can set ACLs on files and directories in order to have
2190 more fine-grained permissions than the standard Unix UGO+RWX.
2191 The MDS must be started with the "-o acl" mount option.
2193 Severity : enhancement
2194 Description: Introduced filesystem quotas
2195 Details : Administrators may now establish per-user quotas on the
2198 Severity : enhancement
2200 Description: Configuration change for the XT3
2201 The PTLLND is now used to run Lustre over Portals on the XT3
2202 The configure option(s) --with-cray-portals are no longer used.
2203 Rather --with-portals=<path-to-portals-includes> is used to
2204 enable building on the XT3. In addition to enable XT3 specific
2205 features the option --enable-cray-xt3 must be used.
2210 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2211 Details : A race between allocating a new llcd and re-getting the llcd_lock
2212 allowed another thread to grab newly-allocated llcd.
2214 Severity : enhancement
2216 Description: 2.6 OST async journal commit and locking fix to improve performance
2217 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2218 2.6 kernels are now async as they already were in 2.4 kernels so
2219 that they can commit concurrently with the network bulk transfer.
2220 For block-allocated files the filter allocation semaphore is held
2221 to avoid filesystem fragmentation during allocation. BKL lock
2222 removed for 2.6 xattr operations where it is no longer needed.
2227 Description: lconf incorrectly determined whether two IP networks could talk
2228 Details : In some more complicated routing and multiple-network
2229 configurations, lconf will avoid trying to make a network
2230 connection to a disjoint part of the IP space. It was doing the
2231 math incorrectly for one set of cases.
2236 Description: Fix for potential infinite loop processing records in an llog.
2237 Details : If an llog record is corrupted/zeroed, it is possible to loop
2238 forever in llog_process(). Validate the llog record length
2239 and skip the remainder of the block on error.
2242 Frequency : occasional (liblustre only)
2244 Description: liblustre could not open files whose last component is a symlink
2245 Details : sysio_path_walk() would incorrectly pass the open intent to
2246 intermediate path components.
2249 Frequency : rare (liblustre only with non-standard tuning)
2250 Bugzilla : 7201 (7350)
2251 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2252 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2253 after it has been released, only do this for referenced locks
2255 Severity : enhancement
2257 Description: specifying an (invalid) directory default stripe_size of -1
2258 would reset the directory default striping
2259 Details : stripe_size -1 was used internally to signal directory stripe
2260 removal, now use "all default" to signal dir stripe removal
2261 as a directory striping of "all default" is not useful
2264 Frequency : common for large clusters running liblustre clients
2266 Description: doing an ls when liblustre clients are running is slow
2267 Details : sending a glimpse AST to a liblustre client waits for every AST
2268 to time out, as liblustre clients will not respond. Since they
2269 cannot cache data we refresh the OST lock LVB from disk instead.
2271 Severity : enhancement
2273 Description: doing an ls at the same time as file IO can be slow
2274 Details : enqueue and other "small" requests can be blocked behind many
2275 large IO requests. Create a new OST IO portal for non-IO
2276 requests so they can be processed faster.
2279 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2281 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2282 created file with mode 000
2283 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2284 allow an ftruncate() as a non-root user to a file with mode 000.
2285 Lustre now respects this flag to disable mode checks when
2286 truncating a file owned by the user
2289 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2292 Description: Revoking locks from clients that went dead or catatonic might take
2294 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2295 cancellation of such locks instant on servers without waiting for
2296 any reply from clients. Clients drops these locks when cancel
2297 notification from server is received without replying.
2300 Frequency : liblustre-only, when liblustre client dies or becomes busy
2302 Description: Doing ls on Linux clients can take a long time with active
2304 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2305 granting such locks to it in the first place if possible. Locks
2306 are taken by proxy on the OST during the read or write and
2307 dropped immediately afterward. Add connect flags handling, do
2308 not grant locks to liblustre clients for glimpse ASTs.
2310 Severity : enhancement
2312 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2313 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2314 sequential reads and behaves badly with applications doing random
2315 reads. Improve it by reading ahead at least read region, and
2316 avoiding excessive large RPC for small reads.
2318 Severity : enhancement
2320 Description: Creating more than 1000 files for a single job may cause a load
2321 imbalance on the OSTs if there are also a large number of OSTs.
2322 Details : qos_prep_create() uses an OST index reseed value that is an
2323 even multiple of the number of available OSTs so that if the
2324 reseed happens in the middle of the object allocation it will
2325 still utilize the OSTs as uniformly as possible.
2330 Description: OST or MDS may oops in ping_evictor_main()
2331 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2332 but doesn't restart at beginning of obd_exports_timed list
2335 Severity : enhancement
2337 Description: improve by-nid export eviction on the MDS and OST
2338 Details : allow multiple exports with the same NID to be evicted at one
2339 time without re-searching the exports list.
2342 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2344 Description: MDS may oops in groups_free()
2345 Details : in rare race conditions a newly allocated group_info struct is
2346 freed again, and this can be NULL. The 2.4 compatibility code
2347 for groups_free() checked for a NULL pointer, but 2.6 did not.
2350 Frequency : common for liblustre clients doing little filesystem IO
2351 Bugzilla : 9352, 7313
2352 Description: server may evict liblustre clients accessing contended locks
2353 Details : if a client is granted a lock or receives a completion AST
2354 with a blocking AST already set it would not reply to the AST
2355 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2358 Frequency : lfs setstripe, only systems with more than 160 OSTs
2360 Description: unable to set striping with a starting offset beyond OST 160
2361 Details : llapi_create_file() incorrectly limited the starting stripe
2362 index to the maximum single-file stripe count.
2365 Frequency : LDAP users only
2367 Description: lconf did not handle in-kernel recovery with LDAP properly
2368 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2370 Severity : enhancement
2372 Description: bind OST threads to NUMA nodes to improve performance
2373 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2374 node and do their allocations there to localize memory access
2376 Severity : enhancement
2378 Description: llmount can determine client NID directly from Myrinet (GM)
2379 Details : the client NID code from gmnalnid was moved directly into
2380 llmount, removing the need to use this or specifying the
2381 client NID explicitly when mounting GM clients with zeroconf
2384 Frequency : if client is started with down MDS
2386 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2387 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2388 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2389 ptlrpc_import_delayed_req() because we didn't check if the
2390 request was interrupted, and we also didn't break out of the
2391 event loop if there was no timeout
2396 Description: data loss during non-page-aligned writes to a single file from
2397 both multiple nodes and multiple threads on one node at same time
2398 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2399 inconsistency led to false short-reads, that were cached and later
2400 used by ->prepare_write() to fill in partially written page,
2401 leading to data loss.
2404 Frequency : always, if lconf --abort_recovery used
2406 Description: lconf --abort_recovery fails with 'Operation not supported'
2407 Details : lconf was attempting to abort recovery on the MDT device and not
2410 Severity : enhancement
2412 Description: remove cleanup logs
2413 Details : replace lconf-generated cleanup logs with lustre internal
2414 cleanup routines. Eliminates the need for client-cleanup and
2417 Severity : enhancement
2419 Description: add support for EAs (user and system) on lustre filesystems
2420 Details : it is now possible to store extended attributes in the Lustre
2421 client filesystem, and with the user_xattr mount option it
2422 is possible to allow users to store EAs on their files also
2424 Severity : enhancement
2426 Description: Add possibility (config option) to show minimal available OST free
2428 Details : When compiled with --enable-mindf configure option, statfs(2)
2429 (and so, df) will return least minimal free space available from
2430 all OSTs as amount of free space on FS, instead of summary of
2431 free spaces of all OSTs.
2433 Severity : enhancement
2435 Description: do not expand extent locks acquired on OST-side
2436 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2437 by server: they are not cached anyway.
2440 Frequency : when mmap is used/binaries executed from Lustre
2442 Description: Unmmap pages before throwing them away from read cache.
2443 Details : llap_shrink cache now attempts to unmap pages before discarding
2444 them (if unmapping failed - do not discard). SLES9 kernel has
2445 extra checks that trigger if this unmapping is not done first.
2450 Description: lconf didn't resolve symlinks before checking to see whether a
2451 given mountpoint was already in use
2454 Frequency : when migrating failover services
2455 Bugzilla : 6395, 9514
2456 Description: When migrating a subset of services from a node (e.g. failback
2457 from a failover service node) the remaining services would
2458 time out and evict clients.
2459 Details : lconf --force (implied by --failover) sets the global obd_timeout
2460 to 5 seconds in order to quickly disconnect, but this caused
2461 other RPCs to time out too quickly. Do not change the global
2462 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2464 Severity : enhancement
2465 Frequency : if MDS is started with down OST
2466 Bugzilla : 9439,5706
2467 Description: Allow startup/shutdown of an MDS without depending on the
2468 availability of the OSTs.
2469 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2470 Add appropriate locking and lov-osc refcounts for safe
2471 cleaning. Add osc abort_inflight calls in case the
2472 synchronize never started.
2475 Frequency : occasional (Cray XT3 only)
2477 Description: root not authorized to access files in CRAY_PORTALS environment
2478 Details : The client process capabilities were not honoured on the MDS in
2479 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2480 been accessed by an authorized user then root was able to access
2481 the file on the local client also. The root user capabilities
2482 are now allowed on the MDS, as this environment has secure UID.
