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: LASSERT_{REQ,REP}SWAB macros are buggy
18 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
19 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
20 these macros and replace them with their difinitions instead.
24 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
25 Details : Update which_patch & target file for SLES10 latest kernel.
28 Description: Remove client patches from SLES 10 kernel.
29 Details : This causes SLES 10 clients to behave as patchless clients
30 even on a Lustre-patched (server) kernel.
32 Severity : enhancement
34 Description: self-adjustable client's lru lists
35 Details : use adaptive algorithm for managing client cached locks lru
36 lists according to current server load, other client's work
37 pattern, memory activities, etc. Both, server and client
38 side namespaces provide number of proc tunables for controlling
41 Severity : enhancement
43 Description: light-weight GSS support
44 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
45 and reduce performance overhead.
47 Severity : enhancement
49 Description: Linux keyring support
50 Details : Support using service of Linux keyring for Lustre GSS internal
51 context refresh/cache mechanism.
55 Description: Fix errors in lfs documentation
56 Details : Fixes man pages
60 Description: don't use GFP_* in generic Lustre code.
61 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
65 Description: obdclass is limited by single OBD_ALLOC(idarray)
66 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
70 Description: warnings with build patchless client with vanila 2.6.19 and up
71 Details : change old ctl_table style and replace ctl_table/ctl_table_header
72 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
76 Frequency : Only for SLES
77 Description: sanity_quota fail test_1
78 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
79 file, which makes TSTID[2] inunique.
83 Frequency : Only for SLES9
84 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
85 Details : Update target/ChangeLog/which_patch .
89 Frequency : Only for test_10 in sanity-quota.sh
90 Description: a bug in quota test code
91 Details : checking a test flag in a wrong place
95 Frequency : Only for test_13 in sanity-quota.sh
96 Description: a bug in quota test script
97 Details : shouldn't check overfull allocated quota
101 Description: jbd statistics
102 Details : Port older jbd statistics patch for sles10
104 Severity : enhancement
106 Description: Add printing inode info into message about error in writepage.
111 Description: reply_lock_interpret crash due to race with it and lock cancel.
112 Details : Do not replay locks that are being cancelled. Do not reference
113 locks by their address during replay, just by their handle.
117 Frequency : When flocks are used.
118 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
119 Details : Flock locks might destroy just granted lock if it could be merged
120 with another existing flock, this is done in completion handler,
121 so teach ldlm_cli_enquque_fini that this is a valid case for
127 Description: Oops in read and write path when failing to allocate lock.
128 Details : Check if lock allocation failed and return error back.
132 Description: lstripe command fails for valid OST index
133 Details : The stripe offset is compared to lov->desc.ld_tgt_count
134 instead of lov->desc.ld_active_tgt_count.
138 Description: sanity.sh failed test 103
139 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
144 Description: df doesn't work properly if diskfs blocksize != 4K
145 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
149 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
150 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
154 Description: obd_health_check_timeout too short
155 Details : set obd_health_check_timeout as 1.5x of obd_timeout
159 Description: llapi_file_create() does not allow some changes
160 Details : add llapi_file_open() that allows specifying the mode and
161 open flags, and also returns an open file handle.
165 Description: merge and cleanup kernel patches.
166 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
170 Description: Add journal checksum support.(Kernel part)
171 Details : The journal checksum feature adds two new flags i.e
172 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
173 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
174 indicates that the commit block contains the checksum for
175 the blocks described by the descriptor blocks. Now commit
176 record can be sent to disk without waiting for descriptor
177 blocks to be written to disk. This behavior is controlled
178 using JBD2_FEATURE_ASYNC_COMMIT flag.
182 Description: OSS needs mutliple precreate threads
183 Details : Add ability to start more than one create thread per OSS.
187 Description: Sanity.sh test_65k failed: setstripe should have succeeded
188 Details : execute lfs setstripe on client
192 Description: mds_obd_create error creating tmp object
193 Details : When the user sets quota on root, llog will be affected and can't
194 create files and write files.
197 Frequency : Always on ia64 patchless client, and possibly others.
199 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
200 Details : This allows the patchless client to be loaded on architectures
205 Description: RedHat Update kernel for RHEL5
206 Details : Add the kernel config file for RHEL5.
210 Description: RedHat Update kernel for RHEL5
211 Details : Modify the kernel config file more closer RHEL5.
215 Description: Build failure against Centos5 (RHEL5)
216 Details : Define PAGE_SIZE when it isn't present.
220 Description: test_7 on sanity-quota.sh can't be used on separate machines
221 Details : add facet to handle it
225 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
226 Details : with NFS, the anon dentry's parent was set to itself in
227 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
228 corresponding dentry other than use rec->ur_name.
230 Severity : enhancement
232 Description: lfs setstripe enhancement
233 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
237 Description: testing performance impact of enabling checksumming
238 Details : enable checksum by default, allow --disable-checksum
239 configure option and "-o nochecksum" mount option
243 Description: lustre support for RHEL5
244 Details : Add support for RHEL5.
248 Description: Client eviction due to failover config
249 Details : after a connection loss, the lustre client should attempt to
250 reconnect to the last active server first before trying the
251 other potential connections.
255 Description: when mds and osts use different quota unit(32bit and 64bit),
256 quota will be released repeatly.
257 Details : void sending multiple quota reqs to mds, which will keep the
258 status between the reqs.
262 Description: osts not allocated evenly to files
263 Details : change the condition to increase offset_idx
266 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
268 Description: Data corruption for OSTs that are formatted larger than 2TB
270 Details : When generating the bio request for lustre file writes the
271 sector number would overflow a temporary variable before being
272 used for the IO. The data reads correctly from Lustre (which
273 will overflow in a similar manner) but other file data or
274 filesystem metadata may be corrupted in some cases.
278 Description: Tune the kernel for good SCSI performance.
279 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
280 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
285 Description: rewrite ext2-derived code in obdclass/uuid.c
286 Details : rewrite inherited code (uuid parsing code from ext2 utils)
287 from scratch preserving functionality.
291 Description: "lfs find -obd UUID" prints directories
292 Details : "lfs find -obd UUID" will return all directory names instead
293 of just file names. It is incorrect because the directories
294 do not reside on the OSTs.
298 Description: use i_size_read and i_size_write in 2.6 port
299 Details : replace inode->i_size access with i_size_read/write()
302 Frequency : when removing large files
304 Description: scheduling issue during removal of large Lustre files
305 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
306 It causes scheduling issues when removing large files (17TB in the
310 Frequency : only with liblustre clients on XT3
312 Description: evictions taking too long
313 Details : allow llrd to evict clients directly on OSTs
316 Frequency : only on ppc
318 Description: /proc/fs/lustre/devices broken on ppc
319 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
320 We should make sure the type of 'index' is loff_t and then cast
321 explicitly as needed below. Do not assign an explicitly cast
325 Frequency : only for rhel5
327 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
328 Details : Modify the target file & which_kernel.
330 Severity : enhancement
332 Description: omit set fsid for export NFS
333 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
334 as FS_REQUIRES_DEV for avoid problems with generate fsid.
338 Frequency : Always, for kernels after 2.6.16
339 Description: Fix warning idr_remove called for id=.. which is not
341 Details : Last kernels save old s_dev before kill super and not allow
342 to restore from callback - restore it before call kill_anon_super.
346 Description: buffer overruns could theoretically occur
347 Details : llapi_semantic_traverse() modifies the "path" argument by
348 appending values to the end of the origin string, and a
349 overrun may occur. Adding buffer overrun check in liblustreapi.
353 Description: Fix error on 'ls .' at the top of the Lustre mount.
354 Details : Don't revalidate dentry if it is a root dentry.
358 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
359 Details : Modify vm-tunables-rhel4.patch.
363 Description: Kernel config for 2.6.18-vanilla.
364 Details : Modify targets/2.6-vanilla.target.in.
365 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
366 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
367 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
368 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
372 Description: improve handling recoverable errors
373 Details : if request processig with error which can be recoverable on server
374 request should be resend, otherwise page released from cache and
379 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
380 Details : Modify target file & which_patch.
381 A flaw was found in the IA32 system call emulation provided
382 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
383 value could be stored in the %RAX register, which could trigger an
384 out-of-bounds system call table access. An untrusted local user
385 could exploit this flaw to run code in the kernel
386 (ie a root privilege escalation). (CVE-2007-4573).
390 Description: Remove 2.4 kernel checks lustre 1.8
391 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
392 quota_interface.c in quota subsystem.
393 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
395 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
396 lvfs_linux.c upcall_cache.c in lvfs subsystem.
400 Description: O_DIRECT bypasses client statistics.
401 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
402 read_ahead_stats were updated. copied the stats section from
403 osc_send_oap_rpc() into async_internal().
407 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
411 Description: change order of libsysio includes
412 Details : '#include sysio.h' should always come before '#include xtio.h'
416 Description: handle "serious error: objid * already exists" more gracefully
417 Details : If LAST_ID value on disk is smaller than the objects existing in
418 the O/0/d* directories, it indicates disk corruption and causes an
419 LBUG(). If the object is 0-length, then we should use the existing
420 object. This will help to avoid a full fsck in most cases.
422 Severity : enhancement
424 Description: adapt the lustre_config script to support the upgrade case
425 Details : Add "-u" option for lustre_config script to support upgrading 1.4
426 server targets to 1.6 in parallel.
430 Description: To avoid grant space > avaible space when the disk is almost
431 full. Without this patch you might see the error "grant XXXX >
432 available" or some LBUG about grant, when the disk is almost
434 Details : In filter_check_grant, for non_grant cache write, we should
435 check the left space by if (*left > ungranted + bytes), instead
436 of (*left > ungranted), because only we are sure the left space
437 is enough for another "bytes", then the ungrant space should be
438 increase. In client, we should update cl_avail_grant only there
439 is OBD_MD_FLGRANT in the reply.
443 Description: Update RHEL 4 kernel to fix local root privilege escalation.
444 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
445 described in CVE-2007-4573. This problem could allow untrusted
446 local users to gain root access.
450 Description: Remove 2.4 kernel checks lustre 1.8
451 Details : Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
452 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
454 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
455 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
458 --------------------------------------------------------------------------------
460 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
462 * Support for kernels:
463 2.6.5-7.283 (SLES 9),
464 2.6.9-55.EL (RHEL 4),
465 2.6.16.46-0.14 (SLES 10),
466 2.6.18.8 vanilla (kernel.org)
467 * Client support for unpatched kernels:
468 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
469 2.6.16 - 2.6.22 vanilla (kernel.org)
470 * Due to recently discovered recovery problems, we do not recommend
471 using patchless RHEL 4 clients with this or any earlier release.
472 * Recommended e2fsprogs version: 1.39.cfs8
473 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
474 * Starting with this release, the ldiskfs backing filesystem required
475 by Lustre is now in its own package, lustre-ldiskfs. This package
476 should be installed. It is versioned separately from Lustre and
477 may be released separately in future.
481 Description: block reactivating mgc import until all deactivates complete
482 Details : Fix race when failing back MDT/MGS to itself (testing)
484 Severity : enhancement
486 Description: add optional extra BUILD_VERSION info
487 Details : add a new environment variable (namely LUSTRE_VERS) which allows
488 to override the lustre version.
491 Frequency : 2.6.18 servers only
493 Description: ll_kern_mount() doesn't release the module reference
494 Details : The ldiskfs module reference count never drops down to 0
495 because ll_kern_mount() doesn't release the module reference.
500 Description: server LBUG when using old ost_num_threads parameter
501 Details : Accept the old ost_num_threads parameter but warn that it
502 is deprecated, and fix an off-by-one error that caused an LBUG.
507 Description: Transient SCSI error results in persistent IO issue
508 Details : iobuf->dr_error is not reinitialized to 0 between two
512 Frequency : sometimes when underlying device returns I/O errors
514 Description: OSTs not going read-only during write failures
515 Details : OSTs are not remounted read-only when the journal commit threads
516 get I/O errors because fsfilt_ext3 calls journal_start/stop()
517 instead of the ext3 wrappers.
521 Description: poor connect scaling with increasing client count
522 Details : Don't run filter_grant_sanity_check for more than 100 exports
523 to improve scaling for large numbers of clients.
526 Frequency : SLES10 only
528 Description: sanity-quota.sh quotacheck failed: rc = -22
529 Details : Quotas cannot be enabled on SLES10.
532 Frequency : liblustre clients only
534 Description: getdirentries does not give error when run on compute nodes
535 Details : getdirentries does not fail when the size specified as an argument
536 is too small to contain at least one entry
538 Severity : enhancement
540 Description: Add LNET router traceability for debug purposes
541 Details : If a checksum failure occurs with a router as part of the
542 IO path, the NID of the last router that forwarded the bulk data
543 is printed so it can be identified.
548 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
549 Details : Due to a race condition, liblustre clients were occasionally
552 Severity : enhancement
554 Description: lfs setstripe use optional parameters instead of postional
557 Severity : enhancement
559 Description: Nanosecond timestamp support for ldiskfs
560 Details : The on-disk ldiskfs filesystem has added support for nanosecond
561 resolution timestamps. There is not yet support for this at
562 the Lustre filesystem level.
565 Frequency : during server recovery
567 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
568 Details : request with rq_no_resend flag not awake l_wait_event if they get a
572 Frequency : nfs export on patchless client
574 Description: connectathon hang when test nfs export over patchless client
575 Details : Disconnected dentry cannot be found with lookup, so we do not need
576 to unhash it or make it invalid
579 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
580 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
581 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
586 Description: open req refcounting wrong on reconnect
587 Details : If reconnect happened between getting open reply from server and
588 call to mdc_set_replay_data in ll_file_open, we will schedule
589 replay for unreferenced request that we are about to free.
590 Subsequent close will crash in variety of ways.
591 Check that request is still eligible for replay in
592 mdc_set_replay_data().
597 Description: disable writes to filesystem when reading health_check file
598 Details : the default for reading the health_check proc file has changed
599 to NOT do a journal transaction and write to disk, because this
600 can cause reads of the /proc file to hang and block HA state
601 checking on a healthy but otherwise heavily loaded system. It
602 is possible to return to the previous behaviour during configure
603 with --enable-health-write.
605 Severity : enhancement
607 Description: 64-bit inode version
608 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
609 to the inode. This will be required for version-based recovery.
614 Description: MDS fails to start if a duplicate client export is detected
615 Details : in some rare cases it was possible for a client to connect to
616 an MDS multiple times. Upon recovery the MDS would detect this
617 and fail during startup. Handle this more gracefully.
619 Severity : enhancement
621 Description: Add -o localflock option to simulate old noflock
623 Details : This will achieve local-only flock/fcntl locks
629 Description: log_commit_thread vs filter_destroy race leads to crash
630 Details : Take import reference before releasing llog record semaphore
635 Description: Wrong request locking in request set processing
636 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
637 imp_delayed_list, in this place should be used imp_lock.
640 Frequency : when reconnection
642 Description: Grant Leak when osc reconnect to OST
643 Details : When osc reconnect ost, OST(filter) should check whether it
644 should grant more space to client by comparing fed_grant and
645 cl_avail_grant, and return the granted space to client instead
646 of "new granted" space, because client will call osc_init_grant
647 to update the client grant space info.
650 Frequency : when client reconnect to OST
652 Description: Grant Leak when osc do resend and replay bulk write
653 Details : When osc reconnect to OST, OST(filter)should clear grant info of
654 bulk write request, because the grant info while be sync between
655 OSC and OST when reconnect, and we should ignore the grant info
656 these of resend/replay write req.
661 Description: Grant space more than avaiable left space sometimes.
662 Details : When then OST is about to be full, if two bulk writing from
663 different clients came to OST. Accord the avaliable space of the
664 OST, the first req should be permitted, and the second one
665 should be denied by ENOSPC. But if the seconde arrived before
666 the first one is commited. The OST might wrongly permit second
667 writing, which will cause grant space > avaiable space.
670 Frequency : when client is evicted
672 Description: Grant might be wrongly erased when osc is evicted by OST
673 Details : when the import is evicted by server, it will fork another
674 thread ptlrpc_invalidate_import_thread to invalidate the
675 import, where the grant will be set to 0. While the original
676 thread will update the grant it got when connecting. So if
677 the former happened latter, the grant will be wrongly errased
678 because of this race.
683 Description: Checking Stale with correct fid
684 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
685 to check whether it is stale, because sometimes, we want the
686 enqueue happened anyway, and op_data.fid2 will not be initialized.