2485 Frequency : occasional
2487 Description: ldiskfs "too long searching" message happens too often
2488 Details : A debugging message (otherwise harmless) prints too often on
2489 the OST console. This has been reduced to only happen when
2490 there are fragmentation problems on the filesystem.
2495 Description: Division by zero in statfs when all OSCs are inactive
2496 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2497 lov_statfs divides by value returned by lov_get_stripecnt().
2501 Bugzilla : 9489, 3273
2502 Description: First write from each client to each OST was only 4kB in size,
2503 to initialize client writeback cache, which caused sub-optimal
2504 RPCs and poor layout on disk for the first writen file.
2505 Details : Clients now request an initial cache grant at (re)connect time
2506 and so that they can start streaming writes to the cache right
2507 away and always do full-sized RPCs if there is enough data.
2508 If the OST is rebooted the client also re-establishes its grant
2509 so that client cached writes will be honoured under the grant.
2514 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2515 Details : Now I/O RPCs go to different portal number and (presumably) fast
2516 lock requests (and glimses) and other RPCs get their own service
2517 threads pool that should be able to service those RPCs
2520 Severity : enhancement
2522 Description: Ability to exchange lustre version between client and servers and
2523 issue warnings at client side if client is too old. Also for
2524 liblustre clients there is ability to refuse connection of too old
2526 Details : New 'version' field is added to connect data structure that is
2527 filled with version info. That info is later checked by server and
2531 Frequency : rare, liblustre only.
2532 Bugzilla : 9296, 9581
2533 Description: Two simultaneous writes from liblustre at offset within same page
2534 might proceed at the same time overwriting eachother with stale
2536 Details : I/O lock withing llu_file_prwv was released too early, before data
2537 actually was hitting the wire. Extended lock-holding time until
2538 server acknowledges receiving data.
2541 Frequency : extremely rare. Never observed in practice.
2543 Description: avoid generating lustre_handle cookie of 0.
2544 Details : class_handle_hash() generates handle cookies by incrementing
2545 global counter, and can hit 0 occasionaly (this is unlikely, but
2546 not impossible, because initial value of cookie counter is
2547 selected randonly). Value of 0 is used as a sentinel meaning
2548 "unassigned handle" --- avoid it. Also coalesce two critical
2549 sections in this function into one.
2551 Severity : enhancement
2553 Description: allow liblustre clients to delegate truncate locking to OST
2554 Details : To avoid overhead of locking, liblustre client instructs OST to
2555 take extent lock in ost_punch() on client's behalf. New connection
2556 flag is added to handle backward compatibility.
2558 Severity : enhancement
2559 Bugzilla : 4928, 7341, 9758
2560 Description: allow number of OST service threads to be specified
2561 Details : a module parameter allows the number of OST service threads
2562 to be specified via "options ost ost_num_threads={N}" in the
2563 OSS's /etc/modules.conf or /etc/modprobe.conf.
2567 Bugzilla : 6146, 9635, 9895
2568 Description: servers crash with bad pointer in target_handle_connect()
2569 Details : In rare cases when a client is reconnecting it was possible that
2570 the connection request was the last reference for that export.
2571 We would temporarily drop the export reference and get a new
2572 one, but this may have been the last reference and the export
2573 was just destroyed. Get new reference before dropping old one.
2575 Severity : enhancement
2576 Frequency : if client is started with failover MDS
2578 Description: Allow multiple MDS hostnames in the mount command
2579 Details : Try to read the configuration from all specified MDS
2580 hostnames during a client mount in case the "primary"
2583 Severity : enhancement
2585 Description: Stop sending data to evicted clients as soon as possible.
2586 Details : Check if the client we are about to send or are sending data to
2587 was evicted already. (Check is done every second of waiting,
2588 for which l_wait_event interface was extended to allow checking
2589 of exit condition at specified intervals).
2592 Frequency : rare, normally only when NFS exporting is done from client
2594 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2596 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2599 Frequency : always, if srand() is called before liblustre initialization
2601 Description: Liblustre uses system PRNG disturbing its usage by user application
2602 Details : Introduce internal to lustre fast and high-quality PRNG for
2603 lustre usage and make liblustre and some other places in generic
2604 lustre code to use it.
2606 Severity : enhancement
2607 Bugzilla : 9477, 9557, 9870
2608 Description: Verify that the MDS configuration logs are updated when xml is
2609 Details : Check if the .xml configuration logs are newer than the config
2610 logs stored on the MDS and report an error if this is the case.
2611 Request --write-conf, or allow starting with --old_conf.
2613 Severity : enhancement
2615 Description: Handle symlinks in the path when checking if Lustre is mounted.
2616 Details : Resolve intermediate symlinks when checking if a client has
2617 mounted a filesystem to avoid duplicate client mounts.
2622 Description: lconf can hit an error exception but still return success.
2623 Details : The lconf command catches the Command error exception at the top
2624 level script context and will exit with the associated exit
2625 status, but doesn't ensure that this exit status is non-zero.
2630 Description: failure of ptlrpc thread startup can cause oops
2631 Details : Starting a ptlrpc service thread can fail if there are a large
2632 number of threads or the server memory is very fragmented.
2633 Handle this without oopsing.
2636 Frequency : always, only if liblustre and non-default acceptor port was used
2638 Description: liblustre cannot connect to servers with non-default acceptor port
2639 Details : tcpnal_set_default_params() was not called and was therefore
2640 ignoring the environment varaible TCPNAL_PORT, as well as other
2641 TCPNAL_ environment variables
2646 Description: two objects could be created on the same OST for a single file
2647 Details : If an OST is down, in some cases it was possible to create two
2648 objects on a single OST for a single file. No problems other
2649 than potential performance impact and spurious error messages.
2653 Bugzilla : 5681, 9562
2654 Description: Client may oops in ll_unhash_aliases
2655 Details : Client dcache may become inconsistent in race condition.
2656 In some cases "getcwd" can fail if the current directory is
2662 Description: Inode refcounting problems in NFS export code
2663 Details : link_raw functions used to call d_instantiate without obtaining
2664 extra inode reference first.
2668 Bugzilla : 9942, 9903
2669 Description: Referencing freed requests leading to crash, memleaks with NFS.
2670 Details : We used to require that call to ll_revalidate_it was always
2671 followed by ll_lookup_it. Also with revalidate_special() it is
2672 possible to call ll_revalidate_it() twice for the same dentry
2673 even if first occurence returned success. This fix changes semantic
2674 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2675 reference on a request referred from the intent.
2676 ll_intent_release() then releases such a request.
2679 Frequency : rare, normally benchmark loads only
2681 Description: unlinked inodes were kept in memory on the client
2682 Details : If a client is repeatedly creating and unlinking files it
2683 can accumulate a lot of stale inodes in the inode slab cache.
2684 If there is no other client load running this can cause the
2685 client node to run out of memory. Instead flush old inodes
2686 from client cache that have the same inode number as a new inode.
2689 Frequency : SLES9 2.6.5 kernel and long filenames only
2690 Bugzilla : 9969, 10379
2691 Description: utime reports stale NFS file handle
2692 Details : SLES9 uses out-of-dentry names in some cases, which confused
2693 the lustre dentry revalidation. Change it to always use the
2697 Frequency : rare, unless heavy write-truncate concurrency is continuous
2698 Bugzilla : 4180, 6984, 7171, 9963, 9331
2699 Description: OST becomes very slow and/or deadlocked during object unlink
2700 Details : filter_destroy() was holding onto the parent directory lock
2701 while truncating+unlinking objects. For very large objects this
2702 may block other threads for a long time and slow overall OST
2703 responsiveness. It may also be possible to get a lock ordering
2704 deadlock in this case, or run out of journal credits because of
2705 the combined truncate+unlink. Solution is to do object truncate
2706 first in one transaction without parent lock, and then do the
2707 final unlink in a new transaction with the parent lock. This
2708 reduces the lock hold time dramatically.
2711 Frequency : rare, 2.4 kernels only
2713 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2714 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2715 before unmounting their filesystem in order to avoid deadlock.
2716 The kernel_locked() function in 2.4 kernels only checks whether
2717 the kernel lock is held, not whether it is this process that is
2718 holding it as 2.6 kernels do.
2723 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2724 Details : The client ptlrpc code may be trying to reconnect to a down
2725 server before a previous connection attempt has timed out.
2726 Increase the reconnect interval to be longer than the connection
2727 timeout interval to avoid sending duplicate connections to
2731 Frequency : echo_client brw_test command
2733 Description: fix echo_client to work with OST preallocated code
2734 Details : OST preallocation code (5137) didn't take echo_client IO path
2735 into account: echo_client calls filter methods outside of any
2736 OST thread and, hence, there is no per-thread preallocated
2737 pages and buffers to use. Solution: hijack pga pages for IO. As
2738 a byproduct, this avoids unnecessary data copying.
2742 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2743 Description: Client can oops in mdc_commit_close() after open replay
2744 Details : It was possible for the MDS to return an open request with no
2745 transaction number in mds_finish_transno() if the client was
2746 evicted, but without actually returning an error. Clients
2747 would later try to replay that open and may trip an assertion
2748 Simplify the client close codepath, and always return an error
2749 from the MDS in case the open is not successful.