688 Severity : enhancement
690 Description: update patchless client
691 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
694 Frequency : only with 2.4 kernel
696 Description: random memory corruption
697 Details : size of struct ll_inode_info is to big for union inode.u and this
698 can be cause of random memory corruption.
703 Description: Memory leak in recovery
704 Details : Lov_mds_md was not free in an error handler in mds_create_object.
705 It should also check obd_fail before fsfilt_start, otherwise if
706 fsfilt_start return -EROFS,(failover mds during mds recovery).
707 then the req will return with repmsg->transno = 0 and rc = EROFS.
708 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
709 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
710 be freed no matter whether fsfilt_commit success or not.
713 Frequency : only with huge count clients
715 Description: Prevents from taking the superblock lock in llap_from_page for
717 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
718 for llap_from_page call in ll_removepage prevents from taking the
719 superblock lock for a soon died page.
724 Description: Not check open intent error before release open handle
725 Details : in some rare cases, the open intent error is not checked before
726 release open handle, which may cause
727 ASSERTION(open_req->rq_transno != 0), because it tries to release
728 the failed open handle.
733 Description: Set cat log bitmap only after create log success.
734 Details : in some rare cases, the cat log bitmap is set too early. and it
735 should be set only after create log success.
739 Description: Accessing a block bevice can re-enable I/O when Lustre is
740 tearing down a device.
741 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
746 Description: service threads may hog cpus when there are a lot of requests
748 Details : Insert cond_resched to give other threads a chance to use some of
754 Description: the cat log was not initialized in recovery
755 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
756 unlink log on mds will not be initialized until mds post
757 recovery. And also in mds post recovery, the unlink log will
758 initialization will be done asynchronausly, so there will be race
759 between add unlink log and unlink log initialization.
763 Description: brw_stats were being printed incorrectly
764 Details : brw_stats were being printed as log2 but all of them were not
765 recorded as log2. Also remove some code duplication arising from
766 filter_tally_{read,write}.
770 Frequency : rare, only in recovery.
771 Description: ASSERTION(req->rq_type != LI_POISON) failed
772 Details : imp_lock should be held while iterating over imp_sending_list for
773 prevent destroy request after get timeout in ptlrpc_queue_wait.
777 Description: replay-single.sh test 52 fails
778 Details : A lock's skiplist need to be cleanup when it being unlinked
779 from its resource list.
783 Description: Short directio read returns full requested size rather than
785 Details : Direct I/O operations should return actual amount of bytes
786 transferred rather than requested size.
788 Severity : enhancement
790 Description: metadata RPC reduction (e.g. for rm performance)
791 Details : decrease the amount of synchronous RPC between clients and servers
792 by canceling conflicing lock before the operation on the client side
793 and packing thier handles into the main operation RPC to server.
795 Severity : enhancement
797 Description: Async OSC create to avoid the blocking unnecessarily.
798 Details : If a OST has no remain object, system will block on the creating
799 when need to create a new object on this OST. Now, ways use
800 pre-created objects when available, instead of blocking on an
801 empty osc while others are not empty. If we must block, we block
802 for the shortest possible period of time.
804 --------------------------------------------------------------------------------
806 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
811 Frequency : on some architectures
813 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
814 Details : Uninitialized flags sometimes cause configuration commands to
818 Frequency : patchless clients only
820 Description: missing __iget() symbol export
821 Details : The __iget() symbol export is missing. To avoid the need for
822 this on patchless clients the deathrow inode reaper is turned
823 off, and we depend on the VM to clean up old inodes. This
824 dependency was during via the fix for bug 12181.
826 --------------------------------------------------------------------------------
828 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
830 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
831 INTEROPERATE with older versions automatically. In many cases a
832 special upgrade step is needed. Please read the
833 user documentation before upgrading any part of a live system.
834 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
835 version will not interoperate with 1.6 betas before beta5 (1.5.95).
836 * WARNING: Lustre configuration and startup changes are required with
837 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
843 Severity : enhancement
845 Description: MountConf
846 Details : Lustre configuration is now managed via mkfs and mount
847 commands instead of lmc and lconf. New obd types (MGS, MGC)
848 are added for dynamic configuration management. See
849 https://mail.clusterfs.com/wikis/lustre/MountConf for
852 Severity : enhancement
854 Description: dynamic OST addition
855 Details : OSTs can now be added to a live filesystem
857 Severity : enhancement
859 Description: startup order invariance
860 Details : MDTs and OSTs can be started in any order. Clients only
861 require the MDT to complete startup.
863 Severity : enhancement
865 Description: parallel, asynchronous orphan cleanup
866 Details : orphan cleanup is now performed in separate threads for each
867 OST, allowing parallel non-blocking operation.
869 Severity : enhancement
871 Description: optimized stripe assignment
872 Details : stripe assignments are now made based on ost space available,
873 ost previous usage, and OSS previous usage, in order to try
874 to optimize storage space and networking resources.
876 Severity : enhancement
878 Description: Permanently set tunables
879 Details : All writable /proc/fs/lustre tunables can now be permanently
880 set on a per-server basis, at mkfs time or on a live
883 Severity : enhancement
885 Description: Lustre message v2
886 Details : Add lustre message format v2.
888 Severity : enhancement
890 Description: client OST exclusion list
891 Details : Clients can be started with a list of OSTs that should be
892 declared "inactive" for known non-responsive OSTs.
896 Description: SPEC SFS validation failure on NFS v2 over lustre.
897 Details : Changes the blocksize for regular files to be 2x RPC size,
898 and not depend on stripe size.
900 Severity : enhancement
902 Description: Multiple MD RPCs in flight.
903 Details : Further unserialise some read-only MDS RPCs - learn about intents.
904 To avoid overly-overloading MDS, introduce a limit on number of
905 MDS RPCs in flight for a single client and add /proc controls
906 to adjust this limit.
908 Severity : enhancement
910 Description: client read/write statistics
911 Details : Add client read/write call usage stats for performance
912 analysis of user processes.
913 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
914 file access. extents_stats shows chunk size distribution.
915 extents_stats_per_process show chunk size distribution per
918 Severity : enhancement
920 Description: mds statistics
921 Details : Add detailed mds operations statistics in
922 /proc/fs/lustre/mds/*/stats.
926 Description: Failure of copying files with lustre special EAs.
927 Details : Client side always return success for setxattr call for lustre
928 special xattr (currently only "trusted.lov").
933 Description: Refcount LNET uuids
934 Details : The global LNET uuid list grew linearly with every startup;
935 refcount repeated list entries instead of always adding to
938 Severity : enhancement
940 Description: Dynamic service threads
941 Details : Within a small range, start extra service threads
942 automatically when the request queue builds up.
945 Frequency : mixed-endian client/server environments
947 Description: mixed-endian crashes
948 Details : The new msg_v2 system had some failures in mixed-endian
951 Severity : enhancement
953 Description: Easy OST removal
954 Details : OSTs can be permanently deactivated with e.g. 'lctl
955 conf_param lustre-OST0001.osc.active=0'
957 Severity : enhancement
959 Description: MGS proc entries
960 Details : Added basic proc entries for the MGS showing what filesystems
963 Severity : enhancement
965 Description: provide MGS failover
966 Details : Added config lock reacquisition after MGS server failover.
968 Severity : enhancement
970 Description: add Linux 2.4 support
971 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
975 Description: a large application tries to do I/O to the same resource and dies
977 Details : Check the req->rq_arrival time after the call to
978 ost_brw_lock_get(), but before we do anything about
979 processing it & sending the BULK transfer request. This
980 should help move old stale pending locks off the queue as
981 quickly as obd_timeout.
984 Frequency : when an incorrect nid is specified during startup
986 Description: ptlrpc connect to non-existant node causes kernel crash
987 Details : LNET can't be re-entered from an event callback, which
988 happened when we expire a message after the export has been
989 cleaned up. Instead, hand the zombie cleanup off to another
992 Severity : enhancement
994 Description: plain/inodebits lock performance improvement
995 Details : Grouping plain/inodebits in granted list by their request modes
996 and bits policy, thus improving the performance of search through
1000 Frequency : only if OST filesystem is corrupted
1002 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1003 Details : for a short time RPCs with bulk IO are in the replay list,
1004 but replay of bulk IOs is unimplemented. If the OST filesystem
1005 is corrupted due to disk cache incoherency and then replay is
1006 started it is possible to trip an assertion. Avoid putting
1007 committed RPCs into the replay list at all to avoid this issue.
1010 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1012 Description: System hang on startup
1013 Details : This bug allowed the liblustre (e.g. catamount) client to
1014 return to the app before handling all startup RPCs. This
1015 could leave the node unresponsive to lustre network traffic
1016 and manifested as a server ptllnd timeout.
1018 Severity : enhancement
1020 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1021 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1022 boolean to control whether to print peer debug info when a
1023 client's RPC times out.
1026 Frequency : only for kernels with patches from Lustre below 1.4.3
1028 Description: Remove old rdonly API
1029 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1032 Frequency : only for devices with external journals
1034 Description: Set external device read-only also
1035 Details : During a commanded failover stop, we set the disk device
1036 read-only while the server shuts down. We now also set any
1037 external journal device read-only at the same time.
1040 Frequency : when upgrading from 1.4 while trying to change parameters
1042 Description: The wrong (new) MDC name was used when setting parameters for
1043 upgraded MDT's. Also allows changing of OSC (and MDC)
1044 parameters if --writeconf is specified at tunefs upgrade time.
1047 Frequency : when setting specific ost indicies
1049 Description: QOS code breaks on skipped indicies
1050 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1051 created with --index need not be sequential.
1053 Severity : enhancement
1055 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1056 Details : The uninit_groups feature works in conjunction with the kernel
1057 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1058 up the pass1 processing of e2fsck. This is a read-only feature
1059 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1060 mount filesystems that have had this feature enabled.
1062 Severity : enhancement
1064 Description: Improve multi-block allocation algorithm to avoid fragmentation
1065 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1066 allocation locality and avoid filesystem fragmentation.
1068 ------------------------------------------------------------------------------
1070 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1072 * Support for kernels:
1073 2.4.21-47.0.1.EL (RHEL 3)
1074 2.6.5-7.283 (SLES 9)
1075 2.6.9-42.0.10.EL (RHEL 4)
1076 2.6.12.6 vanilla (kernel.org)
1077 2.6.16.27-0.9 (SLES 10)
1078 * Recommended e2fsprogs version: 1.39.cfs5
1080 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1084 Frequency : occasional, depends on client load and configuration
1085 Bugzilla : 12181, 12203
1086 Description: data loss for recently-modified files
1088 Details : In some cases it is possible that recently written or created
1089 files may not be written to disk in a timely manner (this should
1090 normally be within 30s unless client IO load is very high).
1091 The problem appears as zero-length files or files that are a
1092 multiple of 1MB in size after a client crash or client eviction
1093 that are missing data at the end of the file.
1095 This problem is more likely to be hit on clients where files are
1096 repeatedly created and unlinked in the same directory, clients
1097 have a large amount of RAM, have many CPUs, the filesystem has
1098 many OSTs, the clients are rebooted frequently, and/or the files
1099 are not accessed by other nodes after being written.
1101 The presence of the problem can be detected by looking at
1102 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1103 smaller than the second (nr_unused) then dirty files will not
1104 be flushed automatically to disk. "sync; sleep 10" should be
1105 run several times on the node before unmounting it to update
1106 Lustre (this is also safe to run on nodes without this problem).
1108 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1109 that can cause this same problem, so customers using that kernel
1110 also need to update the kernel in addition to Lustre. In order
1111 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1113 It is normal that files written just before a client crash (less
1114 than 30s) may not yet have been flushed to disk, even for local
1118 Frequency : frequent on thin XT3 nodes
1120 Description: UUID collision on thin XT3 Linux nodes
1121 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1122 randomly, since we relied on an insufficiently-seeded PRNG.
1127 Description: OSS hangs after "All ost request buffers busy"
1128 Details : A deadlock between quota and journal operations caused OSS
1129 hangs after printing "All ost request buffers busy."
1132 Frequency : always on liblustre builds
1134 Description: Cleanup compiler warnings on liblustre
1137 Frequency : always on liblustre builds on XT3
1139 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1143 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1144 > 32000 subdirectories
1145 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1146 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1147 subdirectory count crosses 32000. This will aid e2fsck to
1148 correctly handle more than 32000 subdirectories.
1151 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1153 Description: System hang on startup
1154 Details : This bug allowed the liblustre (e.g. catamount) client to
1155 return to the app before handling all startup RPCs. This
1156 could leave the node unresponsive to lustre network traffic
1157 and manifested as a server ptllnd timeout.
1159 Severity : enhancement
1161 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1162 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1163 boolean to control whether to print peer debug info when a
1164 client's RPC times out.
1169 Description: make O_SYNC working on 2.6 kernels
1170 Details : 2.6 kernels use different method for mark pages for write,
1171 so need add a code to lustre for O_SYNC work.
1176 Description: Failure to close file and release space on NFS
1177 Details : Put inode details into lock acquired in ll_intent_file_open.
1178 Use mdc_intent_lock in ll_intent_open to properly
1179 detect all kind of errors unhandled by mdc_enqueue.
1184 Description: proc file read during shutdown sometimes raced obd removal,
1186 Details : Add lock to prevent obd access after proc file removal.
1189 Frequency : Only for files larger than 4GB on 32-bit clients.
1191 Description: improperly doing page alignment of locks
1192 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1193 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1198 Description: RPCs being resent when they shouldn't be
1199 Details : Some RPCs that should not be resent are being resent. This
1200 can cause inconsistencies in the RPC state machine. Do not
1201 resend such requests.
1204 Frequency : rare, only with NFS export
1206 Description: Crash on NFS re-export node
1207 Details : under very unusual load conditions an assertion is hit in
1208 ll_intent_file_open()
1211 Frequency : only if OST filesystem is corrupted
1213 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1214 Details : for a short time RPCs with bulk IO are in the replay list,
1215 but replay of bulk IOs is unimplemented. If the OST filesystem
1216 is corrupted due to disk cache incoherency and then replay is
1217 started it is possible to trip an assertion. Avoid putting
1218 committed RPCs into the replay list at all to avoid this issue.
1223 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1224 Details : Large single O_DIRECT read and write calls can fail to allocate
1225 a sufficiently large buffer to process the request. In case of
1226 allocation failure the allocation is retried with a smaller
1227 buffer and broken into smaller requests.
1229 Severity : enhancement
1231 Description: Add -o localflock option to simulate old noflock behaviour.
1232 Details : This will achieve local-only flock/fcntl locks coherentness.
1237 Description: versioning check is incomplete
1238 Details : Checking the version difference of client vs. server, report
1239 error if the gap is too big.
1244 Description: add support PG_writeback bit
1245 Details : add support for PG_writeback bit for Lustre, for more carefull
1246 work with page cache in 2.6 kernel. This also fix some deadlocks
1247 and remove hack for work O_SYNC with 2.6 kernel.
1249 Severity : enhancement
1251 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1252 Details : The uninit_groups feature works in conjunction with the kernel
1253 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1254 up the pass1 processing of e2fsck. This is a read-only feature
1255 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1256 mount filesystems that have had this feature enabled.
1258 Severity : enhancement
1260 Description: Improve multi-block allocation algorithm to avoid fragmentation
1261 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1262 allocation locality and avoid filesystem fragmentation.
1264 ------------------------------------------------------------------------------
1266 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1268 * Support for kernels:
1269 2.6.9-42.0.3.EL (RHEL 4)
1270 2.6.5-7.276 (SLES 9)
1271 2.4.21-47.0.1.EL (RHEL 3)
1272 2.6.12.6 vanilla (kernel.org)
1273 2.6.16.21-0.8 (SLES10)
1274 * Recommended e2fsprogs version: 1.39.cfs2-0
1276 * The backwards-compatible /proc/sys/portals symlink has been removed
1277 in this release. Before upgrading, please ensure that you change
1278 any configuration scripts or /etc/sysctl.conf files that access
1279 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1280 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1281 in advance of the upgrade on any system running Lustre 1.4.6 or
1282 newer, since /proc/sys/lnet was added in that version.
1283 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1287 Frequency : only when quota is used
1289 Description: avoid scanning export list for quota master
1290 Details : Change the algorithms to avoid scanning export list in order
1291 to improve the efficiency.
1294 Frequency : MDS failover only, very rarely
1296 Description: "went back in time" messages on mds failover
1297 Details : The greatest transno may be lost when the current operation
1298 finishes with an error (transno==0) and the client's last_rcvd
1299 record is over-written. Save the greatest transno in the
1300 mds_last_transno for this case.
1303 Frequency : always for specific kernels and striping counts
1305 Description: client may get "Matching packet too big" without ACL support
1306 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1307 when trying to access files in certain configurations. The
1308 clients should in fact be denied when mounting because they do
1309 not understand ACLs.
1312 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1314 Description: many clients connecting with IO in progress causes connect timeouts
1315 Details : Avoid synchronous journal commits to avoid delays caused by many
1316 clients connecting/disconnecting when bulk IO is in progress.
1317 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1318 OST_IO_PORTAL to avoid delays behind potentially many pending
1322 Frequency : occasionally with multiple writers to a single file
1324 Description: shared writes to file may result in wrong size reported by stat()
1325 Details : Allow growing of kms when extent lock is cancelled
1328 Frequency : always with random mmap IO to multi-striped file
1330 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1331 Details : Only if the hole is at the end of OST object so that kms is too
1332 small. Fix is to increase kms accordingly in ll_nopage.
1335 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1337 Description: writes to a missing object would leak memory on the OST
1338 Details : If there is an inconsistency between the MDS and OST filesystems,
1339 such that the MDS references an object that doesn't exist, writes
1340 to that object will leak memory due to incorrect cleanup in the
1341 error handling path, eventually running out of memory on the OST.
1346 Description: Creating too long symlink causes lustre errors
1347 Details : Check symlink and name lengths before sending requests to MDS.
1350 Frequency : only if flock is enabled (not on by default)
1352 Description: posix locks not released on fd closure on 2.6.9+
1353 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1354 kernels, this caused such locks not to be released on fd close and
1355 then assertions on fs unmount about still used locks.
1358 Frequency : MDS failover only, very rarely
1360 Description: clients may get ASSERTION(granted_lock != NULL)
1361 Details : When request was taking a long time, and a client was resending
1362 a getattr by name lock request. The were multiple lock requests
1363 with the same client lock handle and
1364 mds_getattr_name->fixup_handle_for_resent_request found one of the
1365 lock handles but later failed with ASSERTION(granted_lock != NULL).
1370 Description: handle->h_buffer_credits > 0, assertion failure
1371 Details : h_buffer_credits is zero after truncate, causing assertion
1372 failure. This patch extends the transaction or creates a new
1376 Frequency : NFS re-export or patchless client
1377 Bugzilla : 11179, 10796
1378 Description: Crash on NFS re-export node (__d_move)
1379 Details : We do not want to hash the dentry if we don't have a lock.
1380 But if this dentry is later used in d_move, we'd hit uninitialised
1381 list head d_hash, so we just do this to init d_hash field but
1382 leave dentry unhashed.
1385 Frequency : NFS re-export or patchless client
1387 Description: NFS exports has problem with symbolic link
1388 Details : lustre client didn't properly install dentry when re-exported
1389 to NFS or running patchless client.
1392 Frequency : NFS re-export or patchless client
1394 Description: Various nfs/patchless fixes.
1395 Details : fixes reuse disconected alias for lookup process - this fixes
1396 warning "find_exported_dentry: npd != pd",
1397 fix permission error with open files at nfs.
1398 fix apply umask when do revalidate.
1401 Frequency : occasional
1403 Description: Crash on NFS re-export node
1404 Details : calling clear_page() on the wrong pointer triggered oops in
1405 generic_mapping_read().
1408 Frequency : rarely, using O_DIRECT IO
1410 Description: unaligned directio crashes client with LASSERT
1411 Details : check for unaligned buffers before trying any requests.
1414 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1416 Description: stale data returned from RAID cache
1417 Details : If only a small amount of IO is done to the RAID device before
1418 reading it again it is possible to get stale data from the RAID
1419 cache instead of reading it from disk.
1422 Frequency : always for sles10 kernel
1424 Description: sles10 support
1425 Details : ll_follow_link: compile fixes and using of nd_set_link
1426 under newer kernels.
1429 Frequency : depends on arch, kernel and compiler version, always on sles10
1432 Description: recursive or deep enough symlinks cause stack overflow
1433 Details : getting rid of large stack-allocated variable in
1437 Frequency : depends on hardware
1439 Description: lustre write performance loss in the SLES10 kernel
1440 Details : the performance loss is caused by using of write barriers in the
1441 ext3 code. The SLES10 kernel turns barrier support on by
1442 default. The fix is to undo that change for ldiskfs.
1444 ------------------------------------------------------------------------------
1446 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1448 * Support for kernels:
1449 2.6.9-42.0.3EL (RHEL 4)
1450 2.6.5-7.276 (SLES 9)
1451 2.4.21-47.0.1.EL (RHEL 3)
1452 2.6.12.6 vanilla (kernel.org)
1456 Frequency : quota enabled and large files being deleted
1458 Description: releasing more than 4GB of quota at once hangs OST
1459 Details : If a user deletes more than 4GB of files on a single OST it
1460 will cause the OST to spin in an infinite loop. Release
1461 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1466 Description: statfs data retrieved from /proc may be stale or zero
1467 Details : When reading per-device statfs data from /proc, in the
1468 {kbytes,files}_{total,free,avail} files, it may appear
1469 as zero or be out of date.
1472 Frequency : systems with MD RAID1 external journal devices
1474 Description: lconf's call to blkid is confused by RAID1 journal devices
1475 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1476 of returning all block devices that match the journal UUID.
1479 Frequency : always, for aggregate stripe size over 4GB
1481 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1482 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1483 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1486 Frequency : always if "lfs find" used on a local file/directory
1488 Description: "lfs find" segfaults if used on a local file/directory
1489 Details : The case where a directory component was not specified wasn't
1490 handled correctly. Handle this properly.
1493 Frequency : always on ppc64
1495 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1496 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1497 perform endianness conversion.
1500 Frequency : rarely (truncate to non-zero file size after write under load)
1501 Bugzilla : 10730, 10687
1502 Description: Files padded with zeros to next 4K multiple
1503 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1504 it is possible that files that are truncated to a non-zero size
1505 immediately after being written are filled with zero bytes beyond
1506 the truncated size. No file data is lost.
1508 Severity : enhancement
1510 Description: Allow recovery/failover for liblustre clients.
1511 Details : liblustre clients were unaware of failover configurations until
1514 Severity : enhancement
1516 Description: user file locks should fail when not mounting with flock option
1517 Details : Set up an error-returning stub in ll_file_operations.lock field
1518 to prevent incorrect behaviour when client is mounted without
1519 flock option. Also, set up properly f_op->flock field for
1523 Frequency : always on ia64
1525 Description: "lfs df" loops on printing out MDS statfs information
1526 Details : The obd_ioctl_data was not initialized and in some systems
1527 this caused a failure during the ioctl that did not return
1528 an error. Initialize the struct and return an error on failure.
1531 Frequency : SLES 9 only
1533 Description: Error of copying files with lustre special EAs as root
1534 Details : Client side always return success for setxattr call for lustre
1535 special xattr (currently only "trusted.lov").
1538 Frequency : rarely on clusters with both ia64+i386 clients
1540 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1541 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1542 clients are doing concurrent writes to the same file it is
1543 possible that the ia64 clients may LASSERT because the OST
1544 extent locks are not PAGE_SIZE aligned. Ensure that grown
1545 locks are always aligned on the request boundary.
1548 Frequency : specific use, occasional
1550 Description: Overwriting in use executable truncates on-disk binary image
1551 Details : If one node attempts to overwrite an executable in use by
1552 another node, we now correctly return ETXTBSY instead of
1553 truncating the file.
1555 Severity : enhancement
1557 Description: Async OSC create to avoid the blocking unnecessarily.
1558 Details : If a OST has no remain object, system will block on the creating
1559 when need to create a new object on this OST. Now, ways use
1560 pre-created objects when available, instead of blocking on an
1561 empty osc while others are not empty. If we must block, we block
1562 for the shortest possible period of time.
1567 Description: chmod on Lustre root is propagated to other clients
1568 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1569 invalid by the follow_mount time.
1572 Frequency : liblustre clients only
1574 Description: Race in 'instant cancel' lock handling could lead to such locks
1575 never to be granted in case of SMP MDS
1576 Details : Do not destroy not yet granted but cbpending locks in
1580 Frequency : replay/resend of open
1582 Description: non null lock assetion failure in mds_intent_policy
1583 Details : Trying to replay/resend lockless open requests resulted in
1584 mds_open() returning 0 with no lock. Now it sets a flag if
1585 a lock is going to be returned.
1587 Severity : enhancement
1589 Description: Checksum enhancements
1590 Details : New checksum enhancements allow for resending RPCs that failed
1593 Severity : enhancement
1595 Description: Tunables on number of dirty pages in cacche
1596 Details : Allow to set limit on number of dirty pages cached.
1601 Description: client crash on unmount - lock still has references
1602 Details : In some error handling cases it was possible to leak a lock
1603 reference on a client while accessing a file. This was not
1604 harmful to the client during operation, but would cause the
1605 client to crash when the filesystem is unmounted.
1608 Frequency : specific case, rare
1610 Description: ETXTBSY on mds though file not in use by client
1611 Details : ETXTBSY is no longer incorrectly returned when attempting to
1612 chmod or chown a directory that the user previously tried to
1613 execute or a currently-executing binary.
1616 Frequency : extremely rare except on liblustre-based clients
1618 Description: Lustre space not freed when files are deleted
1619 Details : Clean up open-unlinked files after client eviction. Previously
1620 the unlink was skipped and the files remained as orphans.
1625 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1626 Details : In some cases it was possible to send a blocking callback to
1627 a client doing a glimpse, even though that client didn't get
1628 a lock granted. When the glimpse lock is cancelled on the OST
1629 the freed lock is left on the waiting list and corrupted the list.
1632 Frequency : all core dumps
1634 Description: Broke core dumps to lustre
1635 Details : Negative dentry may be unhashed if parent does not have UPDATE
1636 lock, but some callers, e.g. do_coredump, expect dentry to be
1637 hashed after successful create, hash it in ll_create_it.
1639 ------------------------------------------------------------------------------
1641 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1643 * Support for kernels:
1644 2.6.9-42.0.2.EL (RHEL 4)
1645 2.6.5-7.276 (SLES 9)
1646 2.4.21-40.EL (RHEL 3)
1647 2.6.12.6 vanilla (kernel.org)
1651 Frequency : always on RHEL 3
1653 Description: Number of open files grows over time
1654 Details : The number of open files grows over time, whether or not
1655 Lustre is started. This was due to a filp leak introduced
1656 by one of our kernel patches.
1658 ------------------------------------------------------------------------------
1660 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1662 * Support for kernels:
1663 2.6.9-42.EL (RHEL 4)
1664 2.6.5-7.276 (SLES 9)
1665 2.4.21-40.EL (RHEL 3)
1666 2.6.12.6 vanilla (kernel.org)
1671 Bugzilla : 5719, 9635, 9792, 9684
1672 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1673 Details : If a server is under heavy load and cannot reply to new
1674 connection requests before the client resends the (re)connect,
1675 the connection handling code can behave badly if two service
1676 threads are concurrently handing separate (re)connections from
1677 the same client. Add better locking to the connection handling
1678 code, and ensure that only a single connection will be processed
1679 for a given client UUID, even if the lock is dropped.
1681 Severity : enhancement
1683 Description: add TCP zero-copy support to kernel
1684 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1685 sends if the hardware supports scatter-gather and checksumming.
1686 This allows socklnd to do client-write and server-read more
1687 efficiently and reduce CPU utilization from skbuf copying.
1690 Frequency : only if NFS exporting from client
1692 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1693 Details : If setting attributes on a file created under NFS that had
1694 never been opened it would be possible to oops the client
1695 if the file had no objects.
1698 Frequency : always for liblustre
1700 Description: liblustre client does MDS+OSTs setattr RPC for each write
1701 Details : When doing a write from a liblustre client, the client
1702 incorrectly issued an RPC to the MDS and each OST the file was
1703 striped over in order to update the timestamps. When writing
1704 with small chunks and many clients this could overwhelm the MDS
1705 with RPCs. In all cases it would slow down the write because
1706 these RPCs are unnecessary.
1708 Severity : enhancement
1710 Description: allow number of MDS service threads to be changed at module load
1711 Details : It is now possible to change the number of MDS service threads
1712 running. Adding "options mds mds_num_threads={N}" to the MDS's
1713 /etc/modprobe.conf will set the number of threads for the next
1714 time Lustre is restarted (assuming the "mds" module is also
1715 reloaded at that time). The default number of threads will
1716 stay the same, 32 for most systems.
1721 Description: OST crash if filesystem is unformatted or corrupt
1722 Details : If an OST is started on a device that has never been formatted
1723 or if the filesystem is corrupt and cannot even mount then the
1724 error handling cleanup routines would dereference a NULL pointer.
1729 Description: NULL pointer deref in llap_from_page.
1730 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1731 incorrect) mapping if the page was truncated/reclaimed while it was
1732 searched for. Check for this condition and skip such pages when
1733 doing readahead. Introduce extra check to llap_from_page() to
1734 verify page->mapping->host is non-NULL (so page is not anonymous).
1737 Frequency : Sometimes when using sys_sendfile
1739 Description: "page not covered by a lock" warnings from ll_readpage
1740 Details : sendfile called ll_readpage without right page locks present.
1741 Now we introduced ll_file_sendfile that does necessary locking
1742 around call to generic_file_sendfile() much like we do in
1746 Frequency : with certain MDS communication failures at client mount time
1748 Description: NULL pointer deref after failed client mount
1749 Details : a client connection request may delayed by the network layer
1750 and not be sent until after the PTLRPC layer has timed out the
1751 request. If the client fails the mount immediately it will try
1752 to clean up before the network times out the request. Add a
1753 reference from the request import to the obd device and delay
1754 the cleanup until the network drops the request.
1757 Frequency : occasionally during client (re)connect
1759 Description: assertion failure during client (re)connect
1760 Details : processing a client connection request may be delayed by the
1761 client or server longer than the client connect timeout. This
1762 causes the client to resend the connection request. If the
1763 original connection request is replied in this interval, the
1764 client may trip an assertion failure in ptlrpc_connect_interpret()
1765 which thought it would be the only running connect process.
1768 Frequency : only with obd_echo servers and clients that are rebooted
1770 Description: kernel BUG accessing uninitialized data structure
1771 Details : When running an obd_echo server it did not start the ping_evictor
1772 thread, and when a client was evicted an uninitialized data
1773 structure was accessed. Start the ping_evictor in the RPC
1774 service startup instead of the OBD startup.
1776 Severity : enhancement
1777 Bugzilla : 10193 (patchless)
1778 Description: Remove dependency on various unexported kernel interfaces.
1779 Details : No longer need reparent_to_init, exit_mm, exit_files,
1780 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1783 Frequency : rare (only users of deprecated and unsupported LDAP config)
1785 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1786 Details : LDAP apparently contains 'lustreName' attributes instead of
1787 'name'. A simple remapping of the name is sufficient.
1790 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1792 Description: waiting_locks_callback trips kernel BUG if client is evicted
1793 Details : Running with the dump_on_timeout debug flag turned on makes
1794 it possible that the waiting_locks_callback() can try to dump
1795 the Lustre kernel debug logs from an interrupt handler. Defer
1796 this log dumping to the expired_lock_main() thread.
1798 Severity : enhancement
1800 Description: Support NFS exporting on 2.6 kernels.
1801 Details : Implement non-rawops metadata methods for NFS server to use without
1802 changing NFS server code.
1805 Frequency : very rare (synthetic metadata workload only)
1807 Description: two racing renames might cause an MDS thread to deadlock
1808 Details : Running the "racer" program may cause one MDS thread to rename
1809 a file from being the source of a rename to being the target of
1810 a rename at exactly the same time that another thread is doing
1811 so, and the second thread has already enqueued these locks after
1812 doing a lookup of the target and is trying to relock them in
1813 order. Ensure that we don't try to re-lock the same resource.
1816 Frequency : only very large systems with liblustre clients
1818 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1819 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1820 all OSTs in parallel. This allows the eviction of stale liblustre
1821 clients to proceed much faster than if they were done in series,
1822 and also offers similar improvements for other set_info RPCs.
1827 Description: excessive CPU usage during initial read phase on client
1828 Details : During the initial read phase on a client, it would agressively
1829 retry readahead on the file, consuming too much CPU and impacting
1830 performance (since 1.4.5.8). Improve the readahead algorithm
1831 to avoid this, and also improve some other common cases (read
1832 of small files in particular, where "small" is files smaller than
1833 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1838 Description: MDS crash when receiving packet with unknown intent.