2752 Frequency : rare, 2.6 OSTs only
2754 Description: OST may deadlock under high load on fragmented files
2755 Details : If there was a heavy load and highly-fragmented OST filesystems
2756 it was possible to have all the OST threads deadlock waiting on
2757 allocation of biovecs, because the biovecs were not released
2758 until the entire RPC IO was completed. Instead, release biovecs
2759 as soon as they are complete to ensure forward IO progress.
2761 Severity : enhancement
2763 Description: Support for specifying external journal device at mount
2764 Details : If an OST or MDS device is formatted with an external journal
2765 device, this device major/minor is stored in the ext3 superblock
2766 and may not be valid for failover. Allow detecting and
2767 specifying the external journal at mount time.
2772 Description: Mounting an MDS with pending unlinked files may cause oops
2773 Details : target_finish_recovery() calls mds_postrecov() which returned
2774 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2775 considers this an error and immediately begins cleaning up the
2776 lov, just after starting the mds_lov process
2778 Severity : enhancement
2780 Description: Implement 'lfs df' to report actual free space on per-OST basis
2781 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2782 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2783 usage of objects; '-h' to report in human readable format.
2785 ------------------------------------------------------------------------------
2787 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2794 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2795 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2796 at startup during orphan unlinking. Instead, skip buddy update
2797 before setup, as it will be regenerated after recovery is complete.
2802 Description: If an OST is inactive, its locks might reference stale inodes.
2803 Details : lov_change_cbdata() must iterate over all namespaces, even if
2804 they are inactive to clear inode references from the lock.
2806 Severity : enhancement
2807 Frequency : occasional, if non-standard max_dirty_mb used
2809 Description: Client will block write RPCs if not enough grant
2810 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2811 then the client will block writes while waiting for another RPC
2812 to complete instead of consuming its dirty limit. With change
2813 we get improved performance when max_dirty_mb is small.
2815 Severity : enhancement
2816 Bugzilla : 3389, 6253
2817 Description: Add support for supplementary groups on the MDS.
2818 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2819 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2820 lookups for user supplementary groups into a cache.
2824 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2825 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2826 for liblustre O_EXCL is enforced by MDS.
2830 Description: readdir never returns NULL in liblustre.
2831 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2832 dentry in struct dirent.
2836 Frequency : liblustre only, depends on application IO pattern
2837 Description: liblustre clients evicted if not contacting servers
2838 Details : Don't put liblustre clients into the ping_evictor list, so
2839 they will not be evicted by the pinger ever.
2841 Severity : enhancement
2843 Description: Add ability to evict clients by NID from MDS.
2844 Details : By echoing "nid:$NID" string into
2845 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2846 $NID would be instantly evicted from this MDS and from all active
2847 OSTs connected to it.
2851 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2852 Details : lookup_it_finish() used to query file size from OSTs that was not
2857 Description: service threads change working directory to that of init
2858 Details : Starting lustre service threads may pin the working directory
2859 of the parent thread, making that filesystem busy. Threads
2860 now change to the working directory of init to avoid this.
2864 Frequency : during shutdown only
2865 Description: shutdown with a failed MDS or OST can cause unmount to hang
2866 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2871 Frequency : 2.6 only
2872 Description: chmod/chown may include an extra supplementary group
2873 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2874 supplementary group and if none is specified this is used.
2877 Bugzilla : 5479 (6816)
2879 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2880 Details : If lookup is in progress on a file that is unlinked we might try
2881 to revalidate the inode and fail in revalidate after lookup is
2882 complete and ll_file_open() enqueues the open again but
2883 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2886 Frequency : always, if lconf --abort_recovery used
2888 Description: lconf --abort_recovery fails with 'Operation not supported'
2889 Details : lconf was attempting to abort recovery on the MDT device and not
2892 ------------------------------------------------------------------------------
2894 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2899 Frequency : rare (only unsupported configurations with a node running as an
2901 Bugzilla : 6514, 5137
2902 Description: Mounting a Lustre file system on a node running as an OST could
2904 Details : OSTs now preallocates memory needed to write out data at
2905 startup, instead of when needed, to avoid having to
2906 allocate memory in possibly low memory situations.
2907 Specifically, if the file system is mounted on on OST,
2908 memory pressure could force it to try to write out data,
2909 which it needed to allocate memory to do. Due to the low
2910 memory, it would be unable to do so and the node would
2911 become unresponsive.
2913 Severity : enhancement
2915 Description: Addition of lconf --service command line option
2916 Details : lconf now accepts a '--service <arg>' option, which is
2917 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2919 Severity : enhancement
2921 Description: Failover mode is now the default for OSTs.
2922 Details : By default, OSTs will now run in failover mode. To return to
2923 the old behaviour, add '--failout' to the lmc line for OSTs.
2925 Severity : enhancement
2927 Description: Health checks are now provided for MDS and OSTs
2928 Details : Additional detailed health check information on MSD and OSTs
2929 is now provided through the procfs health_check value.
2932 Frequency : occasional, depends on IO load
2934 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2935 after numerous create/delete cycles
2936 Details : The ext3 inode allocation policy would not allocate new inodes
2937 very well on the OSTs because there are no new directories
2938 being created. Instead we look for groups with free space if
2939 the parent directories are nearly full.
2944 Description: Network or server problems during mount may cause partially
2945 mounted clients instead of returning an error.
2946 Details : The config llog parsing code may overwrite the error return
2947 code during mount error handling, returning success instead
2953 Description: MDS can fail to allocate large reply buffers
2954 Details : After long uptimes the MDS can fail to allocate large reply
2955 buffers (e.g. zconf client mount config records) due to memory
2956 fragmentation or consumption by the buffer cache. Preallocate
2957 some large reply buffers so that these replies can be sent even
2958 under memory pressure.
2962 Frequency : rare (liblustre)
2963 Description: fsx running with liblustre complained that using truncate() to
2964 extend the file doesn't work. This patch corrects that issue.
2965 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2966 ATTR_SIZE and lsm use in llu_setattr_raw.
2970 Frequency : rare, only 2.6 kernels
2971 Description: Unusual file access patterns on the MDS may result in inode
2972 data being lost in very rare circumstances.
2973 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2974 code in the 2.6 kernel caused the inode and/or EA data not to
2975 be read from disk, causing single-file corruption.
2979 Frequency : rare, only 2.6 filesystems using extents
2980 Description: Heavy concurrent write and delete load may cause data corruption.
2981 Details : It was possible under high-load situations to have an extent
2982 metadata block in the block device cache from a just-unlinked
2983 file overwrite a newly-allocated data block. We now unmap any
2984 metadata buffers that alias just-allocated data blocks.
2988 Frequency : filesystems with default stripe_count larger than 77
2989 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2990 Details : lconf specifies an inode size of 4096 bytes when the default
2991 stripe_count is larger than 77. This conflicts with the default
2992 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2993 this case to avoid pinning too much memory for large EAs.
2995 ------------------------------------------------------------------------------
2997 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3002 Frequency : rare (extremely heavy IO load with hundreds of clients)
3004 Description: Client is evicted, gets IO error writing to file
3005 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3006 caused clients to be evicted for AST timeouts. The fixes in
3007 bug 5192 mean we no longer need to have such short AST timeouts
3008 so ldlm_timeout has been increased.
3011 Frequency : occasional during --force or --failover shutdown under load
3012 Bugzilla : 5949, 4834
3013 Description: Server oops/LBUG if stopped with --force or --failover under load
3014 Details : a collection of import/export refcount and cleanup ordering
3015 issues fixed for safer force cleanup
3018 Frequency : only filesystems larger than 120 OSTs
3019 Bugzilla : 5990, 6223
3020 Description: lfs getstripe would oops on a very large filesystem
3021 Details : lov_getconfig used kfree on vmalloc'd memory
3024 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3025 Bugzilla : 6242, 6243
3026 Description: reading from files that had been truncated to a non-zero size
3027 but never opened returned no data
3028 Details : ll_file_read() reads zeros from no-object files to EOF
3033 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3034 Details : There's little harm in aborting MDS/OSS recovery and letting it
3035 try again, so I removed the LASSERT and return an error instead.
3037 Severity : enhancement
3039 Description: New debugging infrastructure for tracking down data corruption
3040 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3041 (b) cover more of the client-side code path, and (c) try to narrow
3042 down where problems occurred
3046 Bugzilla : 3819, 4364, 4397, 6313
3047 Description: Racing close and eviction MDS could cause assertion in mds_close
3048 Details : It was possible to get multiple mfd references during close and
3049 client eviction, leading to one thread referencing a freed mfd.
3051 Severity: : enhancement
3052 Bugzilla : 3262, 6359
3053 Description: Attempts to reconnect to servers are now more aggressive.
3054 Details : This builds on the enhanced upcall-less recovery that was added
3055 in 1.4.2. When trying to reconnect to servers, clients will
3056 now try each server in the failover group every 10 seconds. By
3057 default, clients would previously try one server every 25 seconds.
3062 Description: After recovery, certain operations trigger a failed
3063 assertion on a client.