1839 Details : Do not LBUG in unknown intent case, just return -EFAULT
1841 Severity : enhancement
1842 Bugzilla : 9293, 9385
1843 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1844 Details : Do not serialize getattr (non-intent version) and statfs.
1847 Frequency : occasional, when OST network is overloaded/intermittent
1849 Description: client evicted by OST after bulk IO timeout
1850 Details : If a client sends a bulk IO request (read or write) the OST
1851 may evict the client if it is unresposive to its data GET/PUT
1852 request. This is incorrect if the network is overloaded (takes
1853 too long to transfer the RPC data) or dropped the OST GET/PUT
1854 request. There is no need to evict the client at all, since
1855 the pinger and/or lock callbacks will handle this, and the
1856 client can restart the bulk request.
1859 Frequency : Always when mmapping file with no objects
1861 Description: client crashes when mmapping file with no objects
1862 Details : Check that we actually have objects in a file before doing any
1863 operations on objects in ll_vm_open, ll_vm_close and
1869 Description: Request leak when working with deleted CWD
1870 Details : Introduce advanced request refcount tracking for requests
1871 referenced from lustre intent.
1873 Severity : Enhancement
1875 Description: Cache open file handles on client.
1876 Details : MDS now will return special lock along with openhandle, if
1877 requested and client is allowed to hold openhandle, even if unused,
1878 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1879 closing files for every read/write openration.
1881 Severity : Enhancement
1883 Description: Cache open negative dentries on client when possible.
1884 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1885 negative dentries on lock revocation.
1890 Description: Remounting a client read-only wasn't possible with a zconf mount
1891 Details : It wasn't possible to remount a client read-only with llmount.
1893 Severity : enhancement
1894 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1895 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1896 MPICH is included to add Lustre-specific ADIO interfaces.
1897 This is based closely on the UFS ADIO layer and only differs
1898 in file creation, in order to allow the OST striping to be set.
1899 This is user-contributed code and not supported by CFS.
1904 Description: extended inode attributes (immutable, append-only) work improperly
1905 when 2.4 and 2.6 kernels are used on client/server or vice versa
1906 Details : Introduce kernel-independent values for these flags.
1908 Severity : enhancement
1911 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1912 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1913 granularity. Now, Fractional MB granularity is be supported,
1914 this is very useful for low memory system.
1916 Severity : enhancement
1918 Description: Getattr by fid
1919 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1920 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1921 have locks and avoids getattr rpc requests for it on every stat.
1924 Frequency : Always, for filesystems larger than 2TB
1926 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1927 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1928 filesystems larger than 2TB to be mounted with mballoc enabled.
1931 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1933 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1934 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1935 filesystems larger than 2TB is mounted then the kernel will
1936 silently corrupt the start of the filesystem. CONFIG_LBD is
1937 enabled for all CFS-supported kernels, but the possibility of
1938 this happening with a modified kernel config exists.
1940 Severity : enhancement
1942 Description: add client O_DIRECT support for 2.6 kernels
1943 Details : It is now possible to do O_DIRECT reads and writes to files
1944 in the Lustre client mountpoint on 2.6 kernel clients.
1946 Severity : enhancement
1948 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1949 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1950 parallel, not waiting for response from every OST before sending
1951 a rpc to the next OST.
1956 Description: setattr vs write race when updating file timestamps
1957 Details : Client processes that update a file timestamp into the past
1958 right after writing to the file (e.g. tar) it is possible that
1959 the updated file modification time can be reset to the current
1960 time due to a race between processing the setattr and write RPC.
1962 Severity : enhancement
1964 Description: Bring 'lfs find' closer in line with regular Linux find.
1965 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1966 -print0 options and obtains all the needed info through the lustre
1969 Severity : enhancement
1971 Description: support up to 1024 configured devices on one node
1972 Details : change obd_dev array from statically allocated to dynamically
1973 allocated structs as they are first used to reduce memory usage
1978 Description: Flush dirty partially truncated pages during truncate
1979 Details : Immediatelly flush partially truncated pages in filter_setattr,
1980 this way we completely avoid having any pages in page cache on OST
1981 and can retire ugly workarounds during writes to flush such pages.
1986 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1987 Details : protect inode from truncation within vfs_unlink() context
1988 just take a reference before calling vfs_unlink() and release it
1989 when parent's i_sem is free.
1994 Description: last_id value checked outside lock on OST caused LASSERT failure
1995 Details : If there were multiple MDS->OST object precreate requests in
1996 flight, it was possible that the OST's last object id was checked
1997 outside a lock and incorrectly tripped an assertion. Move checks
1998 inside locks, and discard old precreate requests.
2001 Frequency : always, if extents are used on OSTs
2003 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2004 Details : OSTs using the extents format would not zero the high 16 bits of
2005 the index physical block number. This is not a problem for any
2006 OST filesystems smaller than 16TB, and no kernels support ext3
2007 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2008 new/modified files) and can be fixed for existing filesystems
2009 with e2fsprogs-1.39-cfs1.
2014 Description: import connection selection may be incorrect if timer wraps
2015 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2016 import connections to be ignored if the 32-bit jiffies counter
2017 wraps. Use a 64-bit jiffies counter.
2020 Frequency : very large clusters immediately after boot
2022 Description: LNET request buffers exhausted under heavy short-term load
2023 Details : If a large number of client requests are generated on a service
2024 that has previously never seen so many requests it is possible
2025 that the request buffer growth cannot keep up with the spike in
2026 demand. Instead of dropping incoming requests, they are held in
2027 the LND until the RPC service can accept more requests.
2030 Frequency : Sometimes during replay
2032 Description: Assertion failure in ll_local_open after replay.
2033 Details : If replay happened on an open request reply before we were able
2034 to set replay handler, reply will become not swabbed tripping the
2035 assertion in ll_local_open. Now we set the handler right after
2036 recognising of open request
2039 Frequency : very rare
2041 Description: kernel reports "badness in vsnprintf"
2042 Details : Reading from the "recovery_status" /proc file in small chunks
2043 may cause a negative length in lprocfs_obd_rd_recovery_status()
2044 call to vsnprintf() (which is otherwise harmless). Exit early
2045 if there is no more space in the output buffer.
2047 Severity : enhancement
2049 Description: clear OBD RPC statistics by writing to them
2050 Details : It is now possible to clear the OBD RPC statistics by writing
2051 to the "stats" file.
2056 Description: Client mtime is not the same on different clients after utimes
2057 Details : In some cases, the client was using the utimes() syscall on
2058 a file cached on another node. The clients now validate the
2059 ctime from the MDS + OSTs to determine which one is right.
2064 Description: Inability to activate failout mode
2065 Details : lconf script incorrectly assumed that in pythong string's numeric
2066 value is used in comparisons.
2069 Frequency : always with multiple stripes per file
2071 Description: Inefficient object allocation for mutli-stripe files
2072 Details : When selecting which OSTs to stripe files over, for files with
2073 a stripe count that divides evenly into the number of OSTs,
2074 the MDS is always picking the same starting OST for each file.
2075 Return the OST selection heuristic to the original design.
2080 Description: mount failures may take full timeout to return an error
2081 Details : Under some heavy load conditions it is possible that a
2082 failed mount can wait for the full obd_timeout interval,
2083 possibly several minutes, before reporting an error.
2084 Instead return an error as soon as the status is known.
2086 Frequency : quota enabled and large files being deleted
2088 Description: releasing more than 4GB of quota at once hangs OST
2089 Details : If a user deletes more than 4GB of files on a single OST it
2090 will cause the OST to spin in an infinite loop. Release
2091 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2096 Description: statfs data retrieved from /proc may be stale or zero
2097 Details : When reading per-device statfs data from /proc, in the
2098 {kbytes,files}_{total,free,avail} files, it may appear
2099 as zero or be out of date.
2102 Frequency : systems with MD RAID1 external journal devices
2104 Description: lconf's call to blkid is confused by RAID1 journal devices
2105 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2106 of returning all block devices that match the journal UUID.
2109 Frequency : always, for aggregate stripe size over 4GB
2111 Description: assertion fails when trying to use 4GB stripe size
2112 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2113 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2116 Frequency : always on ppc64
2118 Description: the first write on an ext3 filesystem with mballoc got stuck
2119 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2120 perform endianness conversion.
2122 ------------------------------------------------------------------------------
2124 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2126 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2127 INTEROPERATE with older versions automatically. Please read the
2128 user documentation before upgrading any part of a live system.
2129 * WARNING: Lustre networking configuration changes are required with
2130 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2133 * Support for kernels:
2134 2.6.9-22.0.2.EL (RHEL 4)
2135 2.6.5-7.244 (SLES 9)
2136 2.6.12.6 vanilla (kernel.org)
2139 Severity : enhancement
2140 Bugzilla : 7981/8208
2141 Description: Introduced Lustre Networking (LNET)
2142 Details : LNET is new networking infrastructure for Lustre, it includes
2143 a reorganized network configuration mode (see the user
2144 documentation for full details) as well as support for routing
2145 between different network fabrics. Lustre Networking Devices
2146 (LNDs) for the supported network fabrics have also been
2147 created for this new infrastructure.
2149 Severity : enhancement
2150 Description: Introduced Access control lists
2151 Details : clients can set ACLs on files and directories in order to have
2152 more fine-grained permissions than the standard Unix UGO+RWX.
2153 The MDS must be started with the "-o acl" mount option.
2155 Severity : enhancement
2156 Description: Introduced filesystem quotas
2157 Details : Administrators may now establish per-user quotas on the
2160 Severity : enhancement
2162 Description: Configuration change for the XT3
2163 The PTLLND is now used to run Lustre over Portals on the XT3
2164 The configure option(s) --with-cray-portals are no longer used.
2165 Rather --with-portals=<path-to-portals-includes> is used to
2166 enable building on the XT3. In addition to enable XT3 specific
2167 features the option --enable-cray-xt3 must be used.
2172 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2173 Details : A race between allocating a new llcd and re-getting the llcd_lock
2174 allowed another thread to grab newly-allocated llcd.
2176 Severity : enhancement
2178 Description: 2.6 OST async journal commit and locking fix to improve performance
2179 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2180 2.6 kernels are now async as they already were in 2.4 kernels so
2181 that they can commit concurrently with the network bulk transfer.
2182 For block-allocated files the filter allocation semaphore is held
2183 to avoid filesystem fragmentation during allocation. BKL lock
2184 removed for 2.6 xattr operations where it is no longer needed.
2189 Description: lconf incorrectly determined whether two IP networks could talk
2190 Details : In some more complicated routing and multiple-network
2191 configurations, lconf will avoid trying to make a network
2192 connection to a disjoint part of the IP space. It was doing the
2193 math incorrectly for one set of cases.
2198 Description: Fix for potential infinite loop processing records in an llog.
2199 Details : If an llog record is corrupted/zeroed, it is possible to loop
2200 forever in llog_process(). Validate the llog record length
2201 and skip the remainder of the block on error.
2204 Frequency : occasional (liblustre only)
2206 Description: liblustre could not open files whose last component is a symlink
2207 Details : sysio_path_walk() would incorrectly pass the open intent to
2208 intermediate path components.
2211 Frequency : rare (liblustre only with non-standard tuning)
2212 Bugzilla : 7201 (7350)
2213 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2214 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2215 after it has been released, only do this for referenced locks
2217 Severity : enhancement
2219 Description: specifying an (invalid) directory default stripe_size of -1
2220 would reset the directory default striping
2221 Details : stripe_size -1 was used internally to signal directory stripe
2222 removal, now use "all default" to signal dir stripe removal
2223 as a directory striping of "all default" is not useful
2226 Frequency : common for large clusters running liblustre clients
2228 Description: doing an ls when liblustre clients are running is slow
2229 Details : sending a glimpse AST to a liblustre client waits for every AST
2230 to time out, as liblustre clients will not respond. Since they
2231 cannot cache data we refresh the OST lock LVB from disk instead.
2233 Severity : enhancement
2235 Description: doing an ls at the same time as file IO can be slow
2236 Details : enqueue and other "small" requests can be blocked behind many
2237 large IO requests. Create a new OST IO portal for non-IO
2238 requests so they can be processed faster.
2241 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2243 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2244 created file with mode 000
2245 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2246 allow an ftruncate() as a non-root user to a file with mode 000.
2247 Lustre now respects this flag to disable mode checks when
2248 truncating a file owned by the user
2251 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2254 Description: Revoking locks from clients that went dead or catatonic might take
2256 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2257 cancellation of such locks instant on servers without waiting for
2258 any reply from clients. Clients drops these locks when cancel
2259 notification from server is received without replying.
2262 Frequency : liblustre-only, when liblustre client dies or becomes busy
2264 Description: Doing ls on Linux clients can take a long time with active
2266 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2267 granting such locks to it in the first place if possible. Locks
2268 are taken by proxy on the OST during the read or write and
2269 dropped immediately afterward. Add connect flags handling, do
2270 not grant locks to liblustre clients for glimpse ASTs.
2272 Severity : enhancement
2274 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2275 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2276 sequential reads and behaves badly with applications doing random
2277 reads. Improve it by reading ahead at least read region, and
2278 avoiding excessive large RPC for small reads.
2280 Severity : enhancement
2282 Description: Creating more than 1000 files for a single job may cause a load
2283 imbalance on the OSTs if there are also a large number of OSTs.
2284 Details : qos_prep_create() uses an OST index reseed value that is an
2285 even multiple of the number of available OSTs so that if the
2286 reseed happens in the middle of the object allocation it will
2287 still utilize the OSTs as uniformly as possible.
2292 Description: OST or MDS may oops in ping_evictor_main()
2293 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2294 but doesn't restart at beginning of obd_exports_timed list
2297 Severity : enhancement
2299 Description: improve by-nid export eviction on the MDS and OST
2300 Details : allow multiple exports with the same NID to be evicted at one
2301 time without re-searching the exports list.
2304 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2306 Description: MDS may oops in groups_free()
2307 Details : in rare race conditions a newly allocated group_info struct is
2308 freed again, and this can be NULL. The 2.4 compatibility code
2309 for groups_free() checked for a NULL pointer, but 2.6 did not.
2312 Frequency : common for liblustre clients doing little filesystem IO
2313 Bugzilla : 9352, 7313
2314 Description: server may evict liblustre clients accessing contended locks
2315 Details : if a client is granted a lock or receives a completion AST
2316 with a blocking AST already set it would not reply to the AST
2317 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2320 Frequency : lfs setstripe, only systems with more than 160 OSTs
2322 Description: unable to set striping with a starting offset beyond OST 160
2323 Details : llapi_create_file() incorrectly limited the starting stripe
2324 index to the maximum single-file stripe count.
2327 Frequency : LDAP users only
2329 Description: lconf did not handle in-kernel recovery with LDAP properly
2330 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2332 Severity : enhancement
2334 Description: bind OST threads to NUMA nodes to improve performance
2335 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2336 node and do their allocations there to localize memory access
2338 Severity : enhancement
2340 Description: llmount can determine client NID directly from Myrinet (GM)
2341 Details : the client NID code from gmnalnid was moved directly into
2342 llmount, removing the need to use this or specifying the
2343 client NID explicitly when mounting GM clients with zeroconf
2346 Frequency : if client is started with down MDS
2348 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2349 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2350 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2351 ptlrpc_import_delayed_req() because we didn't check if the
2352 request was interrupted, and we also didn't break out of the
2353 event loop if there was no timeout
2358 Description: data loss during non-page-aligned writes to a single file from
2359 both multiple nodes and multiple threads on one node at same time
2360 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2361 inconsistency led to false short-reads, that were cached and later
2362 used by ->prepare_write() to fill in partially written page,
2363 leading to data loss.
2366 Frequency : always, if lconf --abort_recovery used
2368 Description: lconf --abort_recovery fails with 'Operation not supported'
2369 Details : lconf was attempting to abort recovery on the MDT device and not
2372 Severity : enhancement
2374 Description: remove cleanup logs
2375 Details : replace lconf-generated cleanup logs with lustre internal
2376 cleanup routines. Eliminates the need for client-cleanup and
2379 Severity : enhancement
2381 Description: add support for EAs (user and system) on lustre filesystems
2382 Details : it is now possible to store extended attributes in the Lustre
2383 client filesystem, and with the user_xattr mount option it
2384 is possible to allow users to store EAs on their files also
2386 Severity : enhancement
2388 Description: Add possibility (config option) to show minimal available OST free
2390 Details : When compiled with --enable-mindf configure option, statfs(2)
2391 (and so, df) will return least minimal free space available from
2392 all OSTs as amount of free space on FS, instead of summary of
2393 free spaces of all OSTs.
2395 Severity : enhancement
2397 Description: do not expand extent locks acquired on OST-side
2398 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2399 by server: they are not cached anyway.
2402 Frequency : when mmap is used/binaries executed from Lustre
2404 Description: Unmmap pages before throwing them away from read cache.
2405 Details : llap_shrink cache now attempts to unmap pages before discarding
2406 them (if unmapping failed - do not discard). SLES9 kernel has
2407 extra checks that trigger if this unmapping is not done first.
2412 Description: lconf didn't resolve symlinks before checking to see whether a
2413 given mountpoint was already in use
2416 Frequency : when migrating failover services
2417 Bugzilla : 6395, 9514
2418 Description: When migrating a subset of services from a node (e.g. failback
2419 from a failover service node) the remaining services would
2420 time out and evict clients.
2421 Details : lconf --force (implied by --failover) sets the global obd_timeout
2422 to 5 seconds in order to quickly disconnect, but this caused
2423 other RPCs to time out too quickly. Do not change the global
2424 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2426 Severity : enhancement
2427 Frequency : if MDS is started with down OST
2428 Bugzilla : 9439,5706
2429 Description: Allow startup/shutdown of an MDS without depending on the
2430 availability of the OSTs.
2431 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2432 Add appropriate locking and lov-osc refcounts for safe
2433 cleaning. Add osc abort_inflight calls in case the
2434 synchronize never started.
2437 Frequency : occasional (Cray XT3 only)
2439 Description: root not authorized to access files in CRAY_PORTALS environment
2440 Details : The client process capabilities were not honoured on the MDS in
2441 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2442 been accessed by an authorized user then root was able to access
2443 the file on the local client also. The root user capabilities
2444 are now allowed on the MDS, as this environment has secure UID.
2447 Frequency : occasional
2449 Description: ldiskfs "too long searching" message happens too often
2450 Details : A debugging message (otherwise harmless) prints too often on
2451 the OST console. This has been reduced to only happen when
2452 there are fragmentation problems on the filesystem.
2457 Description: Division by zero in statfs when all OSCs are inactive
2458 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2459 lov_statfs divides by value returned by lov_get_stripecnt().
2463 Bugzilla : 9489, 3273
2464 Description: First write from each client to each OST was only 4kB in size,
2465 to initialize client writeback cache, which caused sub-optimal
2466 RPCs and poor layout on disk for the first writen file.
2467 Details : Clients now request an initial cache grant at (re)connect time
2468 and so that they can start streaming writes to the cache right
2469 away and always do full-sized RPCs if there is enough data.
2470 If the OST is rebooted the client also re-establishes its grant
2471 so that client cached writes will be honoured under the grant.
2476 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2477 Details : Now I/O RPCs go to different portal number and (presumably) fast
2478 lock requests (and glimses) and other RPCs get their own service
2479 threads pool that should be able to service those RPCs
2482 Severity : enhancement
2484 Description: Ability to exchange lustre version between client and servers and
2485 issue warnings at client side if client is too old. Also for
2486 liblustre clients there is ability to refuse connection of too old
2488 Details : New 'version' field is added to connect data structure that is
2489 filled with version info. That info is later checked by server and
2493 Frequency : rare, liblustre only.
2494 Bugzilla : 9296, 9581
2495 Description: Two simultaneous writes from liblustre at offset within same page
2496 might proceed at the same time overwriting eachother with stale
2498 Details : I/O lock withing llu_file_prwv was released too early, before data
2499 actually was hitting the wire. Extended lock-holding time until
2500 server acknowledges receiving data.
2503 Frequency : extremely rare. Never observed in practice.
2505 Description: avoid generating lustre_handle cookie of 0.
2506 Details : class_handle_hash() generates handle cookies by incrementing
2507 global counter, and can hit 0 occasionaly (this is unlikely, but
2508 not impossible, because initial value of cookie counter is
2509 selected randonly). Value of 0 is used as a sentinel meaning
2510 "unassigned handle" --- avoid it. Also coalesce two critical
2511 sections in this function into one.
2513 Severity : enhancement
2515 Description: allow liblustre clients to delegate truncate locking to OST
2516 Details : To avoid overhead of locking, liblustre client instructs OST to
2517 take extent lock in ost_punch() on client's behalf. New connection
2518 flag is added to handle backward compatibility.
2520 Severity : enhancement
2521 Bugzilla : 4928, 7341, 9758
2522 Description: allow number of OST service threads to be specified
2523 Details : a module parameter allows the number of OST service threads
2524 to be specified via "options ost ost_num_threads={N}" in the
2525 OSS's /etc/modules.conf or /etc/modprobe.conf.
2529 Bugzilla : 6146, 9635, 9895
2530 Description: servers crash with bad pointer in target_handle_connect()
2531 Details : In rare cases when a client is reconnecting it was possible that
2532 the connection request was the last reference for that export.
2533 We would temporarily drop the export reference and get a new
2534 one, but this may have been the last reference and the export
2535 was just destroyed. Get new reference before dropping old one.
2537 Severity : enhancement
2538 Frequency : if client is started with failover MDS
2540 Description: Allow multiple MDS hostnames in the mount command
2541 Details : Try to read the configuration from all specified MDS
2542 hostnames during a client mount in case the "primary"
2545 Severity : enhancement
2547 Description: Stop sending data to evicted clients as soon as possible.
2548 Details : Check if the client we are about to send or are sending data to
2549 was evicted already. (Check is done every second of waiting,
2550 for which l_wait_event interface was extended to allow checking
2551 of exit condition at specified intervals).
2554 Frequency : rare, normally only when NFS exporting is done from client
2556 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2558 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2561 Frequency : always, if srand() is called before liblustre initialization
2563 Description: Liblustre uses system PRNG disturbing its usage by user application
2564 Details : Introduce internal to lustre fast and high-quality PRNG for
2565 lustre usage and make liblustre and some other places in generic
2566 lustre code to use it.
2568 Severity : enhancement
2569 Bugzilla : 9477, 9557, 9870
2570 Description: Verify that the MDS configuration logs are updated when xml is
2571 Details : Check if the .xml configuration logs are newer than the config
2572 logs stored on the MDS and report an error if this is the case.
2573 Request --write-conf, or allow starting with --old_conf.
2575 Severity : enhancement
2577 Description: Handle symlinks in the path when checking if Lustre is mounted.
2578 Details : Resolve intermediate symlinks when checking if a client has
2579 mounted a filesystem to avoid duplicate client mounts.
2584 Description: lconf can hit an error exception but still return success.
2585 Details : The lconf command catches the Command error exception at the top
2586 level script context and will exit with the associated exit
2587 status, but doesn't ensure that this exit status is non-zero.
2592 Description: failure of ptlrpc thread startup can cause oops
2593 Details : Starting a ptlrpc service thread can fail if there are a large
2594 number of threads or the server memory is very fragmented.
2595 Handle this without oopsing.
2598 Frequency : always, only if liblustre and non-default acceptor port was used
2600 Description: liblustre cannot connect to servers with non-default acceptor port
2601 Details : tcpnal_set_default_params() was not called and was therefore
2602 ignoring the environment varaible TCPNAL_PORT, as well as other
2603 TCPNAL_ environment variables
2608 Description: two objects could be created on the same OST for a single file
2609 Details : If an OST is down, in some cases it was possible to create two
2610 objects on a single OST for a single file. No problems other
2611 than potential performance impact and spurious error messages.
2615 Bugzilla : 5681, 9562
2616 Description: Client may oops in ll_unhash_aliases
2617 Details : Client dcache may become inconsistent in race condition.
2618 In some cases "getcwd" can fail if the current directory is
2624 Description: Inode refcounting problems in NFS export code
2625 Details : link_raw functions used to call d_instantiate without obtaining
2626 extra inode reference first.
2630 Bugzilla : 9942, 9903
2631 Description: Referencing freed requests leading to crash, memleaks with NFS.
2632 Details : We used to require that call to ll_revalidate_it was always
2633 followed by ll_lookup_it. Also with revalidate_special() it is
2634 possible to call ll_revalidate_it() twice for the same dentry
2635 even if first occurence returned success. This fix changes semantic
2636 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2637 reference on a request referred from the intent.
2638 ll_intent_release() then releases such a request.
2641 Frequency : rare, normally benchmark loads only
2643 Description: unlinked inodes were kept in memory on the client
2644 Details : If a client is repeatedly creating and unlinking files it
2645 can accumulate a lot of stale inodes in the inode slab cache.
2646 If there is no other client load running this can cause the
2647 client node to run out of memory. Instead flush old inodes
2648 from client cache that have the same inode number as a new inode.
2651 Frequency : SLES9 2.6.5 kernel and long filenames only
2652 Bugzilla : 9969, 10379
2653 Description: utime reports stale NFS file handle
2654 Details : SLES9 uses out-of-dentry names in some cases, which confused
2655 the lustre dentry revalidation. Change it to always use the
2659 Frequency : rare, unless heavy write-truncate concurrency is continuous
2660 Bugzilla : 4180, 6984, 7171, 9963, 9331
2661 Description: OST becomes very slow and/or deadlocked during object unlink
2662 Details : filter_destroy() was holding onto the parent directory lock
2663 while truncating+unlinking objects. For very large objects this
2664 may block other threads for a long time and slow overall OST
2665 responsiveness. It may also be possible to get a lock ordering
2666 deadlock in this case, or run out of journal credits because of
2667 the combined truncate+unlink. Solution is to do object truncate
2668 first in one transaction without parent lock, and then do the
2669 final unlink in a new transaction with the parent lock. This
2670 reduces the lock hold time dramatically.
2673 Frequency : rare, 2.4 kernels only
2675 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2676 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2677 before unmounting their filesystem in order to avoid deadlock.
2678 The kernel_locked() function in 2.4 kernels only checks whether
2679 the kernel lock is held, not whether it is this process that is
2680 holding it as 2.6 kernels do.
2685 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2686 Details : The client ptlrpc code may be trying to reconnect to a down
2687 server before a previous connection attempt has timed out.
2688 Increase the reconnect interval to be longer than the connection
2689 timeout interval to avoid sending duplicate connections to
2693 Frequency : echo_client brw_test command
2695 Description: fix echo_client to work with OST preallocated code
2696 Details : OST preallocation code (5137) didn't take echo_client IO path
2697 into account: echo_client calls filter methods outside of any
2698 OST thread and, hence, there is no per-thread preallocated
2699 pages and buffers to use. Solution: hijack pga pages for IO. As
2700 a byproduct, this avoids unnecessary data copying.
2704 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2705 Description: Client can oops in mdc_commit_close() after open replay
2706 Details : It was possible for the MDS to return an open request with no
2707 transaction number in mds_finish_transno() if the client was
2708 evicted, but without actually returning an error. Clients
2709 would later try to replay that open and may trip an assertion
2710 Simplify the client close codepath, and always return an error
2711 from the MDS in case the open is not successful.
2714 Frequency : rare, 2.6 OSTs only
2716 Description: OST may deadlock under high load on fragmented files
2717 Details : If there was a heavy load and highly-fragmented OST filesystems
2718 it was possible to have all the OST threads deadlock waiting on
2719 allocation of biovecs, because the biovecs were not released
2720 until the entire RPC IO was completed. Instead, release biovecs
2721 as soon as they are complete to ensure forward IO progress.
2723 Severity : enhancement
2725 Description: Support for specifying external journal device at mount
2726 Details : If an OST or MDS device is formatted with an external journal
2727 device, this device major/minor is stored in the ext3 superblock
2728 and may not be valid for failover. Allow detecting and
2729 specifying the external journal at mount time.
2734 Description: Mounting an MDS with pending unlinked files may cause oops
2735 Details : target_finish_recovery() calls mds_postrecov() which returned
2736 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2737 considers this an error and immediately begins cleaning up the
2738 lov, just after starting the mds_lov process
2740 Severity : enhancement
2742 Description: Implement 'lfs df' to report actual free space on per-OST basis
2743 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2744 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2745 usage of objects; '-h' to report in human readable format.
2747 ------------------------------------------------------------------------------
2749 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2756 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2757 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2758 at startup during orphan unlinking. Instead, skip buddy update
2759 before setup, as it will be regenerated after recovery is complete.
2764 Description: If an OST is inactive, its locks might reference stale inodes.
2765 Details : lov_change_cbdata() must iterate over all namespaces, even if
2766 they are inactive to clear inode references from the lock.
2768 Severity : enhancement
2769 Frequency : occasional, if non-standard max_dirty_mb used
2771 Description: Client will block write RPCs if not enough grant
2772 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2773 then the client will block writes while waiting for another RPC
2774 to complete instead of consuming its dirty limit. With change
2775 we get improved performance when max_dirty_mb is small.
2777 Severity : enhancement
2778 Bugzilla : 3389, 6253
2779 Description: Add support for supplementary groups on the MDS.
2780 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2781 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2782 lookups for user supplementary groups into a cache.
2786 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2787 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2788 for liblustre O_EXCL is enforced by MDS.
2792 Description: readdir never returns NULL in liblustre.
2793 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2794 dentry in struct dirent.
2798 Frequency : liblustre only, depends on application IO pattern
2799 Description: liblustre clients evicted if not contacting servers
2800 Details : Don't put liblustre clients into the ping_evictor list, so
2801 they will not be evicted by the pinger ever.
2803 Severity : enhancement
2805 Description: Add ability to evict clients by NID from MDS.
2806 Details : By echoing "nid:$NID" string into
2807 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2808 $NID would be instantly evicted from this MDS and from all active
2809 OSTs connected to it.
2813 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2814 Details : lookup_it_finish() used to query file size from OSTs that was not
2819 Description: service threads change working directory to that of init
2820 Details : Starting lustre service threads may pin the working directory
2821 of the parent thread, making that filesystem busy. Threads
2822 now change to the working directory of init to avoid this.
2826 Frequency : during shutdown only
2827 Description: shutdown with a failed MDS or OST can cause unmount to hang
2828 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2833 Frequency : 2.6 only
2834 Description: chmod/chown may include an extra supplementary group
2835 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2836 supplementary group and if none is specified this is used.
2839 Bugzilla : 5479 (6816)
2841 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2842 Details : If lookup is in progress on a file that is unlinked we might try
2843 to revalidate the inode and fail in revalidate after lookup is
2844 complete and ll_file_open() enqueues the open again but
2845 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2848 Frequency : always, if lconf --abort_recovery used
2850 Description: lconf --abort_recovery fails with 'Operation not supported'
2851 Details : lconf was attempting to abort recovery on the MDT device and not
2854 ------------------------------------------------------------------------------
2856 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2861 Frequency : rare (only unsupported configurations with a node running as an
2863 Bugzilla : 6514, 5137
2864 Description: Mounting a Lustre file system on a node running as an OST could
2866 Details : OSTs now preallocates memory needed to write out data at
2867 startup, instead of when needed, to avoid having to
2868 allocate memory in possibly low memory situations.
2869 Specifically, if the file system is mounted on on OST,
2870 memory pressure could force it to try to write out data,
2871 which it needed to allocate memory to do. Due to the low
2872 memory, it would be unable to do so and the node would
2873 become unresponsive.
2875 Severity : enhancement
2877 Description: Addition of lconf --service command line option
2878 Details : lconf now accepts a '--service <arg>' option, which is
2879 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2881 Severity : enhancement
2883 Description: Failover mode is now the default for OSTs.
2884 Details : By default, OSTs will now run in failover mode. To return to
2885 the old behaviour, add '--failout' to the lmc line for OSTs.
2887 Severity : enhancement
2889 Description: Health checks are now provided for MDS and OSTs
2890 Details : Additional detailed health check information on MSD and OSTs
2891 is now provided through the procfs health_check value.
2894 Frequency : occasional, depends on IO load
2896 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2897 after numerous create/delete cycles
2898 Details : The ext3 inode allocation policy would not allocate new inodes
2899 very well on the OSTs because there are no new directories
2900 being created. Instead we look for groups with free space if
2901 the parent directories are nearly full.
2906 Description: Network or server problems during mount may cause partially
2907 mounted clients instead of returning an error.