3064 Details : Failing over an mds, using lconf -d --failover, while a
3065 client was doing a readdir() call would cause the client to
3066 LBUG after recovery completed and the readdir() was resent.
3068 Severity : enhancement
3070 Description: Default groups are now added by lconf
3071 Details : You can now run lconf --group <servicename> without having to
3072 manually add groups with lmc.
3075 Frequency : occasional
3077 Description: Nodes with an elan id of 0 trigger a failed assertion
3080 Frequency : always when accessing e.g. tty/console device nodes
3082 Description: tty and some other devices nodes cannot be used on lustre
3083 Details : file's private_data field is used by device data and lustre
3084 values in there got lost. New field was added to struct file to
3085 store fs-specific private data.
3088 Frequency : when exporting Lustre via NFS
3090 Description: NFSD failed occasionally when looking up a path component
3091 Details : NFSD is looking up ".." which was broken in ext3 directories
3092 that had grown large enough to become hashed.
3095 Frequency : Clusters with multiple interfaces not on the same subnet
3097 Description: Nodes will repeatedly try to reconnect to an interface which it
3098 cannot reach and report an error to the log.
3099 Details : Extra peer list entries will be created by lconf with some peers
3100 unreachable. lconf now validates the peer before adding it.
3103 Frequency : Only if a default stripe is set on the filesystem root.
3105 Description: Setting a default stripe on the filesystem root prevented the
3106 filesystem from being remounted.
3107 Details : The client was sending extra request flags in the root getattr
3108 request and did not allocate a reply buffer for the dir EA.
3111 Frequency : occasional, higher if lots of files are accessed by one client
3112 Bugzilla : 6159, 6097
3113 Description: Client trips assertion regarding lsm mismatch/magic
3114 Details : While revalidating inodes the VFS looks up inodes with ifind()
3115 and in rare cases can find an inode that is being freed.
3116 The ll_test_inode() code will free the lsm during ifind()
3117 when it finds an existing inode and then the VFS later attaches
3118 this free lsm to a new inode.
3122 Bugzilla : 6422, 7030
3123 Description: MDS deadlock between mkdir and client eviction
3124 Details : Creating a new file via mkdir or mknod (starting a transaction
3125 and getting the ns lock) can deadlock with client eviction
3126 (gets ns lock and trying to finish a synchronous transaction).
3129 Frequency : occasional
3130 Description: While starting a server, the fsfilt_ext3 module could not be
3132 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3133 kernels. Previously, lconf would still use the ext3 name
3134 when trying to load modules. Now, it will correctly use
3135 ext3 on 2.4 and ldiskfs on 2.6.
3137 Severity : enhancement
3138 Description: The default stripe count has been changed to 1
3139 Details : The interpretation of the default stripe count (0, to lfs
3140 or lmc) has been changed to mean striping across a single
3141 OST, rather than all available. For general usage we have
3142 found a stripe count of 1 or 2 works best.
3144 Severity : enhancement
3145 Description: Add support for compiling against Cray portals.
3146 Details : Conditional compiling for some areas that are different
3150 Frequency : occasional
3151 Bugzilla : 6409, 6834
3152 Description: Creating files with an explicit stripe count may lead to
3153 a failed assertion on the MDS
3154 Details : If some OSTs are full or unavailable, creating files may
3155 trigger a failed assertion on the MDS. Now, Lustre will
3156 try to use other servers or return an error to the
3160 Frequency : occasional
3162 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3163 caused lock timeout during readahead (since 1.4.2).
3164 Details : Processes doing readahead might match a lock that hasn't been
3165 granted yet if there are overlapping and conflicting lock
3166 requests. The readahead process waits on ungranted lock
3167 (original lock is CBPENDING), while OST waits for that process
3168 to cancel CBPENDING read lock and eventually evicts client.
3170 Severity : enhancement
3172 Description: Initial enabling of flock support for clients
3173 Details : Implements fcntl advisory locking and file status functions.
3174 This feature is provided as an optional mount flag (default
3175 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3176 locking are implemented yet, and those that are are not guaranteed
3177 to be completely correct in production environments.
3178 mount -t lustre -o [flock|noflock] ...
3181 Frequency : occasional
3183 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3184 assertion in the ext3 JBD (journaling) layer.
3185 Details : The b_committed_data struct is protected by the big kernel lock
3186 in 2.4 kernels, serializing journal_commit_transaction() and
3187 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3188 access to this struct. In 2.6 kernels there is finer grained
3189 locking to improve SMP performance of the JBD layer.
3193 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3198 Description: merge and cleanup kernel patches.
3201 -----------------------------------------------------------------------------
3203 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3205 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3206 versions of Lustre. Please update all servers and clients to
3207 version 1.4.2 or later at the same time. You must also run
3208 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3209 to update the configuration logs.
3211 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3212 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3213 - swab llog records of type '0' so we get proper header size/idx (5861)
3214 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3215 - fix rename of one directory over another leaking an inode (5953)
3216 - avoid SetPageDirty on 2.6 (5981)
3217 - don't re-add just-being-destroyed locks to the waiting list (5653)
3218 - when creating new directories, inherit the parent's custom
3219 striping settings if present parent (3048)
3220 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3221 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3222 - don't hold client locks on temporary worklist from l_lru (5666)
3223 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3224 - automatically evict dead clients (5921)
3225 - Update file size properly in create+truncate+fstat case (6196)
3226 - Do not unhash mountpoint dentries, do not allow removal of
3228 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3229 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3230 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3231 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3232 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3233 - Correctly handle reads of files with no objects (6243)
3234 - lctl recover will also mark a device active if deactivate used (5933)
3236 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3237 - allow --write-conf on an MDS with different nettype than client (5619)
3238 - don't write config llogs to MDS for mounts not from that MDS (5617)
3239 - lconf should create multiple TCP connections from a client (5201)
3240 - init scripts are now turned off by default; run chkconfig --on
3241 lustre and chkconfig --on lustrefs to use them
3242 - upcalls are no longer needed for clients to recover to failover
3244 - add --abort-recovery option to lconf to abort recovery on device
3246 - add support for an arbitrary number of OSTs (3026)
3247 - Quota support protocol changes.
3248 - forward compatibility changes to wire structs (6007)
3249 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3250 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3251 - improved "lustre status" script
3252 - initialize blocksize for non-regular files (6062)
3253 - added --disable-server and --disable-client configure options (5782)
3254 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3255 - Vanilla 2.4.29 support
3256 - increase maximum number of obd devices to 520 (6242)
3257 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3258 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3259 SLES 9 2.6.5 kernels
3260 - limit stripes per file to 160 (the maximum EA size) (6093)
3262 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3265 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3266 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3267 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3268 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3269 - small ext3 extents cleanups and fixes (5733)
3270 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3271 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3272 - avoid needless client->OST connect, fix handle mismatch (5317)
3273 - fix DLM error path that led to out-of-sync client, long delays (5779)
3274 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3275 - fix several locking issues related to i_size (5492,5624,5654,5672)
3276 - don't move pending lock onto export if it is already evicted (5683)
3277 - fix kernel oops when creating .foo in unlinked directory (5548)
3278 - fix deadlock in obdfilter statistics vs. object create (5811)
3279 - use time_{before,after} to avoid timer jiffies wrap (5882)
3280 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3281 - Do not leak request if server was not able to process it (5154)
3282 - If mds_open unable to find parent dir, make that negative lookup(5154)
3283 - don't create new directories with extent-mapping (5909, 5936)
3285 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3286 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3287 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3288 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3289 - support for RHEL4 kernel on i686 (5773)
3290 - provide error messages when incompatible logs are encountered (5898)
3292 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3293 * version 1.4.0.10 (1.4.1 release candidate 1)
3295 - don't keep a lock reference when lock is not granted (4238)
3296 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3297 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3298 - drop import inflight refcount on signal_completed_replay error (5255)
3299 - unlock page after async write error during send (3677)
3300 - handle missing objects in filter_preprw_read properly (5265)
3301 - no transno return for symlink open, don't save no-trasno open (3440)
3302 - don't try to complete elan receive that already failed (4012)
3303 - free RPC server reply state on error (5406)
3304 - clean up thread from ptlrpc_start_thread() on error (5160)