2908 Details : The config llog parsing code may overwrite the error return
2909 code during mount error handling, returning success instead
2915 Description: MDS can fail to allocate large reply buffers
2916 Details : After long uptimes the MDS can fail to allocate large reply
2917 buffers (e.g. zconf client mount config records) due to memory
2918 fragmentation or consumption by the buffer cache. Preallocate
2919 some large reply buffers so that these replies can be sent even
2920 under memory pressure.
2924 Frequency : rare (liblustre)
2925 Description: fsx running with liblustre complained that using truncate() to
2926 extend the file doesn't work. This patch corrects that issue.
2927 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2928 ATTR_SIZE and lsm use in llu_setattr_raw.
2932 Frequency : rare, only 2.6 kernels
2933 Description: Unusual file access patterns on the MDS may result in inode
2934 data being lost in very rare circumstances.
2935 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2936 code in the 2.6 kernel caused the inode and/or EA data not to
2937 be read from disk, causing single-file corruption.
2941 Frequency : rare, only 2.6 filesystems using extents
2942 Description: Heavy concurrent write and delete load may cause data corruption.
2943 Details : It was possible under high-load situations to have an extent
2944 metadata block in the block device cache from a just-unlinked
2945 file overwrite a newly-allocated data block. We now unmap any
2946 metadata buffers that alias just-allocated data blocks.
2950 Frequency : filesystems with default stripe_count larger than 77
2951 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2952 Details : lconf specifies an inode size of 4096 bytes when the default
2953 stripe_count is larger than 77. This conflicts with the default
2954 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2955 this case to avoid pinning too much memory for large EAs.
2957 ------------------------------------------------------------------------------
2959 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2964 Frequency : rare (extremely heavy IO load with hundreds of clients)
2966 Description: Client is evicted, gets IO error writing to file
2967 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2968 caused clients to be evicted for AST timeouts. The fixes in
2969 bug 5192 mean we no longer need to have such short AST timeouts
2970 so ldlm_timeout has been increased.
2973 Frequency : occasional during --force or --failover shutdown under load
2974 Bugzilla : 5949, 4834
2975 Description: Server oops/LBUG if stopped with --force or --failover under load
2976 Details : a collection of import/export refcount and cleanup ordering
2977 issues fixed for safer force cleanup
2980 Frequency : only filesystems larger than 120 OSTs
2981 Bugzilla : 5990, 6223
2982 Description: lfs getstripe would oops on a very large filesystem
2983 Details : lov_getconfig used kfree on vmalloc'd memory
2986 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2987 Bugzilla : 6242, 6243
2988 Description: reading from files that had been truncated to a non-zero size
2989 but never opened returned no data
2990 Details : ll_file_read() reads zeros from no-object files to EOF
2995 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2996 Details : There's little harm in aborting MDS/OSS recovery and letting it
2997 try again, so I removed the LASSERT and return an error instead.
2999 Severity : enhancement
3001 Description: New debugging infrastructure for tracking down data corruption
3002 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3003 (b) cover more of the client-side code path, and (c) try to narrow
3004 down where problems occurred
3008 Bugzilla : 3819, 4364, 4397, 6313
3009 Description: Racing close and eviction MDS could cause assertion in mds_close
3010 Details : It was possible to get multiple mfd references during close and
3011 client eviction, leading to one thread referencing a freed mfd.
3013 Severity: : enhancement
3014 Bugzilla : 3262, 6359
3015 Description: Attempts to reconnect to servers are now more aggressive.
3016 Details : This builds on the enhanced upcall-less recovery that was added
3017 in 1.4.2. When trying to reconnect to servers, clients will
3018 now try each server in the failover group every 10 seconds. By
3019 default, clients would previously try one server every 25 seconds.
3024 Description: After recovery, certain operations trigger a failed
3025 assertion on a client.
3026 Details : Failing over an mds, using lconf -d --failover, while a
3027 client was doing a readdir() call would cause the client to
3028 LBUG after recovery completed and the readdir() was resent.
3030 Severity : enhancement
3032 Description: Default groups are now added by lconf
3033 Details : You can now run lconf --group <servicename> without having to
3034 manually add groups with lmc.
3037 Frequency : occasional
3039 Description: Nodes with an elan id of 0 trigger a failed assertion
3042 Frequency : always when accessing e.g. tty/console device nodes
3044 Description: tty and some other devices nodes cannot be used on lustre
3045 Details : file's private_data field is used by device data and lustre
3046 values in there got lost. New field was added to struct file to
3047 store fs-specific private data.
3050 Frequency : when exporting Lustre via NFS
3052 Description: NFSD failed occasionally when looking up a path component
3053 Details : NFSD is looking up ".." which was broken in ext3 directories
3054 that had grown large enough to become hashed.
3057 Frequency : Clusters with multiple interfaces not on the same subnet
3059 Description: Nodes will repeatedly try to reconnect to an interface which it
3060 cannot reach and report an error to the log.
3061 Details : Extra peer list entries will be created by lconf with some peers
3062 unreachable. lconf now validates the peer before adding it.
3065 Frequency : Only if a default stripe is set on the filesystem root.
3067 Description: Setting a default stripe on the filesystem root prevented the
3068 filesystem from being remounted.
3069 Details : The client was sending extra request flags in the root getattr
3070 request and did not allocate a reply buffer for the dir EA.
3073 Frequency : occasional, higher if lots of files are accessed by one client
3074 Bugzilla : 6159, 6097
3075 Description: Client trips assertion regarding lsm mismatch/magic
3076 Details : While revalidating inodes the VFS looks up inodes with ifind()
3077 and in rare cases can find an inode that is being freed.
3078 The ll_test_inode() code will free the lsm during ifind()
3079 when it finds an existing inode and then the VFS later attaches
3080 this free lsm to a new inode.
3084 Bugzilla : 6422, 7030
3085 Description: MDS deadlock between mkdir and client eviction
3086 Details : Creating a new file via mkdir or mknod (starting a transaction
3087 and getting the ns lock) can deadlock with client eviction
3088 (gets ns lock and trying to finish a synchronous transaction).
3091 Frequency : occasional
3092 Description: While starting a server, the fsfilt_ext3 module could not be
3094 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3095 kernels. Previously, lconf would still use the ext3 name
3096 when trying to load modules. Now, it will correctly use
3097 ext3 on 2.4 and ldiskfs on 2.6.
3099 Severity : enhancement
3100 Description: The default stripe count has been changed to 1
3101 Details : The interpretation of the default stripe count (0, to lfs
3102 or lmc) has been changed to mean striping across a single
3103 OST, rather than all available. For general usage we have
3104 found a stripe count of 1 or 2 works best.
3106 Severity : enhancement
3107 Description: Add support for compiling against Cray portals.
3108 Details : Conditional compiling for some areas that are different
3112 Frequency : occasional
3113 Bugzilla : 6409, 6834
3114 Description: Creating files with an explicit stripe count may lead to
3115 a failed assertion on the MDS
3116 Details : If some OSTs are full or unavailable, creating files may
3117 trigger a failed assertion on the MDS. Now, Lustre will
3118 try to use other servers or return an error to the
3122 Frequency : occasional
3124 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3125 caused lock timeout during readahead (since 1.4.2).
3126 Details : Processes doing readahead might match a lock that hasn't been
3127 granted yet if there are overlapping and conflicting lock
3128 requests. The readahead process waits on ungranted lock
3129 (original lock is CBPENDING), while OST waits for that process
3130 to cancel CBPENDING read lock and eventually evicts client.
3132 Severity : enhancement
3134 Description: Initial enabling of flock support for clients
3135 Details : Implements fcntl advisory locking and file status functions.
3136 This feature is provided as an optional mount flag (default
3137 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3138 locking are implemented yet, and those that are are not guaranteed
3139 to be completely correct in production environments.
3140 mount -t lustre -o [flock|noflock] ...
3143 Frequency : occasional
3145 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3146 assertion in the ext3 JBD (journaling) layer.
3147 Details : The b_committed_data struct is protected by the big kernel lock
3148 in 2.4 kernels, serializing journal_commit_transaction() and
3149 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3150 access to this struct. In 2.6 kernels there is finer grained
3151 locking to improve SMP performance of the JBD layer.
3155 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3160 Description: merge and cleanup kernel patches.
3163 -----------------------------------------------------------------------------
3165 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3167 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3168 versions of Lustre. Please update all servers and clients to
3169 version 1.4.2 or later at the same time. You must also run
3170 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3171 to update the configuration logs.
3173 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3174 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3175 - swab llog records of type '0' so we get proper header size/idx (5861)
3176 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3177 - fix rename of one directory over another leaking an inode (5953)
3178 - avoid SetPageDirty on 2.6 (5981)
3179 - don't re-add just-being-destroyed locks to the waiting list (5653)
3180 - when creating new directories, inherit the parent's custom
3181 striping settings if present parent (3048)
3182 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3183 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3184 - don't hold client locks on temporary worklist from l_lru (5666)
3185 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3186 - automatically evict dead clients (5921)
3187 - Update file size properly in create+truncate+fstat case (6196)
3188 - Do not unhash mountpoint dentries, do not allow removal of
3190 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3191 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3192 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3193 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3194 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3195 - Correctly handle reads of files with no objects (6243)
3196 - lctl recover will also mark a device active if deactivate used (5933)
3198 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3199 - allow --write-conf on an MDS with different nettype than client (5619)
3200 - don't write config llogs to MDS for mounts not from that MDS (5617)
3201 - lconf should create multiple TCP connections from a client (5201)
3202 - init scripts are now turned off by default; run chkconfig --on
3203 lustre and chkconfig --on lustrefs to use them
3204 - upcalls are no longer needed for clients to recover to failover
3206 - add --abort-recovery option to lconf to abort recovery on device
3208 - add support for an arbitrary number of OSTs (3026)
3209 - Quota support protocol changes.
3210 - forward compatibility changes to wire structs (6007)
3211 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3212 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3213 - improved "lustre status" script
3214 - initialize blocksize for non-regular files (6062)
3215 - added --disable-server and --disable-client configure options (5782)
3216 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3217 - Vanilla 2.4.29 support
3218 - increase maximum number of obd devices to 520 (6242)
3219 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3220 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3221 SLES 9 2.6.5 kernels
3222 - limit stripes per file to 160 (the maximum EA size) (6093)
3224 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3227 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3228 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3229 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3230 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3231 - small ext3 extents cleanups and fixes (5733)
3232 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3233 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3234 - avoid needless client->OST connect, fix handle mismatch (5317)
3235 - fix DLM error path that led to out-of-sync client, long delays (5779)
3236 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3237 - fix several locking issues related to i_size (5492,5624,5654,5672)
3238 - don't move pending lock onto export if it is already evicted (5683)
3239 - fix kernel oops when creating .foo in unlinked directory (5548)
3240 - fix deadlock in obdfilter statistics vs. object create (5811)
3241 - use time_{before,after} to avoid timer jiffies wrap (5882)
3242 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3243 - Do not leak request if server was not able to process it (5154)
3244 - If mds_open unable to find parent dir, make that negative lookup(5154)
3245 - don't create new directories with extent-mapping (5909, 5936)
3247 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3248 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3249 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3250 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3251 - support for RHEL4 kernel on i686 (5773)
3252 - provide error messages when incompatible logs are encountered (5898)
3254 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3255 * version 1.4.0.10 (1.4.1 release candidate 1)
3257 - don't keep a lock reference when lock is not granted (4238)
3258 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3259 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3260 - drop import inflight refcount on signal_completed_replay error (5255)
3261 - unlock page after async write error during send (3677)
3262 - handle missing objects in filter_preprw_read properly (5265)
3263 - no transno return for symlink open, don't save no-trasno open (3440)
3264 - don't try to complete elan receive that already failed (4012)
3265 - free RPC server reply state on error (5406)
3266 - clean up thread from ptlrpc_start_thread() on error (5160)
3267 - readahead could read extra page into cache that wasn't ejected (5388)
3268 - prevent races in class_attach/setup/cleanup/detach (5260)
3269 - don't dereference de->d_inode after l_dput of de (5458)
3270 - use "int" for stripe value returned from lock_to_stripe (5544)
3271 - mballoc allocation and error-checking fixes in 2.6 (5504)
3272 - block device patches to fix I/O request sizes in 2.6 (5482)
3273 - look up hostnames for IB nals (5602)
3274 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3275 - don't start multiple acceptors for the same port (5277)
3276 - fix incorrect LASSERT in mds_getattr_name (5635)
3277 - export a proc file for general "ping" checking (5628)
3278 - fix "lfs check" to not block when the MDS is down (5628)
3280 - service request history (4965)
3281 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3282 - create an "evict_client" /proc entry on OSTs, like the MDS has
3283 - fix mount usage message, return errors per mount(8) (5168)
3284 - change grep [] to grep "[]" in tests so they work in more UMLs
3285 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3286 - remove ancient LOV_MAGIC_V0 EA support (5047)
3287 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3288 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3289 - allow readahead allocations to fail when low on memory (5383)
3290 - mmap locking landed again, after considerable improvement (2828)
3291 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3293 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3296 - send OST transaction number in read/write reply to free req (4966)
3297 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3298 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3299 - the watchdog thread now runs as interruptible (5246)
3300 - flock/lockf fixes (but it's still disabled, pending 5135)
3301 - don't use EXT3 constants in llite code (5094)
3302 - memory shortage at startup could cause assertion (5176)
3304 - reorganization of lov code
3305 - single portals codebase
3307 - add extents/mballoc support (5025)
3308 - direct I/O reads in the obdfilter (4048)
3309 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3311 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3314 - send OST transaction number in read/write reply to free req (4966)
3315 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3316 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3317 - the watchdog thread now runs as interruptible (5246)
3318 - handle missing objects in filter_preprw_read properly (5265)
3319 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3320 - drop import inflight refcount on signal_completed_replay error (5255)
3321 - unlock page after async write error during send (3677)
3322 - return original error code on reconstructed replies (3761)
3323 - no transno return for symlink open, don't save no-trasno open (3440)
3325 - add pid to ldlm debugging output (4922)
3326 - bump the watchdog timeouts -- we can't handle 30sec yet
3327 - extra debugging for orphan dentry/inode bug (5259)
3329 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3332 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3333 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3334 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3335 - fix NULL dereference in /proc/sys/portals/routes (4827)
3336 - allow failed mdc_close() operations to be interrupted (4561)
3337 - stop precreate on OST before MDS would time out on it (4778)
3338 - don't send partial-page writes before EOF from client (4410)
3339 - discard client grant for sub-page writes on large-page clients (4520)
3340 - don't free dentries not owned by NFS code, check generation (4806)
3341 - fix lsm leak if mds_create_objects() fails (4801)
3342 - limit debug_daemon file size, always print CERROR messages (4789)
3343 - use transno after validating reply (3892)
3344 - process timed out requests if import state changes (3754)
3345 - update mtime on OST during writes, return in glimpse (4829)
3346 - add mkfsoptions to LDAP (4679)
3347 - use ->max_readahead method instead of zapping global ra (5039)
3348 - don't interrupt __l_wait_event() during strace
3350 - add software watchdogs to catch hung threads quickly (4941)
3351 - make lustrefs init script start after nfs is mounted
3352 - fix CWARN/ERROR duplication (4930)
3353 - return async write errors to application if possible (2248)
3354 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3355 - print NAL number in %x format (4645)
3356 - update barely-supported suse-2.4.21-171 series (4842)
3357 - support for sles 9 %post scripts
3358 - support for building 2.6 kernel-source packages
3359 - support for sles km_* packages
3361 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3364 - ignore -ENOENT errors in osc_destroy (3639)
3365 - notify osc create thread that OSC is being cleaned up (4600)
3366 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3367 - reconstruct ost_handle() like mds_handle() (4657)
3368 - create a new thread to do import eviction to avoid deadlock (3969)
3369 - let lconf resolve symlinked-to devices (4629)
3370 - don't unlink "objects" from directory with default EA (4554)
3371 - hold socknal file ref over connect in case target is down (4394)
3372 - allow more than 32000 subdirectories in a single directory (3244)
3373 - fix blocks count for O_DIRECT writes (3751)
3374 - OST returns ENOSPC from object create when no space left (4539)
3375 - don't send truncate RPC if file size isn't changing (4410)
3376 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3377 - bind to privileged port in socknal and tcpnal (3689)
3379 - rate limit CERROR/CWARN console message to avoid overload (4519)
3380 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3381 - basic mmap support (3918)
3382 - kernel patch series update from b1_4 (4711)
3384 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3387 - avoid crash during MDS cleanup with OST shut down (2775)
3388 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3389 - don't use bad inodes on the MDS (3744)
3390 - dynamic object preallocation to improve recovery speed (4236)
3391 - don't hold spinlock over lock dumping or change debug flags (4401)
3392 - don't zero obd_dev when it is force cleaned (3651)
3393 - print grants to console if they go negative (4431)
3394 - "lctl deactivate" will stop automatic recovery attempts (3406)
3395 - look for existing locks in ldlm_handle_enqueue() (3764)
3396 - don't resolve lock handle twice in recovery avoiding race (4401)
3397 - revalidate should check working dir is a directory (4134)
3399 - don't always mark "slow" obdfilter messages as errors (4418)
3401 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3404 - don't close LustreDB during write_conf until it is done (3860)
3405 - fix typo in lconf for_each_profile (3821)
3406 - allow dumping logs from multiple threads at one time (3820)
3407 - don't allow multiple threads in OSC recovery (3812)
3408 - fix debug_size parameters (3864)
3409 - fix mds_postrecov to initialize import for llog ctxt (3121)
3410 - replace config semaphore with spinlock (3306)
3411 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3412 - don't allow enqueue to complete on a destroyed export (3822)
3413 - down write_lock before checking llog header bitmap (3825)
3414 - recover from lock replay timeout (3764)
3415 - up llog sem before sending rpc (3652)
3416 - reduce ns lock hold times when setting kms (3267)
3417 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3418 - fix NULL deref and obd_dev leak on setup error (3312)
3419 - replace some LBUG about llog ops with error handling (3841)
3420 - don't match INVALID dentries from d_lookup and spin (3784)
3421 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3422 - fix invalid assertion in ptlrpc_set_wait (3880)
3424 - add libwrap support for the TCP acceptor (3996)
3425 - add /proc/sys/portals/routes for non-root route listing (3994)
3426 - allow setting MDS UUID in .xml (2580)
3427 - print the stack of a process that LBUGs (4228)
3429 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3432 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3433 - make sure to unset replay flag from failed open requests (3440)
3434 - if default stripe count is 0, use OST count for inode size (3636)
3435 - update parent mtime/ctime on client for create/unlink (2611)
3436 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3437 - free recovery state on server during a forced cleanup (3571)