3305 - readahead could read extra page into cache that wasn't ejected (5388)
3306 - prevent races in class_attach/setup/cleanup/detach (5260)
3307 - don't dereference de->d_inode after l_dput of de (5458)
3308 - use "int" for stripe value returned from lock_to_stripe (5544)
3309 - mballoc allocation and error-checking fixes in 2.6 (5504)
3310 - block device patches to fix I/O request sizes in 2.6 (5482)
3311 - look up hostnames for IB nals (5602)
3312 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3313 - don't start multiple acceptors for the same port (5277)
3314 - fix incorrect LASSERT in mds_getattr_name (5635)
3315 - export a proc file for general "ping" checking (5628)
3316 - fix "lfs check" to not block when the MDS is down (5628)
3318 - service request history (4965)
3319 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3320 - create an "evict_client" /proc entry on OSTs, like the MDS has
3321 - fix mount usage message, return errors per mount(8) (5168)
3322 - change grep [] to grep "[]" in tests so they work in more UMLs
3323 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3324 - remove ancient LOV_MAGIC_V0 EA support (5047)
3325 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3326 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3327 - allow readahead allocations to fail when low on memory (5383)
3328 - mmap locking landed again, after considerable improvement (2828)
3329 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3331 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3334 - send OST transaction number in read/write reply to free req (4966)
3335 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3336 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3337 - the watchdog thread now runs as interruptible (5246)
3338 - flock/lockf fixes (but it's still disabled, pending 5135)
3339 - don't use EXT3 constants in llite code (5094)
3340 - memory shortage at startup could cause assertion (5176)
3342 - reorganization of lov code
3343 - single portals codebase
3345 - add extents/mballoc support (5025)
3346 - direct I/O reads in the obdfilter (4048)
3347 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3349 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3352 - send OST transaction number in read/write reply to free req (4966)
3353 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3354 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3355 - the watchdog thread now runs as interruptible (5246)
3356 - handle missing objects in filter_preprw_read properly (5265)
3357 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3358 - drop import inflight refcount on signal_completed_replay error (5255)
3359 - unlock page after async write error during send (3677)
3360 - return original error code on reconstructed replies (3761)
3361 - no transno return for symlink open, don't save no-trasno open (3440)
3363 - add pid to ldlm debugging output (4922)
3364 - bump the watchdog timeouts -- we can't handle 30sec yet
3365 - extra debugging for orphan dentry/inode bug (5259)
3367 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3370 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3371 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3372 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3373 - fix NULL dereference in /proc/sys/portals/routes (4827)
3374 - allow failed mdc_close() operations to be interrupted (4561)
3375 - stop precreate on OST before MDS would time out on it (4778)
3376 - don't send partial-page writes before EOF from client (4410)
3377 - discard client grant for sub-page writes on large-page clients (4520)
3378 - don't free dentries not owned by NFS code, check generation (4806)
3379 - fix lsm leak if mds_create_objects() fails (4801)
3380 - limit debug_daemon file size, always print CERROR messages (4789)
3381 - use transno after validating reply (3892)
3382 - process timed out requests if import state changes (3754)
3383 - update mtime on OST during writes, return in glimpse (4829)
3384 - add mkfsoptions to LDAP (4679)
3385 - use ->max_readahead method instead of zapping global ra (5039)
3386 - don't interrupt __l_wait_event() during strace
3388 - add software watchdogs to catch hung threads quickly (4941)
3389 - make lustrefs init script start after nfs is mounted
3390 - fix CWARN/ERROR duplication (4930)
3391 - return async write errors to application if possible (2248)
3392 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3393 - print NAL number in %x format (4645)
3394 - update barely-supported suse-2.4.21-171 series (4842)
3395 - support for sles 9 %post scripts
3396 - support for building 2.6 kernel-source packages
3397 - support for sles km_* packages
3399 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3402 - ignore -ENOENT errors in osc_destroy (3639)
3403 - notify osc create thread that OSC is being cleaned up (4600)
3404 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3405 - reconstruct ost_handle() like mds_handle() (4657)
3406 - create a new thread to do import eviction to avoid deadlock (3969)
3407 - let lconf resolve symlinked-to devices (4629)
3408 - don't unlink "objects" from directory with default EA (4554)
3409 - hold socknal file ref over connect in case target is down (4394)
3410 - allow more than 32000 subdirectories in a single directory (3244)
3411 - fix blocks count for O_DIRECT writes (3751)
3412 - OST returns ENOSPC from object create when no space left (4539)
3413 - don't send truncate RPC if file size isn't changing (4410)
3414 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3415 - bind to privileged port in socknal and tcpnal (3689)
3417 - rate limit CERROR/CWARN console message to avoid overload (4519)
3418 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3419 - basic mmap support (3918)
3420 - kernel patch series update from b1_4 (4711)
3422 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3425 - avoid crash during MDS cleanup with OST shut down (2775)
3426 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3427 - don't use bad inodes on the MDS (3744)
3428 - dynamic object preallocation to improve recovery speed (4236)
3429 - don't hold spinlock over lock dumping or change debug flags (4401)
3430 - don't zero obd_dev when it is force cleaned (3651)
3431 - print grants to console if they go negative (4431)
3432 - "lctl deactivate" will stop automatic recovery attempts (3406)
3433 - look for existing locks in ldlm_handle_enqueue() (3764)
3434 - don't resolve lock handle twice in recovery avoiding race (4401)
3435 - revalidate should check working dir is a directory (4134)
3437 - don't always mark "slow" obdfilter messages as errors (4418)
3439 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3442 - don't close LustreDB during write_conf until it is done (3860)
3443 - fix typo in lconf for_each_profile (3821)
3444 - allow dumping logs from multiple threads at one time (3820)
3445 - don't allow multiple threads in OSC recovery (3812)
3446 - fix debug_size parameters (3864)
3447 - fix mds_postrecov to initialize import for llog ctxt (3121)
3448 - replace config semaphore with spinlock (3306)
3449 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3450 - don't allow enqueue to complete on a destroyed export (3822)
3451 - down write_lock before checking llog header bitmap (3825)
3452 - recover from lock replay timeout (3764)
3453 - up llog sem before sending rpc (3652)
3454 - reduce ns lock hold times when setting kms (3267)
3455 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3456 - fix NULL deref and obd_dev leak on setup error (3312)
3457 - replace some LBUG about llog ops with error handling (3841)
3458 - don't match INVALID dentries from d_lookup and spin (3784)
3459 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3460 - fix invalid assertion in ptlrpc_set_wait (3880)
3462 - add libwrap support for the TCP acceptor (3996)
3463 - add /proc/sys/portals/routes for non-root route listing (3994)
3464 - allow setting MDS UUID in .xml (2580)
3465 - print the stack of a process that LBUGs (4228)
3467 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3470 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3471 - make sure to unset replay flag from failed open requests (3440)
3472 - if default stripe count is 0, use OST count for inode size (3636)
3473 - update parent mtime/ctime on client for create/unlink (2611)
3474 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3475 - free recovery state on server during a forced cleanup (3571)
3476 - unregister_reply for resent reqs (3063)
3477 - loop back devices mounting and status check on 2.6 (3563)
3478 - fix resource-creation race that can provoke i_size == 0 (3513)
3479 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3480 - more debugging for page-accounting assertion (3746)
3481 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3482 - avoid deadlock after precreation failure (3758)
3483 - fix race and lock order deadlock in orphan handling (3450, 3750)
3484 - add validity checks when grabbing inodes from l_ast_data (3599)
3486 - add /proc/.../recovery_status to obdfilter (3428)
3487 - lightweight CDEBUG infrastructure, debug daemon (3668)
3488 - change default OSC RPC parameters to be better on small clusters
3489 - turn off OST read cache for files smaller than 32MB
3490 - install man pages and include them in rpms (3100)
3491 - add new init script for (un)mounting lustre filesystems (2593)
3492 - run chkconfig in %post for init scripts (3701)
3493 - drop scimac NAL (unmaintained)
3495 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3498 - clean kiobufs before and after use (3485)
3499 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3500 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3501 - change init script to fail more clearly if not run as root (1528)
3502 - allow clients to reconnect during replay (1742)
3503 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3504 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3505 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3506 - fix cond_resched (3554)
3507 - only evict unfinished clients after recovery (3515)
3508 - allow bulk resend, prevent data loss (3570)
3509 - dynamic ptlrpc request buffer allocation (2102)
3510 - don't allow unlinking open directory if it isn't empty (2904)
3511 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3512 - remove extraneous obd dereference causing LASSERT failure (3334)
3513 - don't use get_cycles() when creating temp. files on the mds (3156)
3514 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3515 - handle EEXIST for set-stripe, set proper directory name (3336)
3517 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3518 - fix ksocknal_fmb_callback() error messages (2918)
3520 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3523 - don't copy lvb into (possibly NULL) reply on error (2983)
3524 - don't deref dentry after dput, don't free lvb on error (2922)
3525 - use the kms to determine writeback rpc length (2947)
3526 - increment oti_logcookies when osc is inactive (2948)
3527 - update client's i_blocks count via lvb messages (2543)
3528 - handle intent open/close of special files properly (1557)
3529 - mount MDS with errors=remount-ro, like obdfilter (2009)
3530 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3531 - don't use cancelling-locks' kms values (2947)
3532 - use highest lock extent for kms, not last one (2925)
3533 - don't dereference ERR_PTR() dentry in error handling path (3107)
3534 - fix thread race in portals_debug_dumplog() (3122)
3535 - create lprocfs device entries at setup instead of at attach (1519)
3536 - common AST error handler, don't evict client on completion race (3145)
3537 - zero nameidata in detach_mnt in 2.6 (3118)
3538 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3539 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3540 - initialize RPC timeout timer earlier for 2.6 (3219)
3541 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3542 - print nal/nid for unknown nid (3258)
3543 - additional checks for oscc recovery before doing precreate (3284)
3544 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3545 - don't crash in mdc_close for bad permissions on open (3285)
3546 - zero i_rdev for non-device files (3147)
3547 - clear page->private before handing to FS, better assertion (3119)
3548 - tune the read pipeline (3236)
3549 - fix incorrect decref of invalidated dentry (2350)
3550 - provide read-ahead stats and refine rpc in flight stats (3328)
3551 - don't hold journal transaction open across create RPC (3313)
3552 - update atime on MDS at close time (3265)
3553 - close LDAP connection when recovering to avoid server load (3315)
3554 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3555 - don't leak open file on MDS after open resend (3325)
3556 - serialize filter_precreate and filter_destroy_precreated (3329)
3557 - loop device shouldn't call sync_dev() for nul device (3092)
3558 - clear page cache after eviction (2766)
3559 - resynchronize MDS->OST in background (2824)
3560 - refuse to mount the same filesystem twice on same mountpoint (3394)
3561 - allow llmount to create routes for mounting behind routers (3320)
3562 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3563 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3564 - fix rare problem with rename on htree directories (3417)
3566 - allow default OST striping configuration per directory (1414)
3567 - fix compilation for qswnal for 2.6 kernels (3125)
3568 - increase maximum number of MDS request buffers for large systems
3569 - change liblustreapi to be useful for external progs like lfsck (3098)
3570 - increase local configuration timeout for slow disks (3353)
3571 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3573 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3576 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3577 - don't overwrite extent policy data in reply if lock was blocked (2901)
3578 - drop filter export grants atomically with removal from device (2663)
3579 - del obd_self_export from work_list in class_disconnect_exports (2908)
3580 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3581 - swab reply message in mdc_close, other PPC fixes (2464)
3582 - fix destroying of named logs (2325)
3583 - overwrite old logs when running lconf --write_conf (2264)
3584 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3585 - fix race in target_handle_connect (2898)
3586 - mds_reint_create() should take same inode create lock (2926)
3587 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3588 - don't close files for self_export to avoid uninitialized obd (2936)
3589 - allow MDS with the same name as client node (2939)
3590 - hold dentry reference for closed log files for unlink (2325)
3591 - reserve space for all logs during transactions (2059)
3592 - don't evict page beyond end of stripe extent (2925)
3593 - don't oops on a deleted current working directory (2399)
3594 - handle hard links to targets without a parent properly (2517)
3595 - don't dereference NULL lock when racing during eviction (2867)
3596 - don't grow lock extents when lots of conflicting locks (2919)
3598 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3601 - account for cache space usage on clients to avoid data loss (974)
3602 - lfsck support in lustre kernel code (2349)
3603 - reduce journal credits needed for BRW writes (2370)
3604 - orphan handling to avoid losing space on client/server crashes
3605 - ptlrpcd can be blocked, stopping ALL progress (2477)
3606 - use lock value blocks to assist in proper KMS, faster stat (1021)
3607 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3608 - recovery for initial connections (2355)
3609 - fixes for mds_cleanup_orphans (1934)
3610 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3611 - block all file creations until orphan recovery completes (1901)
3612 - client remove rq_connection from request struct (2423)
3613 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3614 - recovery timer race (2670)
3615 - mdc_close recovey bug (2532)
3616 - ptlrpc cleanup bug (2710)
3617 - mds timeout on local locks (2588)
3618 - namespace lock held during RPCs (2431)
3619 - handle interrupted sync write properly (2503)
3620 - don't try to handle a message that hasn't been replied to (2699)
3621 - client assert failure during cleanup after abort recovery (2701)
3622 - leak mdc device after failed mount (2712)
3623 - ptlrpc_check_set allows timedout requests to complete (2714)
3624 - wait for inflight reqs when ptlrpcd finishes (2710)
3625 - make sure unregistered services are removed from the srv_list
3626 - reset bulk XID's when resending them (caught by 1138 test)
3627 - unregister_bulk after timeout
3628 - fix lconf error (2694)
3629 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3630 - readahead locks pages, leaves pending causing memory pressure (2673)
3631 - increase OST request buffers to 4096 on large machines (2729)
3632 - fix up permission of existing directories in simple_mkdir (2661)
3633 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3634 - don't assign transno to errored transactions (2742)
3635 - don't delete objects on OST if given a bogus objid from MDS (2751)
3636 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3637 - if rq_no_resend, then timeout request after recovery (2432)
3638 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3639 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3640 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3641 - MDS recovery shouldn't delete objects that it has given out (2730)
3642 - if enqueue arrives after completion, don't clobber LVB (2819)
3643 - don't unlock pages twice when trigger_group_io returns error (2814)
3644 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3645 - don't write pages to disk if there was an error (1450)
3646 - don't ping imports that have recovery disabled (2676)
3647 - take buffered bytes into account when balancing socknal conn (2817)
3648 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3649 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3650 - remove little-endian swabbing of llog records (1987)
3651 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3652 - retry reposting request buffers if they fail (1191)
3653 - grow extent at grant time to avoid granting a revoked lock (2809)
3654 - lock revoke doesn't evict page if covered by a second lock (2765)
3655 - disable VM readahead to avoid reading outside lock extents (2805)
3657 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3658 - updated kernel patches for hp-2.4.20 kernel (2681)
3660 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3663 - fix truncated write corruption (2366)
3664 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3666 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3667 - ensure len(uuid) < 37 in lmc (1171)
3668 - fix ia64 OOPS in llog_test (2255)
3669 - zero end of page at obdfilter for partial page writes (2648)
3670 - don't leave stale dentries around after renames (bug 2428)
3671 - fix timeouts when evicting a client with a single lock held (2642)
3672 - set deadline for the initial HELLO message to drain (2634)
3673 - print out dotted-quad IP addresses in the socknal (2302)
3675 - additional debugging for MDS client eviction problem (2443)
3676 - fix mkfsoptions support for osts (2603, 2604)
3678 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3681 - add series for the vanilla 2.6.0 kernel
3682 - add series for the vanilla 2.4.24 kernel
3683 - add series for a cray x86/64 UL kernel drop
3684 - fix xattr patches for the vanilla 2.4.19 series
3686 - generate true UUIDs in lmc (1171)
3687 - have portals stack dumping break in UML (2466)
3688 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3689 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3690 - wake callback waiting threads on client eviction (2460)
3691 - Add --ptldebug and --subsystem to lmc (1719)
3692 - update assertion to allow safe interrupt allocation
3693 - set rq_no_resend for cancel requests (2432)
3694 - recalculate ptlrpcd timeout after resend (2494)
3695 - call vfs_rmdir when removing pending directories (2368)
3696 - fix renaming a file to itself (2429)
3697 - lmc creates a default one-stripe lov (2454)
3698 - expand procfs space to handle large clusters (2326)
3699 - increase UML stack to avoid overflow
3700 - update lconf's list of debug and subsystem masks
3701 - fix lfs find --obd (2510)
3702 - /proc tunable for disabling filter read caching (2591)
3703 - stop rpm packages from altering slapd.conf (2301)
3704 - disable nagle in the socknal under 0conf (2578)
3705 - choose mds inode size based on stripe count (2572)
3706 - fix kernel-source rpm problems (2516)
3708 - add --disable-doc to avoid pdf generation (2421)
3709 - update documentation, tests, type-os, comments
3710 - avoid format warnings on ia64
3711 - remove the TOE NAL
3712 - tiny code cleanups by removing unused fields
3714 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3717 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3718 - remove the most blatant lies from BUILDING file (2371)
3719 - change default debug level to reasonable production setting
3720 - reduce client side cache size to reduce cache flush time
3721 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3722 - make TCP zerocopy and pinger support enabled by default (2476)
3723 - sync writes completed after process exits caused crashes (2319)
3724 - maintain correct mount count on the MDS (2356)
3725 - backout 1557, because 2316 wasn't really fixed
3726 - better file I/O statistics gathering in /proc
3727 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3728 - another kernel patch to fix zero-copy TCP function export
3729 - don't take duplicate lock when processing re-sent getattr (2420)
3730 - lctl uses obd_self_export instead of creating new conn (2353)
3731 - MDS/OST recovery case which requires object creation asserted (2425)
3732 - move lfs from /usr/sbin to /usr/bin in packages
3733 - fix race between mds_client_add and mds_client_free (2417)
3734 - use kmalloc instead of slabs in portals (2430)
3735 - don't create duplicate records when a failover MDS is present (2442)
3736 - remove unnecessary mount age check (2332)
3737 - don't remove directory inodes from locks prematurely (2451)
3738 - don't break if MDS service name is the same as hostname (2103)
3739 - fix races in client write RPC generation when cache full (2482)
3741 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3744 - remove now-unused request->rq_obd (278)
3745 - if an allocation fails, print out how much memory we've used (1933)
3746 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3747 - add the "configurable stack size" patch to most series files (1256)
3748 - ability to write large log records, for 100+ OST configs (2306)
3749 - fix NULL deref when filter_prep fails (2314)
3750 - fix operator precedence error in filter_sync
3751 - dynamic allocation of socknal TX descriptors (2315)
3752 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3753 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3754 - ensure that CWARN messages in Portals always get to the syslog
3755 - __init/__exit are not for prototype decls (ldlm_init/exit)
3756 - x86-64 compile warning fixes
3757 - fix gateway LMC keyword conflict (2318)
3758 - fix MDS lock inversions in getattr/reint paths (1844)
3759 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3760 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3761 - DLM race condition prevented some lock evictions (2328)
3762 - ENOMEM detection and retry on socknal sends (2230)
3763 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3764 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3765 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3766 - better error messages when a client is rejected during recovery (1505)
3767 - avoid cancelling locks which were never granted, after failure (2330)
3768 - fix i_sem/journal inversion in mds_client_add (2333)
3769 - fix truncate/getattr lock cycle deadlock (2334)
3770 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3771 - fix two rare exit paths which could leak an l_lock() ref (2321)
3772 - fencepost error in MDS/OST orphan recovery (2226)
3773 - make log record alignment 8 bytes (1988)
3774 - lstripe now fails when requested offset > ost_count (2237)
3775 - ensure that all kernel series have a complete list.h (1607)
3776 - fix crashes in special-file operations (2316)
3777 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3779 - allow configurable automake binary, for testing new versions
3780 - small update to the lfs documentation
3782 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3784 * fix negative export reference count in fsfilt_sync (2312)
3786 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3787 * release candidate 0.9.1
3789 - orphans are moved into the PENDING directory for possible recovery
3790 - replayed opens now open by fid for orphan/rename safety (1042)
3791 - last close of an orphan inode generates a transno (683)
3792 - chdir() and mount() now pin the directory entry (1020)
3793 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3794 - discard very old requests without processing them (1502)
3795 - remove some common, well-understood CERRORs (1505)
3796 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3797 - clear "grant" flags in OST replies until OST grant code lands (1644)
3798 - fix read performance by not clobbering i_blksize on client (1598)
3799 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3800 - make LRU size a /proc tunable, clears locks when reduced (707)
3801 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3802 - server locks take references on exports now (1558)
3803 - build fixes for 2.4.20-rh trees (1663)
3804 - return an error from lov_create if all OSCs are inactive (1751)
3805 - fix import levels when a reconnect happens without a timeout (1597)
3806 - exit early from mds_open if we get a lookup error (1749)
3807 - partial page read at EOF wouldn't wait for disk before sending (1642)
3808 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3809 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3810 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3811 - write/truncate lock inversion (1639)
3812 - Don't auto-load obdclass, portals modules during cleanup (1495)
3813 - fix timestamps from jumping to "now" (1763)
3814 - extra journal assertions (1648)
3815 - add an extra multiunlink test (1771)
3816 - fix read_record/write_record API (1776)
3817 - fix leak of offset_extent, possible incorrect i_size later (1772)
3818 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3819 - Debugging for the kqswnal_get_idle_tx problems (1820)
3820 - Allow recovery to be attempted multiple times (1536)
3821 - Write out MDS last_rcvd file after it is first created (1600)
3822 - Fix tx_descriptor leak in failed transmit situations (1827)
3823 - ext3 journaling fixes for assertion failure after IO error (1871)
3824 - class_export_put() on freed export after completion AST error (1896)
3825 - Fix revalidate looping in VFS (1322)
3826 - Don't access a freed export during MDS_REINT timeout (1521)
3827 - Add open-unlink recovery support on the MDS (1673,1764)
3828 - Return an error if no MDS data was read from last_rcvd (1946)
3829 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3830 - Don't setup a disk device that doesn't match exported UUID (317)
3831 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3832 - avoid committing NULL handle in force close
3833 - local.sh is now a one-stripe LOV configuration
3834 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3835 - fix ext3 htree duplicate directory entry corruption (1516)
3836 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3837 - update to new LOV EA format (2097)
3838 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3840 2003-06-15 Phil Schwan <phil@clusterfs.com>
3843 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3844 - per-import/export recovery handling (958, 931, 959)
3845 - multiple last-rcvd slots, for serving multiple FSes (949)
3846 - connections are again shared between multiple imp/exports (963, 964)
3847 - "umount -f" would hang if any requests needed to be sent (393, 978)
3848 - avoid pinning large req buffer by copying for queued messages (989)
3849 - add "uuid" to "lctl device" command to help upcalls (991)
3850 - "open" RPCs with transnos would confuse recovery counters (1037)
3851 - do proper endian conversion of all wire messages (288, 340, 891)
3852 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3853 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3854 - fix ptlrpc_connection leak in target_handle_connect (1174)
3855 - fix import refcounting bug in OST and MDS cleanup (1134)
3856 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3857 - fix very unlikely obd_types race condition (501)
3858 - remove osc_open hack for echo_client (1187)
3859 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3860 - a failure in read_inode2 leads to deadlock (1139)
3861 - cancel ack-locks as soon as transaction is committed (1072)
3862 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3863 - make sure to commitrw after any preprw to avoid deadlock (1162)
3864 - failing to execute a file in a lustre FS would lock inode (1203)
3865 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3866 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3867 - obd_uuid2tgt would walk off the end of the list (1255)
3868 - on IA64 the osc would give portals incorrect bulk size (1258)
3869 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3870 - fix lock inversion caused by new llite matching code (1282)
3871 - limit the number of dirty pages on a client to 10MB (1286)
3872 - timed out locks were not being corrected cancelled (1289)
3873 - fix O_DIRECT above 4GB on IA-32 (1292)
3874 * major user-visible changes
3875 - fail out/fail over policy now controlled by the upcall (993)
3877 - add OBD_PING to check server availability and failure (954)
3878 - lustre messages are now sent in sending host order (288, 340, 891)
3879 - add eadatalen to MDS getattr reply (340)
3880 - OST read replies may contain second buffer, with per-page status (593)
3882 2003-03-11 Phil Schwan <phil@clusterfs.com>
3885 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3886 - failed open()s could cause deadlock; fixed (867, 869)
3887 - stop cancelling OST locks when files are closed (481)
3888 - overlapping XID spaces caused network corruption (851, 853)
3889 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3890 - setattr_raw added, to do single-RPC, server-side setattrs
3891 - lmc/lconf syntax change for OST UUIDs
3892 - fix crashy race condition between ptlrpc_free_req and osc_close
3893 - don't use request in mdc_enqueue if we hit a timeout (889)
3894 - don't set the inode i_size for regular files from the MDS (896)
3895 - handle out of order completion AST (842)
3896 - don't LBUG if a lock request times out after receiving AST (913)
3897 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3898 - if a bad lock AST arrives, send an error instead of dropping entirely
3899 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3900 - fix leak in bulk IO when only partially completed (899, 900, 926)
3901 - fix O_DIRECT for ia64 (55)
3902 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3903 - C-z after timeout could hang a process forever; fixed (977)
3905 - client-side I/O cache (678, 924, 929, 941, 970)
3907 - READPAGE and SETATTRs which don't take server-side locks get
3910 2003-02-11 Phil Schwan <phil@clusterfs.com>
3913 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3914 - Fix fsfilt_extN_readpage() to read a full page of directory
3915 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3916 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3917 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3918 - Most intent operations were removed, in favour of a new RPC mode
3919 that does a single RPC to the server and bypasses most of the VFS
3920 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3921 - Aggressively cancel local locks on DLM servers
3922 - mds_reint_unlink sends EA to the client if it's the last nlink.
3923 client uses that EA to unlink OST objects.
3924 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3925 - recursive symlinks were fixed (439)
3926 - fixed NULL deref in DEBUG_REQ
3927 - filter_update_lastobjid no longer calls sync, which annoyed extN
3928 - fixed multi-client small-writes to a single file problem (445)
3929 - fixed mtime updates during file writes (607)
3930 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3931 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3932 - fixed Linux OST size reporting problem (444, 656)
3933 - OST now updates object mtime with writes or setattr (607, 619)
3934 - client verifies file size before zeroing page past EOF (445)
3935 - OST now writes last allocated objid to disk with allocation (108)
3936 - LOV on echo now works (409)
3938 - mds_reint_unlink sends a new buffer, with the EA included. this
3939 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3940 set if a regular file was being unlinked, and it was the last link
3941 - use PtlGet from the target for bulk writes (315)
3942 - OST now updates object mtime with writes or setattr (607, 619)
3943 - LDLM now has a grant-time callback to revalidate locked items, if
3945 - Many MDS operations were reorganized to combat race conditions
3947 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3948 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3950 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3953 - Fully reactivate OST imports after reconnection (512, others)
3954 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3955 - More graceful error handling for truncating on dead OST (515)
3956 - Don't error out unless we're actually accessing dead stripes (474)
3957 - Fix garbage sizes when stripes are missing (410)
3958 - LRU counters were broken, causing constant lock purge (433, 432)
3959 - garbage on read from stripes with failed OSTs (441)
3960 - mark OSCs as active before reconnecting during recovery (438)
3961 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3962 - lfind did not preserve OST order in output (443)
3963 - symlinks cause hung clients, incorrect data (439)
3964 - stop dereferencing request after dropping refcount (457)
3965 - don't LASSERT(spin_is_locked) on non-SMP (455)
3966 - fixes for many rename() bugs
3967 - fstat didn't correctly synchronize attributes (399)
3968 - server must handle lock cancellation during blocking AST prep (487)
3969 - bulk descriptors were free()d too soon (511)
3970 - fix paths in lconf, which would load incorrect modules (451, 507)
3971 - fix confusing lconf 'host not found' error message (386)
3972 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3973 - fix race condition in mdc_blocking_ast() for inode access (526)
3974 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3975 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3976 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3977 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3978 - fix lstripe in several ways (kernel side) (527)
3979 - fix request leak in ldlm_cli_enqueue (262)
3980 - incorrect OSC was marked inactive after OST failure
3981 - call mds_fs_cleanup before unmounting filesystem (524)
3982 - fix races between taking ns_lock and ldlm_lock_change_resource
3983 - fix races updating LOV export open file list
3984 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3985 - fix recovery NULL deref in ldlm_cli_cancel_unused
3986 - fix some DLM races by using new hash table for lock handles (419)
3987 - permit the client to specify desired inodes, at replay
3988 - duplicate requests when we queue them for replay reintegration
3989 - fix last_rcvd offset calculation
3990 - sync after each recovered transaction, so we always make progress
3991 - never, not always, ERESTART requests without transnos
3992 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3994 - skip replay if the MDS says that the client is already connected
3995 - don't check for a recovery-enabled export to match lctl's UUID
3996 - don't INC_USE_COUNT for phantom exports
3997 - don't crash when cleaning up phantom exports (567)
3998 - don't double-finish or set replay data for errored mdc_open requests
3999 - abort requests when they time out, so we don't get old replies
4000 - send/receive replies for AST messages again
4001 - if the client says that it doesn't have the lock, cancel it on the
4003 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4004 mark it as destroyed, it will all work out when decref is called
4005 - fix module use counts (22, 581)
4007 - ASTs now expect a reply (server cancels lock on error reply)
4009 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4012 - fix many simultaneous client startup (392)
4013 - fix dentry->d_it clobbering
4014 - credentials weren't being shipped for readdir/getattr operations
4015 - remove invalid assertions triggered during some concurrent MD
4017 - proper Lustre versions added (336, 389)
4018 - fix memory leak for create error case (398)
4019 - fix LOV locking bug that would get cli/srv out of sync
4020 - fix echo client over LOV (409)
4021 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4022 - fix double-O_EXCL intent crash (424)
4023 - avoid sending multiple lock CANCELs (352)
4025 - MDS can do multi-client recovery (modulo bugs in new code)
4027 - many updates, edits, cleanups
4029 2002-11-18 Phil Schwan <phil@clusterfs.com>
4032 - fix null d_it dereference (346)
4033 - fix full OST/dbench hang (333)
4034 - fix permission problem with file removal (286)
4035 - fix removal of OSCs from LOV when they fail
4036 - fix NULL deref during bulk timeout (214)
4037 - fix problems related to multiple filesystems on one MDS (241)
4038 - fixed serious subtle metadata locking bugs
4039 - free locks on clients when inodes are removed due to memory
4041 - fix inode pointer in lock data (285)
4042 - partial support for multiple MDS on a single host (241)
4043 - data locks weren't cancelled at clear_inode time (290, 311)
4044 - intent locks could lead to unbounded lock growth (205)
4045 - added a maximum lock count, an LRU list, and a flusher
4046 - fix multiple rename (365)
4047 - properly abstracted the echo client
4048 - OSC locked 1 byte too many; fixed
4049 - rewrote brw callback code:
4050 - fixed recovery bugs related to LOVs (306)
4051 - fixed too-many-pages-in-one-write crash (191)
4052 - fixed (again) crash in sync_io_timeout (214)
4053 - probably fixed callback-related race (385)
4055 - Add capability to MDS protocol
4056 - LDLM cancellations and callbacks on different portals
4058 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4061 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4062 - defer open object destruction to close time (601981, 138)
4063 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4064 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4065 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4066 - don't call obd_setattr in ll_file_release for destroyed objects
4067 * protocol change to lustre_msg: move |version| and add |flags|
4068 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4069 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4070 * added replay of create, unlink, link and rename operations during
4071 MDS failover; recovery should be much more robust now
4072 * remove failed OSCs from LOVs (only lov_create uses this so far)
4073 * the lustre-HOWTO was brought (more) up to date (582544)
4075 2002-10-23 Phil Schwan <phil@clusterfs.com>
4078 - in-use dentries weren't being reused properly (617851)
4079 - prevent multiple LDLM setup (599178)
4080 - fix LOV size calculations for truncate (617853)
4081 - fix client handling of MDS intent errors (POSIX)
4082 - fix permission bug in lovstripe.c test (624321)
4083 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4084 - truncate past end of file could corrupt data
4085 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4086 - a race in recovery could return ETIMEDOUT to apps (623947)
4087 - building outside the source directory was fixed
4088 * the lustre-HOWTO was brought (more) up to date (582544)
4089 * major progress was made on recovery functionality
4091 2002-10-10 Phil Schwan <phil@clusterfs.com>
4094 - recovery deadlock fix
4095 - rm -rf causes LBUG fix (617817)
4096 - file open by multiple tasks fix (618962)
4097 - directory permissions bugs (602707 and 620007)
4098 - journal_stop fixed with locking (611313)
4099 - O_APPEND failures resolved (618273, perhaps 614459)
4100 - lconf PATH fix (619770)
4101 - IA64 build fix (621450)
4102 - RPC buffer sizes scale with amount of memory
4104 2002-10-01 Phil Schwan <phil@clusterfs.com>
4107 - locks would be cancelled without throwing away data pages,
4108 resulting in inconsistent data (605627)
4109 - inode attributes were not always being refreshed (605627, 612449)
4110 - lconf now continues to cleanup after lctl reports an error
4111 - MDS now enforces user permissions (602707)
4112 - lprocfs cleanup fixed, but not yet enabled (614157)
4113 - fixed infinite server hang, should a client not respond to an AST
4114 - avoid going into recovery if user calls readlink() with a buffer
4115 that's too small (613941)
4116 - AST RPCs no longer require replies (614867) -- this may be changed
4117 - don't crash server if client sends an IOV that's too big (611336)
4118 - fixed lock conversion deadlock (611892)
4119 - fixed the following of symlinks (614622)
4120 * recovery: the server can remove locks from a client that dies, other
4121 clients can make progress
4122 * more extN patch fixes
4123 * compile-time configurable ptlrpc buffer allocations
4125 - collaborative read cache document
4126 - Lustre Lite Performance CDR document-in-progress
4128 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4131 - fix typo in patch-2.4.18
4133 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4136 - clear ptlrpc request each time in handle_incoming_request()
4137 - unlink of files now destroys the object on the OST
4139 2002-09-19 Peter Braam <braam@clusterfs.com>
4141 * add hard link support
4142 * change obdfile creation method
4143 * kernel patch changed
4145 2002-09-19 Peter Braam <braam@clusterfs.com>
4148 - stack overflow bug in extN fixed
4150 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4152 * documentation updates
4153 - add man pages for config tools
4154 - update tests/README to describe testing with new config tools
4155 - finish metadata API descriptions
4156 * bug fixes and cleanups
4157 - statfs workaround for 16TB limit
4158 - LOV stripe allocation improved, can stripe on subset of OSTs
4159 - LOV file size/IO offset was wrong for files > 4GB in size
4160 - object EA data was being dropped, caused files to be unreadable
4161 - memory overflow with non-LOV OST caused memory corruption
4162 - fixed regression tests to work with new config tools, obdfilter
4163 - fixed bug when directory size became larger than 1 block
4164 - fixed bug (for single client case) when PWD was deleted
4165 - invalidate local directory pages when doing intent-based ops
4166 - avoid LDLM oops when lock callback contained bad data
4168 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4170 * documentation updates
4171 * bug fixes and cleanups
4172 - configuration tools
4175 - 64-bit compile warnings
4176 - 64-bit internal statfs data
4178 * test_brw on persistent OST devices
4180 * lprocfs (disabled)
4182 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4184 * documentation updates
4185 * bug fixes and cleanups
4186 * configuration tools
4188 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4191 * many small fixes to 0_5_4
4192 * io/network handling
4193 * thinkos in MDS operations
4195 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4198 * crucial basic fixes to 0.5.3
4199 * IOR, Iozone work over Elan
4202 2002-08-07 Phil Schwan <phil@clusterfs.com>
4203 * version 0_5_3, our first alpha
4204 * we use the new Portals iovs
4205 * documentation updates
4206 * bug fixes and cleanups
4207 * small changes in the DLM wire protocol
4209 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4210 * version 0_5_1 with some initial stability,
4211 * locking on MD and file I/O.
4212 * documentation updates
4213 * several bug fixes since 0.5.0
4214 * small changes in wire protocol
4216 2002-07-18 Phil Schwan <phil@clusterfs.com>
4218 * delivered as Lustre Light Alpha
4219 * fixed a crash after handling invalid MDS requests
4220 * fixed directory pages for architectures with non-4k pages sizes
4222 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4223 * release version v0_4_4
4224 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4226 * Fixes a number of interruption problems with OST operations.
4227 * Update documentation for portals header changes
4228 * Move all wire protocol structs/defines to lustre_idl.h
4229 * Fixes symlink length bug.
4230 * Add tcpdump to repository.
4232 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4233 * release version v0_4_3
4234 * Fixes statfs for inodes on extN.
4235 * Fixes bug in runtests which would delete /etc/hosts.
4236 * Use 64-bit object IDs wherever possible (not into VFS though)
4237 Remove ost_get_info, which is unused by lustre, and out of date.
4239 2002-07-03 Peter Braam <braam@clusterfs.com>
4240 * release version v0_4_2 Fixes a lookup error (type not passed)
4241 * move forward to head of Portals
4242 * move forward to latest Lustre kernel
4244 2002-06-25 Peter Braam <braam@clusterfs.com>
4245 * release version v0_4_1. Hopefully stable on single node use.