3438 - unregister_reply for resent reqs (3063)
3439 - loop back devices mounting and status check on 2.6 (3563)
3440 - fix resource-creation race that can provoke i_size == 0 (3513)
3441 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3442 - more debugging for page-accounting assertion (3746)
3443 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3444 - avoid deadlock after precreation failure (3758)
3445 - fix race and lock order deadlock in orphan handling (3450, 3750)
3446 - add validity checks when grabbing inodes from l_ast_data (3599)
3448 - add /proc/.../recovery_status to obdfilter (3428)
3449 - lightweight CDEBUG infrastructure, debug daemon (3668)
3450 - change default OSC RPC parameters to be better on small clusters
3451 - turn off OST read cache for files smaller than 32MB
3452 - install man pages and include them in rpms (3100)
3453 - add new init script for (un)mounting lustre filesystems (2593)
3454 - run chkconfig in %post for init scripts (3701)
3455 - drop scimac NAL (unmaintained)
3457 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3460 - clean kiobufs before and after use (3485)
3461 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3462 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3463 - change init script to fail more clearly if not run as root (1528)
3464 - allow clients to reconnect during replay (1742)
3465 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3466 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3467 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3468 - fix cond_resched (3554)
3469 - only evict unfinished clients after recovery (3515)
3470 - allow bulk resend, prevent data loss (3570)
3471 - dynamic ptlrpc request buffer allocation (2102)
3472 - don't allow unlinking open directory if it isn't empty (2904)
3473 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3474 - remove extraneous obd dereference causing LASSERT failure (3334)
3475 - don't use get_cycles() when creating temp. files on the mds (3156)
3476 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3477 - handle EEXIST for set-stripe, set proper directory name (3336)
3479 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3480 - fix ksocknal_fmb_callback() error messages (2918)
3482 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3485 - don't copy lvb into (possibly NULL) reply on error (2983)
3486 - don't deref dentry after dput, don't free lvb on error (2922)
3487 - use the kms to determine writeback rpc length (2947)
3488 - increment oti_logcookies when osc is inactive (2948)
3489 - update client's i_blocks count via lvb messages (2543)
3490 - handle intent open/close of special files properly (1557)
3491 - mount MDS with errors=remount-ro, like obdfilter (2009)
3492 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3493 - don't use cancelling-locks' kms values (2947)
3494 - use highest lock extent for kms, not last one (2925)
3495 - don't dereference ERR_PTR() dentry in error handling path (3107)
3496 - fix thread race in portals_debug_dumplog() (3122)
3497 - create lprocfs device entries at setup instead of at attach (1519)
3498 - common AST error handler, don't evict client on completion race (3145)
3499 - zero nameidata in detach_mnt in 2.6 (3118)
3500 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3501 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3502 - initialize RPC timeout timer earlier for 2.6 (3219)
3503 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3504 - print nal/nid for unknown nid (3258)
3505 - additional checks for oscc recovery before doing precreate (3284)
3506 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3507 - don't crash in mdc_close for bad permissions on open (3285)
3508 - zero i_rdev for non-device files (3147)
3509 - clear page->private before handing to FS, better assertion (3119)
3510 - tune the read pipeline (3236)
3511 - fix incorrect decref of invalidated dentry (2350)
3512 - provide read-ahead stats and refine rpc in flight stats (3328)
3513 - don't hold journal transaction open across create RPC (3313)
3514 - update atime on MDS at close time (3265)
3515 - close LDAP connection when recovering to avoid server load (3315)
3516 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3517 - don't leak open file on MDS after open resend (3325)
3518 - serialize filter_precreate and filter_destroy_precreated (3329)
3519 - loop device shouldn't call sync_dev() for nul device (3092)
3520 - clear page cache after eviction (2766)
3521 - resynchronize MDS->OST in background (2824)
3522 - refuse to mount the same filesystem twice on same mountpoint (3394)
3523 - allow llmount to create routes for mounting behind routers (3320)
3524 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3525 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3526 - fix rare problem with rename on htree directories (3417)
3528 - allow default OST striping configuration per directory (1414)
3529 - fix compilation for qswnal for 2.6 kernels (3125)
3530 - increase maximum number of MDS request buffers for large systems
3531 - change liblustreapi to be useful for external progs like lfsck (3098)
3532 - increase local configuration timeout for slow disks (3353)
3533 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3535 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3538 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3539 - don't overwrite extent policy data in reply if lock was blocked (2901)
3540 - drop filter export grants atomically with removal from device (2663)
3541 - del obd_self_export from work_list in class_disconnect_exports (2908)
3542 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3543 - swab reply message in mdc_close, other PPC fixes (2464)
3544 - fix destroying of named logs (2325)
3545 - overwrite old logs when running lconf --write_conf (2264)
3546 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3547 - fix race in target_handle_connect (2898)
3548 - mds_reint_create() should take same inode create lock (2926)
3549 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3550 - don't close files for self_export to avoid uninitialized obd (2936)
3551 - allow MDS with the same name as client node (2939)
3552 - hold dentry reference for closed log files for unlink (2325)
3553 - reserve space for all logs during transactions (2059)
3554 - don't evict page beyond end of stripe extent (2925)
3555 - don't oops on a deleted current working directory (2399)
3556 - handle hard links to targets without a parent properly (2517)
3557 - don't dereference NULL lock when racing during eviction (2867)
3558 - don't grow lock extents when lots of conflicting locks (2919)
3560 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3563 - account for cache space usage on clients to avoid data loss (974)
3564 - lfsck support in lustre kernel code (2349)
3565 - reduce journal credits needed for BRW writes (2370)
3566 - orphan handling to avoid losing space on client/server crashes
3567 - ptlrpcd can be blocked, stopping ALL progress (2477)
3568 - use lock value blocks to assist in proper KMS, faster stat (1021)
3569 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3570 - recovery for initial connections (2355)
3571 - fixes for mds_cleanup_orphans (1934)
3572 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3573 - block all file creations until orphan recovery completes (1901)
3574 - client remove rq_connection from request struct (2423)
3575 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3576 - recovery timer race (2670)
3577 - mdc_close recovey bug (2532)
3578 - ptlrpc cleanup bug (2710)
3579 - mds timeout on local locks (2588)
3580 - namespace lock held during RPCs (2431)
3581 - handle interrupted sync write properly (2503)
3582 - don't try to handle a message that hasn't been replied to (2699)
3583 - client assert failure during cleanup after abort recovery (2701)
3584 - leak mdc device after failed mount (2712)
3585 - ptlrpc_check_set allows timedout requests to complete (2714)
3586 - wait for inflight reqs when ptlrpcd finishes (2710)
3587 - make sure unregistered services are removed from the srv_list
3588 - reset bulk XID's when resending them (caught by 1138 test)
3589 - unregister_bulk after timeout
3590 - fix lconf error (2694)
3591 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3592 - readahead locks pages, leaves pending causing memory pressure (2673)
3593 - increase OST request buffers to 4096 on large machines (2729)
3594 - fix up permission of existing directories in simple_mkdir (2661)
3595 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3596 - don't assign transno to errored transactions (2742)
3597 - don't delete objects on OST if given a bogus objid from MDS (2751)
3598 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3599 - if rq_no_resend, then timeout request after recovery (2432)
3600 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3601 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3602 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3603 - MDS recovery shouldn't delete objects that it has given out (2730)
3604 - if enqueue arrives after completion, don't clobber LVB (2819)
3605 - don't unlock pages twice when trigger_group_io returns error (2814)
3606 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3607 - don't write pages to disk if there was an error (1450)
3608 - don't ping imports that have recovery disabled (2676)
3609 - take buffered bytes into account when balancing socknal conn (2817)
3610 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3611 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3612 - remove little-endian swabbing of llog records (1987)
3613 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3614 - retry reposting request buffers if they fail (1191)
3615 - grow extent at grant time to avoid granting a revoked lock (2809)
3616 - lock revoke doesn't evict page if covered by a second lock (2765)
3617 - disable VM readahead to avoid reading outside lock extents (2805)
3619 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3620 - updated kernel patches for hp-2.4.20 kernel (2681)
3622 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3625 - fix truncated write corruption (2366)
3626 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3628 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3629 - ensure len(uuid) < 37 in lmc (1171)
3630 - fix ia64 OOPS in llog_test (2255)
3631 - zero end of page at obdfilter for partial page writes (2648)
3632 - don't leave stale dentries around after renames (bug 2428)
3633 - fix timeouts when evicting a client with a single lock held (2642)
3634 - set deadline for the initial HELLO message to drain (2634)
3635 - print out dotted-quad IP addresses in the socknal (2302)
3637 - additional debugging for MDS client eviction problem (2443)
3638 - fix mkfsoptions support for osts (2603, 2604)
3640 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3643 - add series for the vanilla 2.6.0 kernel
3644 - add series for the vanilla 2.4.24 kernel
3645 - add series for a cray x86/64 UL kernel drop
3646 - fix xattr patches for the vanilla 2.4.19 series
3648 - generate true UUIDs in lmc (1171)
3649 - have portals stack dumping break in UML (2466)
3650 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3651 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3652 - wake callback waiting threads on client eviction (2460)
3653 - Add --ptldebug and --subsystem to lmc (1719)
3654 - update assertion to allow safe interrupt allocation
3655 - set rq_no_resend for cancel requests (2432)
3656 - recalculate ptlrpcd timeout after resend (2494)
3657 - call vfs_rmdir when removing pending directories (2368)
3658 - fix renaming a file to itself (2429)
3659 - lmc creates a default one-stripe lov (2454)
3660 - expand procfs space to handle large clusters (2326)
3661 - increase UML stack to avoid overflow
3662 - update lconf's list of debug and subsystem masks
3663 - fix lfs find --obd (2510)
3664 - /proc tunable for disabling filter read caching (2591)
3665 - stop rpm packages from altering slapd.conf (2301)
3666 - disable nagle in the socknal under 0conf (2578)
3667 - choose mds inode size based on stripe count (2572)
3668 - fix kernel-source rpm problems (2516)
3670 - add --disable-doc to avoid pdf generation (2421)
3671 - update documentation, tests, type-os, comments
3672 - avoid format warnings on ia64
3673 - remove the TOE NAL
3674 - tiny code cleanups by removing unused fields
3676 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3679 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3680 - remove the most blatant lies from BUILDING file (2371)
3681 - change default debug level to reasonable production setting
3682 - reduce client side cache size to reduce cache flush time
3683 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3684 - make TCP zerocopy and pinger support enabled by default (2476)
3685 - sync writes completed after process exits caused crashes (2319)
3686 - maintain correct mount count on the MDS (2356)
3687 - backout 1557, because 2316 wasn't really fixed
3688 - better file I/O statistics gathering in /proc
3689 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3690 - another kernel patch to fix zero-copy TCP function export
3691 - don't take duplicate lock when processing re-sent getattr (2420)
3692 - lctl uses obd_self_export instead of creating new conn (2353)
3693 - MDS/OST recovery case which requires object creation asserted (2425)
3694 - move lfs from /usr/sbin to /usr/bin in packages
3695 - fix race between mds_client_add and mds_client_free (2417)
3696 - use kmalloc instead of slabs in portals (2430)
3697 - don't create duplicate records when a failover MDS is present (2442)
3698 - remove unnecessary mount age check (2332)
3699 - don't remove directory inodes from locks prematurely (2451)
3700 - don't break if MDS service name is the same as hostname (2103)
3701 - fix races in client write RPC generation when cache full (2482)
3703 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3706 - remove now-unused request->rq_obd (278)
3707 - if an allocation fails, print out how much memory we've used (1933)
3708 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3709 - add the "configurable stack size" patch to most series files (1256)
3710 - ability to write large log records, for 100+ OST configs (2306)
3711 - fix NULL deref when filter_prep fails (2314)
3712 - fix operator precedence error in filter_sync
3713 - dynamic allocation of socknal TX descriptors (2315)
3714 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3715 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3716 - ensure that CWARN messages in Portals always get to the syslog
3717 - __init/__exit are not for prototype decls (ldlm_init/exit)
3718 - x86-64 compile warning fixes
3719 - fix gateway LMC keyword conflict (2318)
3720 - fix MDS lock inversions in getattr/reint paths (1844)
3721 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3722 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3723 - DLM race condition prevented some lock evictions (2328)
3724 - ENOMEM detection and retry on socknal sends (2230)
3725 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3726 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3727 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3728 - better error messages when a client is rejected during recovery (1505)
3729 - avoid cancelling locks which were never granted, after failure (2330)
3730 - fix i_sem/journal inversion in mds_client_add (2333)
3731 - fix truncate/getattr lock cycle deadlock (2334)
3732 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3733 - fix two rare exit paths which could leak an l_lock() ref (2321)
3734 - fencepost error in MDS/OST orphan recovery (2226)
3735 - make log record alignment 8 bytes (1988)
3736 - lstripe now fails when requested offset > ost_count (2237)
3737 - ensure that all kernel series have a complete list.h (1607)
3738 - fix crashes in special-file operations (2316)
3739 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3741 - allow configurable automake binary, for testing new versions
3742 - small update to the lfs documentation
3744 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3746 * fix negative export reference count in fsfilt_sync (2312)
3748 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3749 * release candidate 0.9.1
3751 - orphans are moved into the PENDING directory for possible recovery
3752 - replayed opens now open by fid for orphan/rename safety (1042)
3753 - last close of an orphan inode generates a transno (683)
3754 - chdir() and mount() now pin the directory entry (1020)
3755 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3756 - discard very old requests without processing them (1502)
3757 - remove some common, well-understood CERRORs (1505)
3758 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3759 - clear "grant" flags in OST replies until OST grant code lands (1644)
3760 - fix read performance by not clobbering i_blksize on client (1598)
3761 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3762 - make LRU size a /proc tunable, clears locks when reduced (707)
3763 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3764 - server locks take references on exports now (1558)
3765 - build fixes for 2.4.20-rh trees (1663)
3766 - return an error from lov_create if all OSCs are inactive (1751)
3767 - fix import levels when a reconnect happens without a timeout (1597)
3768 - exit early from mds_open if we get a lookup error (1749)
3769 - partial page read at EOF wouldn't wait for disk before sending (1642)
3770 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3771 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3772 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3773 - write/truncate lock inversion (1639)
3774 - Don't auto-load obdclass, portals modules during cleanup (1495)
3775 - fix timestamps from jumping to "now" (1763)
3776 - extra journal assertions (1648)
3777 - add an extra multiunlink test (1771)
3778 - fix read_record/write_record API (1776)
3779 - fix leak of offset_extent, possible incorrect i_size later (1772)
3780 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3781 - Debugging for the kqswnal_get_idle_tx problems (1820)
3782 - Allow recovery to be attempted multiple times (1536)
3783 - Write out MDS last_rcvd file after it is first created (1600)
3784 - Fix tx_descriptor leak in failed transmit situations (1827)
3785 - ext3 journaling fixes for assertion failure after IO error (1871)
3786 - class_export_put() on freed export after completion AST error (1896)
3787 - Fix revalidate looping in VFS (1322)
3788 - Don't access a freed export during MDS_REINT timeout (1521)
3789 - Add open-unlink recovery support on the MDS (1673,1764)
3790 - Return an error if no MDS data was read from last_rcvd (1946)
3791 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3792 - Don't setup a disk device that doesn't match exported UUID (317)
3793 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3794 - avoid committing NULL handle in force close
3795 - local.sh is now a one-stripe LOV configuration
3796 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3797 - fix ext3 htree duplicate directory entry corruption (1516)
3798 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3799 - update to new LOV EA format (2097)
3800 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3802 2003-06-15 Phil Schwan <phil@clusterfs.com>
3805 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3806 - per-import/export recovery handling (958, 931, 959)
3807 - multiple last-rcvd slots, for serving multiple FSes (949)
3808 - connections are again shared between multiple imp/exports (963, 964)
3809 - "umount -f" would hang if any requests needed to be sent (393, 978)
3810 - avoid pinning large req buffer by copying for queued messages (989)
3811 - add "uuid" to "lctl device" command to help upcalls (991)
3812 - "open" RPCs with transnos would confuse recovery counters (1037)
3813 - do proper endian conversion of all wire messages (288, 340, 891)
3814 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3815 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3816 - fix ptlrpc_connection leak in target_handle_connect (1174)
3817 - fix import refcounting bug in OST and MDS cleanup (1134)
3818 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3819 - fix very unlikely obd_types race condition (501)
3820 - remove osc_open hack for echo_client (1187)
3821 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3822 - a failure in read_inode2 leads to deadlock (1139)
3823 - cancel ack-locks as soon as transaction is committed (1072)
3824 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3825 - make sure to commitrw after any preprw to avoid deadlock (1162)
3826 - failing to execute a file in a lustre FS would lock inode (1203)
3827 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3828 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3829 - obd_uuid2tgt would walk off the end of the list (1255)
3830 - on IA64 the osc would give portals incorrect bulk size (1258)
3831 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3832 - fix lock inversion caused by new llite matching code (1282)
3833 - limit the number of dirty pages on a client to 10MB (1286)
3834 - timed out locks were not being corrected cancelled (1289)
3835 - fix O_DIRECT above 4GB on IA-32 (1292)
3836 * major user-visible changes
3837 - fail out/fail over policy now controlled by the upcall (993)
3839 - add OBD_PING to check server availability and failure (954)
3840 - lustre messages are now sent in sending host order (288, 340, 891)
3841 - add eadatalen to MDS getattr reply (340)
3842 - OST read replies may contain second buffer, with per-page status (593)
3844 2003-03-11 Phil Schwan <phil@clusterfs.com>
3847 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3848 - failed open()s could cause deadlock; fixed (867, 869)
3849 - stop cancelling OST locks when files are closed (481)
3850 - overlapping XID spaces caused network corruption (851, 853)
3851 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3852 - setattr_raw added, to do single-RPC, server-side setattrs
3853 - lmc/lconf syntax change for OST UUIDs
3854 - fix crashy race condition between ptlrpc_free_req and osc_close
3855 - don't use request in mdc_enqueue if we hit a timeout (889)
3856 - don't set the inode i_size for regular files from the MDS (896)
3857 - handle out of order completion AST (842)
3858 - don't LBUG if a lock request times out after receiving AST (913)
3859 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3860 - if a bad lock AST arrives, send an error instead of dropping entirely
3861 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3862 - fix leak in bulk IO when only partially completed (899, 900, 926)
3863 - fix O_DIRECT for ia64 (55)
3864 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3865 - C-z after timeout could hang a process forever; fixed (977)
3867 - client-side I/O cache (678, 924, 929, 941, 970)
3869 - READPAGE and SETATTRs which don't take server-side locks get
3872 2003-02-11 Phil Schwan <phil@clusterfs.com>
3875 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3876 - Fix fsfilt_extN_readpage() to read a full page of directory
3877 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3878 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3879 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3880 - Most intent operations were removed, in favour of a new RPC mode
3881 that does a single RPC to the server and bypasses most of the VFS
3882 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3883 - Aggressively cancel local locks on DLM servers
3884 - mds_reint_unlink sends EA to the client if it's the last nlink.
3885 client uses that EA to unlink OST objects.
3886 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3887 - recursive symlinks were fixed (439)
3888 - fixed NULL deref in DEBUG_REQ
3889 - filter_update_lastobjid no longer calls sync, which annoyed extN
3890 - fixed multi-client small-writes to a single file problem (445)
3891 - fixed mtime updates during file writes (607)
3892 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3893 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3894 - fixed Linux OST size reporting problem (444, 656)
3895 - OST now updates object mtime with writes or setattr (607, 619)
3896 - client verifies file size before zeroing page past EOF (445)
3897 - OST now writes last allocated objid to disk with allocation (108)
3898 - LOV on echo now works (409)
3900 - mds_reint_unlink sends a new buffer, with the EA included. this
3901 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3902 set if a regular file was being unlinked, and it was the last link
3903 - use PtlGet from the target for bulk writes (315)
3904 - OST now updates object mtime with writes or setattr (607, 619)
3905 - LDLM now has a grant-time callback to revalidate locked items, if
3907 - Many MDS operations were reorganized to combat race conditions
3909 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3910 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3912 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3915 - Fully reactivate OST imports after reconnection (512, others)
3916 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3917 - More graceful error handling for truncating on dead OST (515)
3918 - Don't error out unless we're actually accessing dead stripes (474)
3919 - Fix garbage sizes when stripes are missing (410)
3920 - LRU counters were broken, causing constant lock purge (433, 432)
3921 - garbage on read from stripes with failed OSTs (441)
3922 - mark OSCs as active before reconnecting during recovery (438)
3923 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3924 - lfind did not preserve OST order in output (443)
3925 - symlinks cause hung clients, incorrect data (439)
3926 - stop dereferencing request after dropping refcount (457)
3927 - don't LASSERT(spin_is_locked) on non-SMP (455)
3928 - fixes for many rename() bugs
3929 - fstat didn't correctly synchronize attributes (399)
3930 - server must handle lock cancellation during blocking AST prep (487)
3931 - bulk descriptors were free()d too soon (511)
3932 - fix paths in lconf, which would load incorrect modules (451, 507)
3933 - fix confusing lconf 'host not found' error message (386)
3934 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3935 - fix race condition in mdc_blocking_ast() for inode access (526)
3936 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3937 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3938 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3939 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3940 - fix lstripe in several ways (kernel side) (527)
3941 - fix request leak in ldlm_cli_enqueue (262)
3942 - incorrect OSC was marked inactive after OST failure
3943 - call mds_fs_cleanup before unmounting filesystem (524)
3944 - fix races between taking ns_lock and ldlm_lock_change_resource
3945 - fix races updating LOV export open file list
3946 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3947 - fix recovery NULL deref in ldlm_cli_cancel_unused
3948 - fix some DLM races by using new hash table for lock handles (419)
3949 - permit the client to specify desired inodes, at replay
3950 - duplicate requests when we queue them for replay reintegration
3951 - fix last_rcvd offset calculation
3952 - sync after each recovered transaction, so we always make progress
3953 - never, not always, ERESTART requests without transnos
3954 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3956 - skip replay if the MDS says that the client is already connected
3957 - don't check for a recovery-enabled export to match lctl's UUID
3958 - don't INC_USE_COUNT for phantom exports
3959 - don't crash when cleaning up phantom exports (567)
3960 - don't double-finish or set replay data for errored mdc_open requests
3961 - abort requests when they time out, so we don't get old replies
3962 - send/receive replies for AST messages again
3963 - if the client says that it doesn't have the lock, cancel it on the
3965 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3966 mark it as destroyed, it will all work out when decref is called
3967 - fix module use counts (22, 581)
3969 - ASTs now expect a reply (server cancels lock on error reply)
3971 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3974 - fix many simultaneous client startup (392)
3975 - fix dentry->d_it clobbering
3976 - credentials weren't being shipped for readdir/getattr operations
3977 - remove invalid assertions triggered during some concurrent MD
3979 - proper Lustre versions added (336, 389)
3980 - fix memory leak for create error case (398)
3981 - fix LOV locking bug that would get cli/srv out of sync
3982 - fix echo client over LOV (409)
3983 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3984 - fix double-O_EXCL intent crash (424)
3985 - avoid sending multiple lock CANCELs (352)
3987 - MDS can do multi-client recovery (modulo bugs in new code)
3989 - many updates, edits, cleanups
3991 2002-11-18 Phil Schwan <phil@clusterfs.com>
3994 - fix null d_it dereference (346)
3995 - fix full OST/dbench hang (333)
3996 - fix permission problem with file removal (286)
3997 - fix removal of OSCs from LOV when they fail
3998 - fix NULL deref during bulk timeout (214)
3999 - fix problems related to multiple filesystems on one MDS (241)
4000 - fixed serious subtle metadata locking bugs
4001 - free locks on clients when inodes are removed due to memory
4003 - fix inode pointer in lock data (285)
4004 - partial support for multiple MDS on a single host (241)
4005 - data locks weren't cancelled at clear_inode time (290, 311)
4006 - intent locks could lead to unbounded lock growth (205)
4007 - added a maximum lock count, an LRU list, and a flusher
4008 - fix multiple rename (365)
4009 - properly abstracted the echo client
4010 - OSC locked 1 byte too many; fixed
4011 - rewrote brw callback code:
4012 - fixed recovery bugs related to LOVs (306)
4013 - fixed too-many-pages-in-one-write crash (191)
4014 - fixed (again) crash in sync_io_timeout (214)
4015 - probably fixed callback-related race (385)
4017 - Add capability to MDS protocol
4018 - LDLM cancellations and callbacks on different portals
4020 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4023 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4024 - defer open object destruction to close time (601981, 138)
4025 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4026 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4027 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4028 - don't call obd_setattr in ll_file_release for destroyed objects
4029 * protocol change to lustre_msg: move |version| and add |flags|
4030 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4031 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4032 * added replay of create, unlink, link and rename operations during
4033 MDS failover; recovery should be much more robust now
4034 * remove failed OSCs from LOVs (only lov_create uses this so far)
4035 * the lustre-HOWTO was brought (more) up to date (582544)
4037 2002-10-23 Phil Schwan <phil@clusterfs.com>
4040 - in-use dentries weren't being reused properly (617851)
4041 - prevent multiple LDLM setup (599178)
4042 - fix LOV size calculations for truncate (617853)
4043 - fix client handling of MDS intent errors (POSIX)
4044 - fix permission bug in lovstripe.c test (624321)
4045 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4046 - truncate past end of file could corrupt data
4047 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4048 - a race in recovery could return ETIMEDOUT to apps (623947)
4049 - building outside the source directory was fixed
4050 * the lustre-HOWTO was brought (more) up to date (582544)
4051 * major progress was made on recovery functionality
4053 2002-10-10 Phil Schwan <phil@clusterfs.com>
4056 - recovery deadlock fix
4057 - rm -rf causes LBUG fix (617817)
4058 - file open by multiple tasks fix (618962)
4059 - directory permissions bugs (602707 and 620007)
4060 - journal_stop fixed with locking (611313)
4061 - O_APPEND failures resolved (618273, perhaps 614459)
4062 - lconf PATH fix (619770)
4063 - IA64 build fix (621450)
4064 - RPC buffer sizes scale with amount of memory
4066 2002-10-01 Phil Schwan <phil@clusterfs.com>
4069 - locks would be cancelled without throwing away data pages,
4070 resulting in inconsistent data (605627)
4071 - inode attributes were not always being refreshed (605627, 612449)
4072 - lconf now continues to cleanup after lctl reports an error
4073 - MDS now enforces user permissions (602707)
4074 - lprocfs cleanup fixed, but not yet enabled (614157)
4075 - fixed infinite server hang, should a client not respond to an AST
4076 - avoid going into recovery if user calls readlink() with a buffer
4077 that's too small (613941)
4078 - AST RPCs no longer require replies (614867) -- this may be changed
4079 - don't crash server if client sends an IOV that's too big (611336)
4080 - fixed lock conversion deadlock (611892)
4081 - fixed the following of symlinks (614622)
4082 * recovery: the server can remove locks from a client that dies, other
4083 clients can make progress
4084 * more extN patch fixes
4085 * compile-time configurable ptlrpc buffer allocations
4087 - collaborative read cache document
4088 - Lustre Lite Performance CDR document-in-progress
4090 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4093 - fix typo in patch-2.4.18
4095 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4098 - clear ptlrpc request each time in handle_incoming_request()
4099 - unlink of files now destroys the object on the OST
4101 2002-09-19 Peter Braam <braam@clusterfs.com>
4103 * add hard link support
4104 * change obdfile creation method
4105 * kernel patch changed
4107 2002-09-19 Peter Braam <braam@clusterfs.com>
4110 - stack overflow bug in extN fixed
4112 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4114 * documentation updates
4115 - add man pages for config tools
4116 - update tests/README to describe testing with new config tools
4117 - finish metadata API descriptions
4118 * bug fixes and cleanups
4119 - statfs workaround for 16TB limit
4120 - LOV stripe allocation improved, can stripe on subset of OSTs
4121 - LOV file size/IO offset was wrong for files > 4GB in size
4122 - object EA data was being dropped, caused files to be unreadable
4123 - memory overflow with non-LOV OST caused memory corruption
4124 - fixed regression tests to work with new config tools, obdfilter
4125 - fixed bug when directory size became larger than 1 block
4126 - fixed bug (for single client case) when PWD was deleted
4127 - invalidate local directory pages when doing intent-based ops
4128 - avoid LDLM oops when lock callback contained bad data
4130 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4132 * documentation updates
4133 * bug fixes and cleanups
4134 - configuration tools
4137 - 64-bit compile warnings
4138 - 64-bit internal statfs data
4140 * test_brw on persistent OST devices
4142 * lprocfs (disabled)
4144 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4146 * documentation updates
4147 * bug fixes and cleanups
4148 * configuration tools
4150 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4153 * many small fixes to 0_5_4
4154 * io/network handling
4155 * thinkos in MDS operations
4157 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4160 * crucial basic fixes to 0.5.3
4161 * IOR, Iozone work over Elan
4164 2002-08-07 Phil Schwan <phil@clusterfs.com>
4165 * version 0_5_3, our first alpha
4166 * we use the new Portals iovs
4167 * documentation updates
4168 * bug fixes and cleanups
4169 * small changes in the DLM wire protocol
4171 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4172 * version 0_5_1 with some initial stability,
4173 * locking on MD and file I/O.
4174 * documentation updates
4175 * several bug fixes since 0.5.0
4176 * small changes in wire protocol
4178 2002-07-18 Phil Schwan <phil@clusterfs.com>
4180 * delivered as Lustre Light Alpha
4181 * fixed a crash after handling invalid MDS requests
4182 * fixed directory pages for architectures with non-4k pages sizes
4184 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4185 * release version v0_4_4
4186 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4188 * Fixes a number of interruption problems with OST operations.
4189 * Update documentation for portals header changes
4190 * Move all wire protocol structs/defines to lustre_idl.h
4191 * Fixes symlink length bug.
4192 * Add tcpdump to repository.
4194 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4195 * release version v0_4_3
4196 * Fixes statfs for inodes on extN.
4197 * Fixes bug in runtests which would delete /etc/hosts.
4198 * Use 64-bit object IDs wherever possible (not into VFS though)
4199 Remove ost_get_info, which is unused by lustre, and out of date.
4201 2002-07-03 Peter Braam <braam@clusterfs.com>
4202 * release version v0_4_2 Fixes a lookup error (type not passed)
4203 * move forward to head of Portals
4204 * move forward to latest Lustre kernel
4206 2002-06-25 Peter Braam <braam@clusterfs.com>
4207 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob