4 2.6.9-55.0.9.EL (RHEL 4),
5 2.6.16.53-0.8 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: Remove 2.4 kernel support from lustre 1.8.0
18 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
19 quota_interface.c in quota subsystem.
20 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
22 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
23 lvfs_linux.c upcall_cache.c in lvfs subsystem.
24 Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
25 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
27 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
28 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
30 Remove 2.4 kernel check from lproc_osc.c, osc_request.c in osc
32 Remove 2.4 kernel check from filter.c, lproc_obdfilter.c in
34 Remove 2.4 kernel code from lustre/obdecho/echo_client.c.
35 Remove 2.4 kernel check from lustre/mgs/lproc_mgs.c,
37 Remove 2.4 kernel code from lustre/mgc/mgc_request.c.
38 Remove 2.4 kernel check from lustre/mdt/mdt_lproc.c.
40 Severity : enhancement
42 Description: Build SLES10 patchless client fails
43 Details : The configure was broken by run ./configure with
44 --with-linux-obj=.... argument for patchless client. When the
45 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
46 can't search header adequately. Use absolute path such as
47 -I($LINUX)/include instead.
51 Description: interrupt oig_wait produce painc on resend.
52 Details : brw_redo_request can be used for resend requests from ptlrpcd and
53 private set, and this produce situation when rq_ptlrpcd_data not
54 copyed to new allocated request and triggered LBUG on assert
55 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
56 ptlrpcd set if request is changed and can be safety changed to use
61 Description: LASSERT_{REQ,REP}SWAB macros are buggy
62 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
63 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
64 these macros and replace them with their difinitions instead.
68 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
69 Details : Update which_patch & target file for SLES10 latest kernel.
72 Description: Remove client patches from SLES 10 kernel.
73 Details : This causes SLES 10 clients to behave as patchless clients
74 even on a Lustre-patched (server) kernel.
76 Severity : enhancement
78 Description: self-adjustable client's lru lists
79 Details : use adaptive algorithm for managing client cached locks lru
80 lists according to current server load, other client's work
81 pattern, memory activities, etc. Both, server and client
82 side namespaces provide number of proc tunables for controlling
85 Severity : enhancement
87 Description: light-weight GSS support
88 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
89 and reduce performance overhead.
91 Severity : enhancement
93 Description: Linux keyring support
94 Details : Support using service of Linux keyring for Lustre GSS internal
95 context refresh/cache mechanism.
99 Description: Fix errors in lfs documentation
100 Details : Fixes man pages
104 Description: don't use GFP_* in generic Lustre code.
105 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
109 Description: obdclass is limited by single OBD_ALLOC(idarray)
110 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
114 Description: warnings with build patchless client with vanila 2.6.19 and up
115 Details : change old ctl_table style and replace ctl_table/ctl_table_header
116 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
120 Frequency : Only for SLES
121 Description: sanity_quota fail test_1
122 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
123 file, which makes TSTID[2] inunique.
127 Frequency : Only for SLES9
128 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
129 Details : Update target/ChangeLog/which_patch .
133 Frequency : Only for test_10 in sanity-quota.sh
134 Description: a bug in quota test code
135 Details : checking a test flag in a wrong place
139 Frequency : Only for test_13 in sanity-quota.sh
140 Description: a bug in quota test script
141 Details : shouldn't check overfull allocated quota
145 Description: jbd statistics
146 Details : Port older jbd statistics patch for sles10
148 Severity : enhancement
150 Description: Add printing inode info into message about error in writepage.
155 Description: reply_lock_interpret crash due to race with it and lock cancel.
156 Details : Do not replay locks that are being cancelled. Do not reference
157 locks by their address during replay, just by their handle.
161 Frequency : When flocks are used.
162 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
163 Details : Flock locks might destroy just granted lock if it could be merged
164 with another existing flock, this is done in completion handler,
165 so teach ldlm_cli_enquque_fini that this is a valid case for
171 Description: Oops in read and write path when failing to allocate lock.
172 Details : Check if lock allocation failed and return error back.
176 Description: lstripe command fails for valid OST index
177 Details : The stripe offset is compared to lov->desc.ld_tgt_count
178 instead of lov->desc.ld_active_tgt_count.
182 Description: sanity.sh failed test 103
183 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
188 Description: df doesn't work properly if diskfs blocksize != 4K
189 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
193 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
194 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
198 Description: obd_health_check_timeout too short
199 Details : set obd_health_check_timeout as 1.5x of obd_timeout
203 Description: llapi_file_create() does not allow some changes
204 Details : add llapi_file_open() that allows specifying the mode and
205 open flags, and also returns an open file handle.
209 Description: merge and cleanup kernel patches.
210 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
214 Description: Add journal checksum support.(Kernel part)
215 Details : The journal checksum feature adds two new flags i.e
216 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
217 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
218 indicates that the commit block contains the checksum for
219 the blocks described by the descriptor blocks. Now commit
220 record can be sent to disk without waiting for descriptor
221 blocks to be written to disk. This behavior is controlled
222 using JBD2_FEATURE_ASYNC_COMMIT flag.
226 Description: OSS needs mutliple precreate threads
227 Details : Add ability to start more than one create thread per OSS.
231 Description: Sanity.sh test_65k failed: setstripe should have succeeded
232 Details : execute lfs setstripe on client
236 Description: mds_obd_create error creating tmp object
237 Details : When the user sets quota on root, llog will be affected and can't
238 create files and write files.
241 Frequency : Always on ia64 patchless client, and possibly others.
243 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
244 Details : This allows the patchless client to be loaded on architectures
249 Description: RedHat Update kernel for RHEL5
250 Details : Add the kernel config file for RHEL5.
254 Description: RedHat Update kernel for RHEL5
255 Details : Modify the kernel config file more closer RHEL5.
259 Description: Build failure against Centos5 (RHEL5)
260 Details : Define PAGE_SIZE when it isn't present.
264 Description: test_7 on sanity-quota.sh can't be used on separate machines
265 Details : add facet to handle it
269 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
270 Details : with NFS, the anon dentry's parent was set to itself in
271 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
272 corresponding dentry other than use rec->ur_name.
274 Severity : enhancement
276 Description: lfs setstripe enhancement
277 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
281 Description: testing performance impact of enabling checksumming
282 Details : enable checksum by default, allow --disable-checksum
283 configure option and "-o nochecksum" mount option
287 Description: lustre support for RHEL5
288 Details : Add support for RHEL5.
292 Description: Client eviction due to failover config
293 Details : after a connection loss, the lustre client should attempt to
294 reconnect to the last active server first before trying the
295 other potential connections.
299 Description: when mds and osts use different quota unit(32bit and 64bit),
300 quota will be released repeatly.
301 Details : void sending multiple quota reqs to mds, which will keep the
302 status between the reqs.
306 Description: osts not allocated evenly to files
307 Details : change the condition to increase offset_idx
310 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
312 Description: Data corruption for OSTs that are formatted larger than 2TB
314 Details : When generating the bio request for lustre file writes the
315 sector number would overflow a temporary variable before being
316 used for the IO. The data reads correctly from Lustre (which
317 will overflow in a similar manner) but other file data or
318 filesystem metadata may be corrupted in some cases.
322 Description: Tune the kernel for good SCSI performance.
323 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
324 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
329 Description: rewrite ext2-derived code in obdclass/uuid.c
330 Details : rewrite inherited code (uuid parsing code from ext2 utils)
331 from scratch preserving functionality.
335 Description: "lfs find -obd UUID" prints directories
336 Details : "lfs find -obd UUID" will return all directory names instead
337 of just file names. It is incorrect because the directories
338 do not reside on the OSTs.
342 Description: use i_size_read and i_size_write in 2.6 port
343 Details : replace inode->i_size access with i_size_read/write()
346 Frequency : when removing large files
348 Description: scheduling issue during removal of large Lustre files
349 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
350 It causes scheduling issues when removing large files (17TB in the
354 Frequency : only with liblustre clients on XT3
356 Description: evictions taking too long
357 Details : allow llrd to evict clients directly on OSTs
360 Frequency : only on ppc
362 Description: /proc/fs/lustre/devices broken on ppc
363 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
364 We should make sure the type of 'index' is loff_t and then cast
365 explicitly as needed below. Do not assign an explicitly cast
369 Frequency : only for rhel5
371 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
372 Details : Modify the target file & which_kernel.
374 Severity : enhancement
376 Description: omit set fsid for export NFS
377 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
378 as FS_REQUIRES_DEV for avoid problems with generate fsid.
382 Frequency : Always, for kernels after 2.6.16
383 Description: Fix warning idr_remove called for id=.. which is not
385 Details : Last kernels save old s_dev before kill super and not allow
386 to restore from callback - restore it before call kill_anon_super.
390 Description: buffer overruns could theoretically occur
391 Details : llapi_semantic_traverse() modifies the "path" argument by
392 appending values to the end of the origin string, and a
393 overrun may occur. Adding buffer overrun check in liblustreapi.
397 Description: Fix error on 'ls .' at the top of the Lustre mount.
398 Details : Don't revalidate dentry if it is a root dentry.
402 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
403 Details : Modify vm-tunables-rhel4.patch.
407 Description: Kernel config for 2.6.18-vanilla.
408 Details : Modify targets/2.6-vanilla.target.in.
409 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
410 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
411 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
412 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
416 Description: improve handling recoverable errors
417 Details : if request processig with error which can be recoverable on server
418 request should be resend, otherwise page released from cache and
423 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
424 Details : Modify target file & which_patch.
425 A flaw was found in the IA32 system call emulation provided
426 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
427 value could be stored in the %RAX register, which could trigger an
428 out-of-bounds system call table access. An untrusted local user
429 could exploit this flaw to run code in the kernel
430 (ie a root privilege escalation). (CVE-2007-4573).
434 Description: O_DIRECT bypasses client statistics.
435 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
436 read_ahead_stats were updated. copied the stats section from
437 osc_send_oap_rpc() into async_internal().
441 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
445 Description: change order of libsysio includes
446 Details : '#include sysio.h' should always come before '#include xtio.h'
450 Description: handle "serious error: objid * already exists" more gracefully
451 Details : If LAST_ID value on disk is smaller than the objects existing in
452 the O/0/d* directories, it indicates disk corruption and causes an
453 LBUG(). If the object is 0-length, then we should use the existing
454 object. This will help to avoid a full fsck in most cases.
456 Severity : enhancement
458 Description: adapt the lustre_config script to support the upgrade case
459 Details : Add "-u" option for lustre_config script to support upgrading 1.4
460 server targets to 1.6 in parallel.
464 Description: To avoid grant space > avaible space when the disk is almost
465 full. Without this patch you might see the error "grant XXXX >
466 available" or some LBUG about grant, when the disk is almost
468 Details : In filter_check_grant, for non_grant cache write, we should
469 check the left space by if (*left > ungranted + bytes), instead
470 of (*left > ungranted), because only we are sure the left space
471 is enough for another "bytes", then the ungrant space should be
472 increase. In client, we should update cl_avail_grant only there
473 is OBD_MD_FLGRANT in the reply.
477 Description: Update RHEL 4 kernel to fix local root privilege escalation.
478 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
479 described in CVE-2007-4573. This problem could allow untrusted
480 local users to gain root access.
484 Frequency : when using O_DIRECT and quotas
486 Description: Incorrect file ownership on O_DIRECT output files
487 Details : block usage reported by 'lfs quota' does not take into account
488 files that have been written with O_DIRECT.
490 --------------------------------------------------------------------------------
492 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
494 * Support for kernels:
495 2.6.5-7.283 (SLES 9),
496 2.6.9-55.EL (RHEL 4),
497 2.6.16.46-0.14 (SLES 10),
498 2.6.18.8 vanilla (kernel.org)
499 * Client support for unpatched kernels:
500 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
501 2.6.16 - 2.6.22 vanilla (kernel.org)
502 * Due to recently discovered recovery problems, we do not recommend
503 using patchless RHEL 4 clients with this or any earlier release.
504 * Recommended e2fsprogs version: 1.39.cfs8
505 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
506 * Starting with this release, the ldiskfs backing filesystem required
507 by Lustre is now in its own package, lustre-ldiskfs. This package
508 should be installed. It is versioned separately from Lustre and
509 may be released separately in future.
513 Description: block reactivating mgc import until all deactivates complete
514 Details : Fix race when failing back MDT/MGS to itself (testing)
517 Frequency : at statup only
519 Description: mds_lov_synchronize race leads to various problems
520 Details : simultaneous MDT->OST connections at startup can cause the
521 sync to abort, leaving the OSC in a bad state.
523 Severity : enhancement
525 Description: add optional extra BUILD_VERSION info
526 Details : add a new environment variable (namely LUSTRE_VERS) which allows
527 to override the lustre version.
530 Frequency : 2.6.18 servers only
532 Description: ll_kern_mount() doesn't release the module reference
533 Details : The ldiskfs module reference count never drops down to 0
534 because ll_kern_mount() doesn't release the module reference.
539 Description: server LBUG when using old ost_num_threads parameter
540 Details : Accept the old ost_num_threads parameter but warn that it
541 is deprecated, and fix an off-by-one error that caused an LBUG.
546 Description: Transient SCSI error results in persistent IO issue
547 Details : iobuf->dr_error is not reinitialized to 0 between two
551 Frequency : sometimes when underlying device returns I/O errors
553 Description: OSTs not going read-only during write failures
554 Details : OSTs are not remounted read-only when the journal commit threads
555 get I/O errors because fsfilt_ext3 calls journal_start/stop()
556 instead of the ext3 wrappers.
560 Description: poor connect scaling with increasing client count
561 Details : Don't run filter_grant_sanity_check for more than 100 exports
562 to improve scaling for large numbers of clients.
565 Frequency : SLES10 only
567 Description: sanity-quota.sh quotacheck failed: rc = -22
568 Details : Quotas cannot be enabled on SLES10.
571 Frequency : liblustre clients only
573 Description: getdirentries does not give error when run on compute nodes
574 Details : getdirentries does not fail when the size specified as an argument
575 is too small to contain at least one entry
577 Severity : enhancement
579 Description: Add LNET router traceability for debug purposes
580 Details : If a checksum failure occurs with a router as part of the
581 IO path, the NID of the last router that forwarded the bulk data
582 is printed so it can be identified.
587 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
588 Details : Due to a race condition, liblustre clients were occasionally
591 Severity : enhancement
593 Description: lfs setstripe use optional parameters instead of postional
596 Severity : enhancement
598 Description: Nanosecond timestamp support for ldiskfs
599 Details : The on-disk ldiskfs filesystem has added support for nanosecond
600 resolution timestamps. There is not yet support for this at
601 the Lustre filesystem level.
604 Frequency : during server recovery
606 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
607 Details : request with rq_no_resend flag not awake l_wait_event if they get a
611 Frequency : nfs export on patchless client
613 Description: connectathon hang when test nfs export over patchless client
614 Details : Disconnected dentry cannot be found with lookup, so we do not need
615 to unhash it or make it invalid
618 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
619 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
620 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
625 Description: open req refcounting wrong on reconnect
626 Details : If reconnect happened between getting open reply from server and
627 call to mdc_set_replay_data in ll_file_open, we will schedule
628 replay for unreferenced request that we are about to free.
629 Subsequent close will crash in variety of ways.
630 Check that request is still eligible for replay in
631 mdc_set_replay_data().
636 Description: disable writes to filesystem when reading health_check file
637 Details : the default for reading the health_check proc file has changed
638 to NOT do a journal transaction and write to disk, because this
639 can cause reads of the /proc file to hang and block HA state
640 checking on a healthy but otherwise heavily loaded system. It
641 is possible to return to the previous behaviour during configure
642 with --enable-health-write.
644 Severity : enhancement
646 Description: 64-bit inode version
647 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
648 to the inode. This will be required for version-based recovery.
653 Description: MDS fails to start if a duplicate client export is detected
654 Details : in some rare cases it was possible for a client to connect to
655 an MDS multiple times. Upon recovery the MDS would detect this
656 and fail during startup. Handle this more gracefully.
658 Severity : enhancement
660 Description: Add -o localflock option to simulate old noflock
662 Details : This will achieve local-only flock/fcntl locks
668 Description: log_commit_thread vs filter_destroy race leads to crash
669 Details : Take import reference before releasing llog record semaphore
674 Description: Wrong request locking in request set processing
675 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
676 imp_delayed_list, in this place should be used imp_lock.
679 Frequency : when reconnection
681 Description: Grant Leak when osc reconnect to OST
682 Details : When osc reconnect ost, OST(filter) should check whether it
683 should grant more space to client by comparing fed_grant and
684 cl_avail_grant, and return the granted space to client instead
685 of "new granted" space, because client will call osc_init_grant
686 to update the client grant space info.
689 Frequency : when client reconnect to OST
691 Description: Grant Leak when osc do resend and replay bulk write
692 Details : When osc reconnect to OST, OST(filter)should clear grant info of
693 bulk write request, because the grant info while be sync between
694 OSC and OST when reconnect, and we should ignore the grant info
695 these of resend/replay write req.
700 Description: Grant space more than avaiable left space sometimes.
701 Details : When then OST is about to be full, if two bulk writing from
702 different clients came to OST. Accord the avaliable space of the
703 OST, the first req should be permitted, and the second one
704 should be denied by ENOSPC. But if the seconde arrived before
705 the first one is commited. The OST might wrongly permit second
706 writing, which will cause grant space > avaiable space.
709 Frequency : when client is evicted
711 Description: Grant might be wrongly erased when osc is evicted by OST
712 Details : when the import is evicted by server, it will fork another
713 thread ptlrpc_invalidate_import_thread to invalidate the
714 import, where the grant will be set to 0. While the original
715 thread will update the grant it got when connecting. So if
716 the former happened latter, the grant will be wrongly errased
717 because of this race.
722 Description: Checking Stale with correct fid
723 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
724 to check whether it is stale, because sometimes, we want the
725 enqueue happened anyway, and op_data.fid2 will not be initialized.
727 Severity : enhancement
729 Description: update patchless client
730 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
733 Frequency : only with 2.4 kernel
735 Description: random memory corruption
736 Details : size of struct ll_inode_info is to big for union inode.u and this
737 can be cause of random memory corruption.
742 Description: Memory leak in recovery
743 Details : Lov_mds_md was not free in an error handler in mds_create_object.
744 It should also check obd_fail before fsfilt_start, otherwise if
745 fsfilt_start return -EROFS,(failover mds during mds recovery).
746 then the req will return with repmsg->transno = 0 and rc = EROFS.
747 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
748 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
749 be freed no matter whether fsfilt_commit success or not.
752 Frequency : only with huge count clients
754 Description: Prevents from taking the superblock lock in llap_from_page for
756 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
757 for llap_from_page call in ll_removepage prevents from taking the
758 superblock lock for a soon died page.
763 Description: Not check open intent error before release open handle
764 Details : in some rare cases, the open intent error is not checked before
765 release open handle, which may cause
766 ASSERTION(open_req->rq_transno != 0), because it tries to release
767 the failed open handle.
772 Description: Set cat log bitmap only after create log success.
773 Details : in some rare cases, the cat log bitmap is set too early. and it
774 should be set only after create log success.
778 Description: Accessing a block bevice can re-enable I/O when Lustre is
779 tearing down a device.
780 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
785 Description: service threads may hog cpus when there are a lot of requests
787 Details : Insert cond_resched to give other threads a chance to use some of
793 Description: the cat log was not initialized in recovery
794 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
795 unlink log on mds will not be initialized until mds post
796 recovery. And also in mds post recovery, the unlink log will
797 initialization will be done asynchronausly, so there will be race
798 between add unlink log and unlink log initialization.
802 Description: brw_stats were being printed incorrectly
803 Details : brw_stats were being printed as log2 but all of them were not
804 recorded as log2. Also remove some code duplication arising from
805 filter_tally_{read,write}.
809 Frequency : rare, only in recovery.
810 Description: ASSERTION(req->rq_type != LI_POISON) failed
811 Details : imp_lock should be held while iterating over imp_sending_list for
812 prevent destroy request after get timeout in ptlrpc_queue_wait.
816 Description: replay-single.sh test 52 fails
817 Details : A lock's skiplist need to be cleanup when it being unlinked
818 from its resource list.
822 Description: Short directio read returns full requested size rather than
824 Details : Direct I/O operations should return actual amount of bytes
825 transferred rather than requested size.
827 Severity : enhancement
829 Description: metadata RPC reduction (e.g. for rm performance)
830 Details : decrease the amount of synchronous RPC between clients and servers
831 by canceling conflicing lock before the operation on the client side
832 and packing thier handles into the main operation RPC to server.
834 Severity : enhancement
836 Description: Async OSC create to avoid the blocking unnecessarily.
837 Details : If a OST has no remain object, system will block on the creating
838 when need to create a new object on this OST. Now, ways use
839 pre-created objects when available, instead of blocking on an
840 empty osc while others are not empty. If we must block, we block
841 for the shortest possible period of time.
843 --------------------------------------------------------------------------------
845 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
850 Frequency : on some architectures
852 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
853 Details : Uninitialized flags sometimes cause configuration commands to
857 Frequency : patchless clients only
859 Description: missing __iget() symbol export
860 Details : The __iget() symbol export is missing. To avoid the need for
861 this on patchless clients the deathrow inode reaper is turned
862 off, and we depend on the VM to clean up old inodes. This
863 dependency was during via the fix for bug 12181.
865 --------------------------------------------------------------------------------
867 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
869 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
870 INTEROPERATE with older versions automatically. In many cases a
871 special upgrade step is needed. Please read the
872 user documentation before upgrading any part of a live system.
873 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
874 version will not interoperate with 1.6 betas before beta5 (1.5.95).
875 * WARNING: Lustre configuration and startup changes are required with
876 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
882 Severity : enhancement
884 Description: MountConf
885 Details : Lustre configuration is now managed via mkfs and mount
886 commands instead of lmc and lconf. New obd types (MGS, MGC)
887 are added for dynamic configuration management. See
888 https://mail.clusterfs.com/wikis/lustre/MountConf for
891 Severity : enhancement
893 Description: dynamic OST addition
894 Details : OSTs can now be added to a live filesystem
896 Severity : enhancement
898 Description: startup order invariance
899 Details : MDTs and OSTs can be started in any order. Clients only
900 require the MDT to complete startup.
902 Severity : enhancement
904 Description: parallel, asynchronous orphan cleanup
905 Details : orphan cleanup is now performed in separate threads for each
906 OST, allowing parallel non-blocking operation.
908 Severity : enhancement
910 Description: optimized stripe assignment
911 Details : stripe assignments are now made based on ost space available,
912 ost previous usage, and OSS previous usage, in order to try
913 to optimize storage space and networking resources.
915 Severity : enhancement
917 Description: Permanently set tunables
918 Details : All writable /proc/fs/lustre tunables can now be permanently
919 set on a per-server basis, at mkfs time or on a live
922 Severity : enhancement
924 Description: Lustre message v2
925 Details : Add lustre message format v2.
927 Severity : enhancement
929 Description: client OST exclusion list
930 Details : Clients can be started with a list of OSTs that should be
931 declared "inactive" for known non-responsive OSTs.
935 Description: SPEC SFS validation failure on NFS v2 over lustre.
936 Details : Changes the blocksize for regular files to be 2x RPC size,
937 and not depend on stripe size.
939 Severity : enhancement
941 Description: Multiple MD RPCs in flight.
942 Details : Further unserialise some read-only MDS RPCs - learn about intents.
943 To avoid overly-overloading MDS, introduce a limit on number of
944 MDS RPCs in flight for a single client and add /proc controls
945 to adjust this limit.
947 Severity : enhancement
949 Description: client read/write statistics
950 Details : Add client read/write call usage stats for performance
951 analysis of user processes.
952 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
953 file access. extents_stats shows chunk size distribution.
954 extents_stats_per_process show chunk size distribution per
957 Severity : enhancement
959 Description: mds statistics
960 Details : Add detailed mds operations statistics in
961 /proc/fs/lustre/mds/*/stats.
965 Description: Failure of copying files with lustre special EAs.
966 Details : Client side always return success for setxattr call for lustre
967 special xattr (currently only "trusted.lov").
972 Description: Refcount LNET uuids
973 Details : The global LNET uuid list grew linearly with every startup;
974 refcount repeated list entries instead of always adding to
977 Severity : enhancement
979 Description: Dynamic service threads
980 Details : Within a small range, start extra service threads
981 automatically when the request queue builds up.
984 Frequency : mixed-endian client/server environments
986 Description: mixed-endian crashes
987 Details : The new msg_v2 system had some failures in mixed-endian
990 Severity : enhancement
992 Description: Easy OST removal
993 Details : OSTs can be permanently deactivated with e.g. 'lctl
994 conf_param lustre-OST0001.osc.active=0'
996 Severity : enhancement
998 Description: MGS proc entries
999 Details : Added basic proc entries for the MGS showing what filesystems
1002 Severity : enhancement
1004 Description: provide MGS failover
1005 Details : Added config lock reacquisition after MGS server failover.
1007 Severity : enhancement
1009 Description: add Linux 2.4 support
1010 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1014 Description: a large application tries to do I/O to the same resource and dies
1015 in the middle of it.
1016 Details : Check the req->rq_arrival time after the call to
1017 ost_brw_lock_get(), but before we do anything about
1018 processing it & sending the BULK transfer request. This
1019 should help move old stale pending locks off the queue as
1020 quickly as obd_timeout.
1023 Frequency : when an incorrect nid is specified during startup
1025 Description: ptlrpc connect to non-existant node causes kernel crash
1026 Details : LNET can't be re-entered from an event callback, which
1027 happened when we expire a message after the export has been
1028 cleaned up. Instead, hand the zombie cleanup off to another
1031 Severity : enhancement
1033 Description: plain/inodebits lock performance improvement
1034 Details : Grouping plain/inodebits in granted list by their request modes
1035 and bits policy, thus improving the performance of search through
1039 Frequency : only if OST filesystem is corrupted
1041 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1042 Details : for a short time RPCs with bulk IO are in the replay list,
1043 but replay of bulk IOs is unimplemented. If the OST filesystem
1044 is corrupted due to disk cache incoherency and then replay is
1045 started it is possible to trip an assertion. Avoid putting
1046 committed RPCs into the replay list at all to avoid this issue.
1049 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1051 Description: System hang on startup
1052 Details : This bug allowed the liblustre (e.g. catamount) client to
1053 return to the app before handling all startup RPCs. This
1054 could leave the node unresponsive to lustre network traffic
1055 and manifested as a server ptllnd timeout.
1057 Severity : enhancement
1059 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1060 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1061 boolean to control whether to print peer debug info when a
1062 client's RPC times out.
1065 Frequency : only for kernels with patches from Lustre below 1.4.3
1067 Description: Remove old rdonly API
1068 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1071 Frequency : only for devices with external journals
1073 Description: Set external device read-only also
1074 Details : During a commanded failover stop, we set the disk device
1075 read-only while the server shuts down. We now also set any
1076 external journal device read-only at the same time.
1079 Frequency : when upgrading from 1.4 while trying to change parameters
1081 Description: The wrong (new) MDC name was used when setting parameters for
1082 upgraded MDT's. Also allows changing of OSC (and MDC)
1083 parameters if --writeconf is specified at tunefs upgrade time.
1086 Frequency : when setting specific ost indicies
1088 Description: QOS code breaks on skipped indicies
1089 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1090 created with --index need not be sequential.
1092 Severity : enhancement
1094 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1095 Details : The uninit_groups feature works in conjunction with the kernel
1096 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1097 up the pass1 processing of e2fsck. This is a read-only feature
1098 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1099 mount filesystems that have had this feature enabled.
1101 Severity : enhancement
1103 Description: Improve multi-block allocation algorithm to avoid fragmentation
1104 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1105 allocation locality and avoid filesystem fragmentation.
1107 ------------------------------------------------------------------------------
1109 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1111 * Support for kernels:
1112 2.4.21-47.0.1.EL (RHEL 3)
1113 2.6.5-7.283 (SLES 9)
1114 2.6.9-42.0.10.EL (RHEL 4)
1115 2.6.12.6 vanilla (kernel.org)
1116 2.6.16.27-0.9 (SLES 10)
1117 * Recommended e2fsprogs version: 1.39.cfs5
1119 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1123 Frequency : occasional, depends on client load and configuration
1124 Bugzilla : 12181, 12203
1125 Description: data loss for recently-modified files
1127 Details : In some cases it is possible that recently written or created
1128 files may not be written to disk in a timely manner (this should
1129 normally be within 30s unless client IO load is very high).
1130 The problem appears as zero-length files or files that are a
1131 multiple of 1MB in size after a client crash or client eviction
1132 that are missing data at the end of the file.
1134 This problem is more likely to be hit on clients where files are
1135 repeatedly created and unlinked in the same directory, clients
1136 have a large amount of RAM, have many CPUs, the filesystem has
1137 many OSTs, the clients are rebooted frequently, and/or the files
1138 are not accessed by other nodes after being written.
1140 The presence of the problem can be detected by looking at
1141 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1142 smaller than the second (nr_unused) then dirty files will not
1143 be flushed automatically to disk. "sync; sleep 10" should be
1144 run several times on the node before unmounting it to update
1145 Lustre (this is also safe to run on nodes without this problem).
1147 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1148 that can cause this same problem, so customers using that kernel
1149 also need to update the kernel in addition to Lustre. In order
1150 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1152 It is normal that files written just before a client crash (less
1153 than 30s) may not yet have been flushed to disk, even for local
1157 Frequency : frequent on thin XT3 nodes
1159 Description: UUID collision on thin XT3 Linux nodes
1160 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1161 randomly, since we relied on an insufficiently-seeded PRNG.
1166 Description: OSS hangs after "All ost request buffers busy"
1167 Details : A deadlock between quota and journal operations caused OSS
1168 hangs after printing "All ost request buffers busy."
1171 Frequency : always on liblustre builds
1173 Description: Cleanup compiler warnings on liblustre
1176 Frequency : always on liblustre builds on XT3
1178 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1182 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1183 > 32000 subdirectories
1184 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1185 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1186 subdirectory count crosses 32000. This will aid e2fsck to
1187 correctly handle more than 32000 subdirectories.
1190 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1192 Description: System hang on startup
1193 Details : This bug allowed the liblustre (e.g. catamount) client to
1194 return to the app before handling all startup RPCs. This
1195 could leave the node unresponsive to lustre network traffic
1196 and manifested as a server ptllnd timeout.
1198 Severity : enhancement
1200 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1201 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1202 boolean to control whether to print peer debug info when a
1203 client's RPC times out.
1208 Description: make O_SYNC working on 2.6 kernels
1209 Details : 2.6 kernels use different method for mark pages for write,
1210 so need add a code to lustre for O_SYNC work.
1215 Description: Failure to close file and release space on NFS
1216 Details : Put inode details into lock acquired in ll_intent_file_open.
1217 Use mdc_intent_lock in ll_intent_open to properly
1218 detect all kind of errors unhandled by mdc_enqueue.
1223 Description: proc file read during shutdown sometimes raced obd removal,
1225 Details : Add lock to prevent obd access after proc file removal.
1228 Frequency : Only for files larger than 4GB on 32-bit clients.
1230 Description: improperly doing page alignment of locks
1231 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1232 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1237 Description: RPCs being resent when they shouldn't be
1238 Details : Some RPCs that should not be resent are being resent. This
1239 can cause inconsistencies in the RPC state machine. Do not
1240 resend such requests.
1243 Frequency : rare, only with NFS export
1245 Description: Crash on NFS re-export node
1246 Details : under very unusual load conditions an assertion is hit in
1247 ll_intent_file_open()
1250 Frequency : only if OST filesystem is corrupted
1252 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1253 Details : for a short time RPCs with bulk IO are in the replay list,
1254 but replay of bulk IOs is unimplemented. If the OST filesystem
1255 is corrupted due to disk cache incoherency and then replay is
1256 started it is possible to trip an assertion. Avoid putting
1257 committed RPCs into the replay list at all to avoid this issue.
1262 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1263 Details : Large single O_DIRECT read and write calls can fail to allocate
1264 a sufficiently large buffer to process the request. In case of
1265 allocation failure the allocation is retried with a smaller
1266 buffer and broken into smaller requests.
1268 Severity : enhancement
1270 Description: Add -o localflock option to simulate old noflock behaviour.
1271 Details : This will achieve local-only flock/fcntl locks coherentness.
1276 Description: versioning check is incomplete
1277 Details : Checking the version difference of client vs. server, report
1278 error if the gap is too big.
1283 Description: add support PG_writeback bit
1284 Details : add support for PG_writeback bit for Lustre, for more carefull
1285 work with page cache in 2.6 kernel. This also fix some deadlocks
1286 and remove hack for work O_SYNC with 2.6 kernel.
1288 Severity : enhancement
1290 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1291 Details : The uninit_groups feature works in conjunction with the kernel
1292 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1293 up the pass1 processing of e2fsck. This is a read-only feature
1294 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1295 mount filesystems that have had this feature enabled.
1297 Severity : enhancement
1299 Description: Improve multi-block allocation algorithm to avoid fragmentation
1300 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1301 allocation locality and avoid filesystem fragmentation.
1303 ------------------------------------------------------------------------------
1305 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1307 * Support for kernels:
1308 2.6.9-42.0.3.EL (RHEL 4)
1309 2.6.5-7.276 (SLES 9)
1310 2.4.21-47.0.1.EL (RHEL 3)
1311 2.6.12.6 vanilla (kernel.org)
1312 2.6.16.21-0.8 (SLES10)
1313 * Recommended e2fsprogs version: 1.39.cfs2-0
1315 * The backwards-compatible /proc/sys/portals symlink has been removed
1316 in this release. Before upgrading, please ensure that you change
1317 any configuration scripts or /etc/sysctl.conf files that access
1318 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1319 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1320 in advance of the upgrade on any system running Lustre 1.4.6 or
1321 newer, since /proc/sys/lnet was added in that version.
1322 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1326 Frequency : only when quota is used
1328 Description: avoid scanning export list for quota master
1329 Details : Change the algorithms to avoid scanning export list in order
1330 to improve the efficiency.
1333 Frequency : MDS failover only, very rarely
1335 Description: "went back in time" messages on mds failover
1336 Details : The greatest transno may be lost when the current operation
1337 finishes with an error (transno==0) and the client's last_rcvd
1338 record is over-written. Save the greatest transno in the
1339 mds_last_transno for this case.
1342 Frequency : always for specific kernels and striping counts
1344 Description: client may get "Matching packet too big" without ACL support
1345 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1346 when trying to access files in certain configurations. The
1347 clients should in fact be denied when mounting because they do
1348 not understand ACLs.
1351 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1353 Description: many clients connecting with IO in progress causes connect timeouts
1354 Details : Avoid synchronous journal commits to avoid delays caused by many
1355 clients connecting/disconnecting when bulk IO is in progress.
1356 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1357 OST_IO_PORTAL to avoid delays behind potentially many pending
1361 Frequency : occasionally with multiple writers to a single file
1363 Description: shared writes to file may result in wrong size reported by stat()
1364 Details : Allow growing of kms when extent lock is cancelled
1367 Frequency : always with random mmap IO to multi-striped file
1369 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1370 Details : Only if the hole is at the end of OST object so that kms is too
1371 small. Fix is to increase kms accordingly in ll_nopage.
1374 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1376 Description: writes to a missing object would leak memory on the OST
1377 Details : If there is an inconsistency between the MDS and OST filesystems,
1378 such that the MDS references an object that doesn't exist, writes
1379 to that object will leak memory due to incorrect cleanup in the
1380 error handling path, eventually running out of memory on the OST.
1385 Description: Creating too long symlink causes lustre errors
1386 Details : Check symlink and name lengths before sending requests to MDS.
1389 Frequency : only if flock is enabled (not on by default)
1391 Description: posix locks not released on fd closure on 2.6.9+
1392 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1393 kernels, this caused such locks not to be released on fd close and
1394 then assertions on fs unmount about still used locks.
1397 Frequency : MDS failover only, very rarely
1399 Description: clients may get ASSERTION(granted_lock != NULL)
1400 Details : When request was taking a long time, and a client was resending
1401 a getattr by name lock request. The were multiple lock requests
1402 with the same client lock handle and
1403 mds_getattr_name->fixup_handle_for_resent_request found one of the
1404 lock handles but later failed with ASSERTION(granted_lock != NULL).
1409 Description: handle->h_buffer_credits > 0, assertion failure
1410 Details : h_buffer_credits is zero after truncate, causing assertion
1411 failure. This patch extends the transaction or creates a new
1415 Frequency : NFS re-export or patchless client
1416 Bugzilla : 11179, 10796
1417 Description: Crash on NFS re-export node (__d_move)
1418 Details : We do not want to hash the dentry if we don't have a lock.
1419 But if this dentry is later used in d_move, we'd hit uninitialised
1420 list head d_hash, so we just do this to init d_hash field but
1421 leave dentry unhashed.
1424 Frequency : NFS re-export or patchless client
1426 Description: NFS exports has problem with symbolic link
1427 Details : lustre client didn't properly install dentry when re-exported
1428 to NFS or running patchless client.
1431 Frequency : NFS re-export or patchless client
1433 Description: Various nfs/patchless fixes.
1434 Details : fixes reuse disconected alias for lookup process - this fixes
1435 warning "find_exported_dentry: npd != pd",
1436 fix permission error with open files at nfs.
1437 fix apply umask when do revalidate.
1440 Frequency : occasional
1442 Description: Crash on NFS re-export node
1443 Details : calling clear_page() on the wrong pointer triggered oops in
1444 generic_mapping_read().
1447 Frequency : rarely, using O_DIRECT IO
1449 Description: unaligned directio crashes client with LASSERT
1450 Details : check for unaligned buffers before trying any requests.
1453 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1455 Description: stale data returned from RAID cache
1456 Details : If only a small amount of IO is done to the RAID device before
1457 reading it again it is possible to get stale data from the RAID
1458 cache instead of reading it from disk.
1461 Frequency : always for sles10 kernel
1463 Description: sles10 support
1464 Details : ll_follow_link: compile fixes and using of nd_set_link
1465 under newer kernels.
1468 Frequency : depends on arch, kernel and compiler version, always on sles10
1471 Description: recursive or deep enough symlinks cause stack overflow
1472 Details : getting rid of large stack-allocated variable in
1476 Frequency : depends on hardware
1478 Description: lustre write performance loss in the SLES10 kernel
1479 Details : the performance loss is caused by using of write barriers in the
1480 ext3 code. The SLES10 kernel turns barrier support on by
1481 default. The fix is to undo that change for ldiskfs.
1483 ------------------------------------------------------------------------------
1485 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1487 * Support for kernels:
1488 2.6.9-42.0.3EL (RHEL 4)
1489 2.6.5-7.276 (SLES 9)
1490 2.4.21-47.0.1.EL (RHEL 3)
1491 2.6.12.6 vanilla (kernel.org)
1495 Frequency : quota enabled and large files being deleted
1497 Description: releasing more than 4GB of quota at once hangs OST
1498 Details : If a user deletes more than 4GB of files on a single OST it
1499 will cause the OST to spin in an infinite loop. Release
1500 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1505 Description: statfs data retrieved from /proc may be stale or zero
1506 Details : When reading per-device statfs data from /proc, in the
1507 {kbytes,files}_{total,free,avail} files, it may appear
1508 as zero or be out of date.
1511 Frequency : systems with MD RAID1 external journal devices
1513 Description: lconf's call to blkid is confused by RAID1 journal devices
1514 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1515 of returning all block devices that match the journal UUID.
1518 Frequency : always, for aggregate stripe size over 4GB
1520 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1521 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1522 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1525 Frequency : always if "lfs find" used on a local file/directory
1527 Description: "lfs find" segfaults if used on a local file/directory
1528 Details : The case where a directory component was not specified wasn't
1529 handled correctly. Handle this properly.
1532 Frequency : always on ppc64
1534 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1535 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1536 perform endianness conversion.
1539 Frequency : rarely (truncate to non-zero file size after write under load)
1540 Bugzilla : 10730, 10687
1541 Description: Files padded with zeros to next 4K multiple
1542 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1543 it is possible that files that are truncated to a non-zero size
1544 immediately after being written are filled with zero bytes beyond
1545 the truncated size. No file data is lost.
1547 Severity : enhancement
1549 Description: Allow recovery/failover for liblustre clients.
1550 Details : liblustre clients were unaware of failover configurations until
1553 Severity : enhancement
1555 Description: user file locks should fail when not mounting with flock option
1556 Details : Set up an error-returning stub in ll_file_operations.lock field
1557 to prevent incorrect behaviour when client is mounted without
1558 flock option. Also, set up properly f_op->flock field for
1562 Frequency : always on ia64
1564 Description: "lfs df" loops on printing out MDS statfs information
1565 Details : The obd_ioctl_data was not initialized and in some systems
1566 this caused a failure during the ioctl that did not return
1567 an error. Initialize the struct and return an error on failure.
1570 Frequency : SLES 9 only
1572 Description: Error of copying files with lustre special EAs as root
1573 Details : Client side always return success for setxattr call for lustre
1574 special xattr (currently only "trusted.lov").
1577 Frequency : rarely on clusters with both ia64+i386 clients
1579 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1580 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1581 clients are doing concurrent writes to the same file it is
1582 possible that the ia64 clients may LASSERT because the OST
1583 extent locks are not PAGE_SIZE aligned. Ensure that grown
1584 locks are always aligned on the request boundary.
1587 Frequency : specific use, occasional
1589 Description: Overwriting in use executable truncates on-disk binary image
1590 Details : If one node attempts to overwrite an executable in use by
1591 another node, we now correctly return ETXTBSY instead of
1592 truncating the file.
1594 Severity : enhancement
1596 Description: Async OSC create to avoid the blocking unnecessarily.
1597 Details : If a OST has no remain object, system will block on the creating
1598 when need to create a new object on this OST. Now, ways use
1599 pre-created objects when available, instead of blocking on an
1600 empty osc while others are not empty. If we must block, we block
1601 for the shortest possible period of time.
1606 Description: chmod on Lustre root is propagated to other clients
1607 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1608 invalid by the follow_mount time.
1611 Frequency : liblustre clients only
1613 Description: Race in 'instant cancel' lock handling could lead to such locks
1614 never to be granted in case of SMP MDS
1615 Details : Do not destroy not yet granted but cbpending locks in
1619 Frequency : replay/resend of open
1621 Description: non null lock assetion failure in mds_intent_policy
1622 Details : Trying to replay/resend lockless open requests resulted in
1623 mds_open() returning 0 with no lock. Now it sets a flag if
1624 a lock is going to be returned.
1626 Severity : enhancement
1628 Description: Checksum enhancements
1629 Details : New checksum enhancements allow for resending RPCs that failed
1632 Severity : enhancement
1634 Description: Tunables on number of dirty pages in cacche
1635 Details : Allow to set limit on number of dirty pages cached.
1640 Description: client crash on unmount - lock still has references
1641 Details : In some error handling cases it was possible to leak a lock
1642 reference on a client while accessing a file. This was not
1643 harmful to the client during operation, but would cause the
1644 client to crash when the filesystem is unmounted.
1647 Frequency : specific case, rare
1649 Description: ETXTBSY on mds though file not in use by client
1650 Details : ETXTBSY is no longer incorrectly returned when attempting to
1651 chmod or chown a directory that the user previously tried to
1652 execute or a currently-executing binary.
1655 Frequency : extremely rare except on liblustre-based clients
1657 Description: Lustre space not freed when files are deleted
1658 Details : Clean up open-unlinked files after client eviction. Previously
1659 the unlink was skipped and the files remained as orphans.
1664 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1665 Details : In some cases it was possible to send a blocking callback to
1666 a client doing a glimpse, even though that client didn't get
1667 a lock granted. When the glimpse lock is cancelled on the OST
1668 the freed lock is left on the waiting list and corrupted the list.
1671 Frequency : all core dumps
1673 Description: Broke core dumps to lustre
1674 Details : Negative dentry may be unhashed if parent does not have UPDATE
1675 lock, but some callers, e.g. do_coredump, expect dentry to be
1676 hashed after successful create, hash it in ll_create_it.
1678 ------------------------------------------------------------------------------
1680 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1682 * Support for kernels:
1683 2.6.9-42.0.2.EL (RHEL 4)
1684 2.6.5-7.276 (SLES 9)
1685 2.4.21-40.EL (RHEL 3)
1686 2.6.12.6 vanilla (kernel.org)
1690 Frequency : always on RHEL 3
1692 Description: Number of open files grows over time
1693 Details : The number of open files grows over time, whether or not
1694 Lustre is started. This was due to a filp leak introduced
1695 by one of our kernel patches.
1697 ------------------------------------------------------------------------------
1699 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1701 * Support for kernels:
1702 2.6.9-42.EL (RHEL 4)
1703 2.6.5-7.276 (SLES 9)
1704 2.4.21-40.EL (RHEL 3)
1705 2.6.12.6 vanilla (kernel.org)
1710 Bugzilla : 5719, 9635, 9792, 9684
1711 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1712 Details : If a server is under heavy load and cannot reply to new
1713 connection requests before the client resends the (re)connect,
1714 the connection handling code can behave badly if two service
1715 threads are concurrently handing separate (re)connections from
1716 the same client. Add better locking to the connection handling
1717 code, and ensure that only a single connection will be processed
1718 for a given client UUID, even if the lock is dropped.
1720 Severity : enhancement
1722 Description: add TCP zero-copy support to kernel
1723 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1724 sends if the hardware supports scatter-gather and checksumming.
1725 This allows socklnd to do client-write and server-read more
1726 efficiently and reduce CPU utilization from skbuf copying.
1729 Frequency : only if NFS exporting from client
1731 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1732 Details : If setting attributes on a file created under NFS that had
1733 never been opened it would be possible to oops the client
1734 if the file had no objects.
1737 Frequency : always for liblustre
1739 Description: liblustre client does MDS+OSTs setattr RPC for each write
1740 Details : When doing a write from a liblustre client, the client
1741 incorrectly issued an RPC to the MDS and each OST the file was
1742 striped over in order to update the timestamps. When writing
1743 with small chunks and many clients this could overwhelm the MDS
1744 with RPCs. In all cases it would slow down the write because
1745 these RPCs are unnecessary.
1747 Severity : enhancement
1749 Description: allow number of MDS service threads to be changed at module load
1750 Details : It is now possible to change the number of MDS service threads
1751 running. Adding "options mds mds_num_threads={N}" to the MDS's
1752 /etc/modprobe.conf will set the number of threads for the next
1753 time Lustre is restarted (assuming the "mds" module is also
1754 reloaded at that time). The default number of threads will
1755 stay the same, 32 for most systems.
1760 Description: OST crash if filesystem is unformatted or corrupt
1761 Details : If an OST is started on a device that has never been formatted
1762 or if the filesystem is corrupt and cannot even mount then the
1763 error handling cleanup routines would dereference a NULL pointer.
1768 Description: NULL pointer deref in llap_from_page.
1769 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1770 incorrect) mapping if the page was truncated/reclaimed while it was
1771 searched for. Check for this condition and skip such pages when
1772 doing readahead. Introduce extra check to llap_from_page() to
1773 verify page->mapping->host is non-NULL (so page is not anonymous).
1776 Frequency : Sometimes when using sys_sendfile
1778 Description: "page not covered by a lock" warnings from ll_readpage
1779 Details : sendfile called ll_readpage without right page locks present.
1780 Now we introduced ll_file_sendfile that does necessary locking
1781 around call to generic_file_sendfile() much like we do in
1785 Frequency : with certain MDS communication failures at client mount time
1787 Description: NULL pointer deref after failed client mount
1788 Details : a client connection request may delayed by the network layer
1789 and not be sent until after the PTLRPC layer has timed out the
1790 request. If the client fails the mount immediately it will try
1791 to clean up before the network times out the request. Add a
1792 reference from the request import to the obd device and delay
1793 the cleanup until the network drops the request.
1796 Frequency : occasionally during client (re)connect
1798 Description: assertion failure during client (re)connect
1799 Details : processing a client connection request may be delayed by the
1800 client or server longer than the client connect timeout. This
1801 causes the client to resend the connection request. If the
1802 original connection request is replied in this interval, the
1803 client may trip an assertion failure in ptlrpc_connect_interpret()
1804 which thought it would be the only running connect process.
1807 Frequency : only with obd_echo servers and clients that are rebooted
1809 Description: kernel BUG accessing uninitialized data structure
1810 Details : When running an obd_echo server it did not start the ping_evictor
1811 thread, and when a client was evicted an uninitialized data
1812 structure was accessed. Start the ping_evictor in the RPC
1813 service startup instead of the OBD startup.
1815 Severity : enhancement
1816 Bugzilla : 10193 (patchless)
1817 Description: Remove dependency on various unexported kernel interfaces.
1818 Details : No longer need reparent_to_init, exit_mm, exit_files,
1819 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1822 Frequency : rare (only users of deprecated and unsupported LDAP config)
1824 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1825 Details : LDAP apparently contains 'lustreName' attributes instead of
1826 'name'. A simple remapping of the name is sufficient.
1829 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1831 Description: waiting_locks_callback trips kernel BUG if client is evicted
1832 Details : Running with the dump_on_timeout debug flag turned on makes
1833 it possible that the waiting_locks_callback() can try to dump
1834 the Lustre kernel debug logs from an interrupt handler. Defer
1835 this log dumping to the expired_lock_main() thread.
1837 Severity : enhancement
1839 Description: Support NFS exporting on 2.6 kernels.
1840 Details : Implement non-rawops metadata methods for NFS server to use without
1841 changing NFS server code.
1844 Frequency : very rare (synthetic metadata workload only)
1846 Description: two racing renames might cause an MDS thread to deadlock
1847 Details : Running the "racer" program may cause one MDS thread to rename
1848 a file from being the source of a rename to being the target of
1849 a rename at exactly the same time that another thread is doing
1850 so, and the second thread has already enqueued these locks after
1851 doing a lookup of the target and is trying to relock them in
1852 order. Ensure that we don't try to re-lock the same resource.
1855 Frequency : only very large systems with liblustre clients
1857 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1858 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1859 all OSTs in parallel. This allows the eviction of stale liblustre
1860 clients to proceed much faster than if they were done in series,
1861 and also offers similar improvements for other set_info RPCs.
1866 Description: excessive CPU usage during initial read phase on client
1867 Details : During the initial read phase on a client, it would agressively
1868 retry readahead on the file, consuming too much CPU and impacting
1869 performance (since 1.4.5.8). Improve the readahead algorithm
1870 to avoid this, and also improve some other common cases (read
1871 of small files in particular, where "small" is files smaller than
1872 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1877 Description: MDS crash when receiving packet with unknown intent.
1878 Details : Do not LBUG in unknown intent case, just return -EFAULT
1880 Severity : enhancement
1881 Bugzilla : 9293, 9385
1882 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1883 Details : Do not serialize getattr (non-intent version) and statfs.
1886 Frequency : occasional, when OST network is overloaded/intermittent
1888 Description: client evicted by OST after bulk IO timeout
1889 Details : If a client sends a bulk IO request (read or write) the OST
1890 may evict the client if it is unresposive to its data GET/PUT
1891 request. This is incorrect if the network is overloaded (takes
1892 too long to transfer the RPC data) or dropped the OST GET/PUT
1893 request. There is no need to evict the client at all, since
1894 the pinger and/or lock callbacks will handle this, and the
1895 client can restart the bulk request.
1898 Frequency : Always when mmapping file with no objects
1900 Description: client crashes when mmapping file with no objects
1901 Details : Check that we actually have objects in a file before doing any
1902 operations on objects in ll_vm_open, ll_vm_close and
1908 Description: Request leak when working with deleted CWD
1909 Details : Introduce advanced request refcount tracking for requests
1910 referenced from lustre intent.
1912 Severity : Enhancement
1914 Description: Cache open file handles on client.
1915 Details : MDS now will return special lock along with openhandle, if
1916 requested and client is allowed to hold openhandle, even if unused,
1917 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1918 closing files for every read/write openration.
1920 Severity : Enhancement
1922 Description: Cache open negative dentries on client when possible.
1923 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1924 negative dentries on lock revocation.
1929 Description: Remounting a client read-only wasn't possible with a zconf mount
1930 Details : It wasn't possible to remount a client read-only with llmount.
1932 Severity : enhancement
1933 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1934 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1935 MPICH is included to add Lustre-specific ADIO interfaces.
1936 This is based closely on the UFS ADIO layer and only differs
1937 in file creation, in order to allow the OST striping to be set.
1938 This is user-contributed code and not supported by CFS.
1943 Description: extended inode attributes (immutable, append-only) work improperly
1944 when 2.4 and 2.6 kernels are used on client/server or vice versa
1945 Details : Introduce kernel-independent values for these flags.
1947 Severity : enhancement
1950 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1951 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1952 granularity. Now, Fractional MB granularity is be supported,
1953 this is very useful for low memory system.
1955 Severity : enhancement
1957 Description: Getattr by fid
1958 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1959 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1960 have locks and avoids getattr rpc requests for it on every stat.
1963 Frequency : Always, for filesystems larger than 2TB
1965 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1966 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1967 filesystems larger than 2TB to be mounted with mballoc enabled.
1970 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1972 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1973 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1974 filesystems larger than 2TB is mounted then the kernel will
1975 silently corrupt the start of the filesystem. CONFIG_LBD is
1976 enabled for all CFS-supported kernels, but the possibility of
1977 this happening with a modified kernel config exists.
1979 Severity : enhancement
1981 Description: add client O_DIRECT support for 2.6 kernels
1982 Details : It is now possible to do O_DIRECT reads and writes to files
1983 in the Lustre client mountpoint on 2.6 kernel clients.
1985 Severity : enhancement
1987 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1988 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1989 parallel, not waiting for response from every OST before sending
1990 a rpc to the next OST.
1995 Description: setattr vs write race when updating file timestamps
1996 Details : Client processes that update a file timestamp into the past
1997 right after writing to the file (e.g. tar) it is possible that
1998 the updated file modification time can be reset to the current
1999 time due to a race between processing the setattr and write RPC.
2001 Severity : enhancement
2003 Description: Bring 'lfs find' closer in line with regular Linux find.
2004 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2005 -print0 options and obtains all the needed info through the lustre
2008 Severity : enhancement
2010 Description: support up to 1024 configured devices on one node
2011 Details : change obd_dev array from statically allocated to dynamically
2012 allocated structs as they are first used to reduce memory usage
2017 Description: Flush dirty partially truncated pages during truncate
2018 Details : Immediatelly flush partially truncated pages in filter_setattr,
2019 this way we completely avoid having any pages in page cache on OST
2020 and can retire ugly workarounds during writes to flush such pages.
2025 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2026 Details : protect inode from truncation within vfs_unlink() context
2027 just take a reference before calling vfs_unlink() and release it
2028 when parent's i_sem is free.
2033 Description: last_id value checked outside lock on OST caused LASSERT failure
2034 Details : If there were multiple MDS->OST object precreate requests in
2035 flight, it was possible that the OST's last object id was checked
2036 outside a lock and incorrectly tripped an assertion. Move checks
2037 inside locks, and discard old precreate requests.
2040 Frequency : always, if extents are used on OSTs
2042 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2043 Details : OSTs using the extents format would not zero the high 16 bits of
2044 the index physical block number. This is not a problem for any
2045 OST filesystems smaller than 16TB, and no kernels support ext3
2046 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2047 new/modified files) and can be fixed for existing filesystems
2048 with e2fsprogs-1.39-cfs1.
2053 Description: import connection selection may be incorrect if timer wraps
2054 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2055 import connections to be ignored if the 32-bit jiffies counter
2056 wraps. Use a 64-bit jiffies counter.
2059 Frequency : very large clusters immediately after boot
2061 Description: LNET request buffers exhausted under heavy short-term load
2062 Details : If a large number of client requests are generated on a service
2063 that has previously never seen so many requests it is possible
2064 that the request buffer growth cannot keep up with the spike in
2065 demand. Instead of dropping incoming requests, they are held in
2066 the LND until the RPC service can accept more requests.
2069 Frequency : Sometimes during replay
2071 Description: Assertion failure in ll_local_open after replay.
2072 Details : If replay happened on an open request reply before we were able
2073 to set replay handler, reply will become not swabbed tripping the
2074 assertion in ll_local_open. Now we set the handler right after
2075 recognising of open request
2078 Frequency : very rare
2080 Description: kernel reports "badness in vsnprintf"
2081 Details : Reading from the "recovery_status" /proc file in small chunks
2082 may cause a negative length in lprocfs_obd_rd_recovery_status()
2083 call to vsnprintf() (which is otherwise harmless). Exit early
2084 if there is no more space in the output buffer.
2086 Severity : enhancement
2088 Description: clear OBD RPC statistics by writing to them
2089 Details : It is now possible to clear the OBD RPC statistics by writing
2090 to the "stats" file.
2095 Description: Client mtime is not the same on different clients after utimes
2096 Details : In some cases, the client was using the utimes() syscall on
2097 a file cached on another node. The clients now validate the
2098 ctime from the MDS + OSTs to determine which one is right.
2103 Description: Inability to activate failout mode
2104 Details : lconf script incorrectly assumed that in pythong string's numeric
2105 value is used in comparisons.
2108 Frequency : always with multiple stripes per file
2110 Description: Inefficient object allocation for mutli-stripe files
2111 Details : When selecting which OSTs to stripe files over, for files with
2112 a stripe count that divides evenly into the number of OSTs,
2113 the MDS is always picking the same starting OST for each file.
2114 Return the OST selection heuristic to the original design.
2119 Description: mount failures may take full timeout to return an error
2120 Details : Under some heavy load conditions it is possible that a
2121 failed mount can wait for the full obd_timeout interval,
2122 possibly several minutes, before reporting an error.
2123 Instead return an error as soon as the status is known.
2125 Frequency : quota enabled and large files being deleted
2127 Description: releasing more than 4GB of quota at once hangs OST
2128 Details : If a user deletes more than 4GB of files on a single OST it
2129 will cause the OST to spin in an infinite loop. Release
2130 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2135 Description: statfs data retrieved from /proc may be stale or zero
2136 Details : When reading per-device statfs data from /proc, in the
2137 {kbytes,files}_{total,free,avail} files, it may appear
2138 as zero or be out of date.
2141 Frequency : systems with MD RAID1 external journal devices
2143 Description: lconf's call to blkid is confused by RAID1 journal devices
2144 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2145 of returning all block devices that match the journal UUID.
2148 Frequency : always, for aggregate stripe size over 4GB
2150 Description: assertion fails when trying to use 4GB stripe size
2151 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2152 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2155 Frequency : always on ppc64
2157 Description: the first write on an ext3 filesystem with mballoc got stuck
2158 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2159 perform endianness conversion.
2161 ------------------------------------------------------------------------------
2163 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2165 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2166 INTEROPERATE with older versions automatically. Please read the
2167 user documentation before upgrading any part of a live system.
2168 * WARNING: Lustre networking configuration changes are required with
2169 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2172 * Support for kernels:
2173 2.6.9-22.0.2.EL (RHEL 4)
2174 2.6.5-7.244 (SLES 9)
2175 2.6.12.6 vanilla (kernel.org)
2178 Severity : enhancement
2179 Bugzilla : 7981/8208
2180 Description: Introduced Lustre Networking (LNET)
2181 Details : LNET is new networking infrastructure for Lustre, it includes
2182 a reorganized network configuration mode (see the user
2183 documentation for full details) as well as support for routing
2184 between different network fabrics. Lustre Networking Devices
2185 (LNDs) for the supported network fabrics have also been
2186 created for this new infrastructure.
2188 Severity : enhancement
2189 Description: Introduced Access control lists
2190 Details : clients can set ACLs on files and directories in order to have
2191 more fine-grained permissions than the standard Unix UGO+RWX.
2192 The MDS must be started with the "-o acl" mount option.
2194 Severity : enhancement
2195 Description: Introduced filesystem quotas
2196 Details : Administrators may now establish per-user quotas on the
2199 Severity : enhancement
2201 Description: Configuration change for the XT3
2202 The PTLLND is now used to run Lustre over Portals on the XT3
2203 The configure option(s) --with-cray-portals are no longer used.
2204 Rather --with-portals=<path-to-portals-includes> is used to
2205 enable building on the XT3. In addition to enable XT3 specific
2206 features the option --enable-cray-xt3 must be used.
2211 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2212 Details : A race between allocating a new llcd and re-getting the llcd_lock
2213 allowed another thread to grab newly-allocated llcd.
2215 Severity : enhancement
2217 Description: 2.6 OST async journal commit and locking fix to improve performance
2218 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2219 2.6 kernels are now async as they already were in 2.4 kernels so
2220 that they can commit concurrently with the network bulk transfer.
2221 For block-allocated files the filter allocation semaphore is held
2222 to avoid filesystem fragmentation during allocation. BKL lock
2223 removed for 2.6 xattr operations where it is no longer needed.
2228 Description: lconf incorrectly determined whether two IP networks could talk
2229 Details : In some more complicated routing and multiple-network
2230 configurations, lconf will avoid trying to make a network
2231 connection to a disjoint part of the IP space. It was doing the
2232 math incorrectly for one set of cases.
2237 Description: Fix for potential infinite loop processing records in an llog.
2238 Details : If an llog record is corrupted/zeroed, it is possible to loop
2239 forever in llog_process(). Validate the llog record length
2240 and skip the remainder of the block on error.
2243 Frequency : occasional (liblustre only)
2245 Description: liblustre could not open files whose last component is a symlink
2246 Details : sysio_path_walk() would incorrectly pass the open intent to
2247 intermediate path components.
2250 Frequency : rare (liblustre only with non-standard tuning)
2251 Bugzilla : 7201 (7350)
2252 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2253 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2254 after it has been released, only do this for referenced locks
2256 Severity : enhancement
2258 Description: specifying an (invalid) directory default stripe_size of -1
2259 would reset the directory default striping
2260 Details : stripe_size -1 was used internally to signal directory stripe
2261 removal, now use "all default" to signal dir stripe removal
2262 as a directory striping of "all default" is not useful
2265 Frequency : common for large clusters running liblustre clients
2267 Description: doing an ls when liblustre clients are running is slow
2268 Details : sending a glimpse AST to a liblustre client waits for every AST
2269 to time out, as liblustre clients will not respond. Since they
2270 cannot cache data we refresh the OST lock LVB from disk instead.
2272 Severity : enhancement
2274 Description: doing an ls at the same time as file IO can be slow
2275 Details : enqueue and other "small" requests can be blocked behind many
2276 large IO requests. Create a new OST IO portal for non-IO
2277 requests so they can be processed faster.
2280 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2282 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2283 created file with mode 000
2284 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2285 allow an ftruncate() as a non-root user to a file with mode 000.
2286 Lustre now respects this flag to disable mode checks when
2287 truncating a file owned by the user
2290 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2293 Description: Revoking locks from clients that went dead or catatonic might take
2295 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2296 cancellation of such locks instant on servers without waiting for
2297 any reply from clients. Clients drops these locks when cancel
2298 notification from server is received without replying.
2301 Frequency : liblustre-only, when liblustre client dies or becomes busy
2303 Description: Doing ls on Linux clients can take a long time with active
2305 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2306 granting such locks to it in the first place if possible. Locks
2307 are taken by proxy on the OST during the read or write and
2308 dropped immediately afterward. Add connect flags handling, do
2309 not grant locks to liblustre clients for glimpse ASTs.
2311 Severity : enhancement
2313 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2314 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2315 sequential reads and behaves badly with applications doing random
2316 reads. Improve it by reading ahead at least read region, and
2317 avoiding excessive large RPC for small reads.
2319 Severity : enhancement
2321 Description: Creating more than 1000 files for a single job may cause a load
2322 imbalance on the OSTs if there are also a large number of OSTs.
2323 Details : qos_prep_create() uses an OST index reseed value that is an
2324 even multiple of the number of available OSTs so that if the
2325 reseed happens in the middle of the object allocation it will
2326 still utilize the OSTs as uniformly as possible.
2331 Description: OST or MDS may oops in ping_evictor_main()
2332 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2333 but doesn't restart at beginning of obd_exports_timed list
2336 Severity : enhancement
2338 Description: improve by-nid export eviction on the MDS and OST
2339 Details : allow multiple exports with the same NID to be evicted at one
2340 time without re-searching the exports list.
2343 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2345 Description: MDS may oops in groups_free()
2346 Details : in rare race conditions a newly allocated group_info struct is
2347 freed again, and this can be NULL. The 2.4 compatibility code
2348 for groups_free() checked for a NULL pointer, but 2.6 did not.
2351 Frequency : common for liblustre clients doing little filesystem IO
2352 Bugzilla : 9352, 7313
2353 Description: server may evict liblustre clients accessing contended locks
2354 Details : if a client is granted a lock or receives a completion AST
2355 with a blocking AST already set it would not reply to the AST
2356 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2359 Frequency : lfs setstripe, only systems with more than 160 OSTs
2361 Description: unable to set striping with a starting offset beyond OST 160
2362 Details : llapi_create_file() incorrectly limited the starting stripe
2363 index to the maximum single-file stripe count.
2366 Frequency : LDAP users only
2368 Description: lconf did not handle in-kernel recovery with LDAP properly
2369 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2371 Severity : enhancement
2373 Description: bind OST threads to NUMA nodes to improve performance
2374 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2375 node and do their allocations there to localize memory access
2377 Severity : enhancement
2379 Description: llmount can determine client NID directly from Myrinet (GM)
2380 Details : the client NID code from gmnalnid was moved directly into
2381 llmount, removing the need to use this or specifying the
2382 client NID explicitly when mounting GM clients with zeroconf
2385 Frequency : if client is started with down MDS
2387 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2388 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2389 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2390 ptlrpc_import_delayed_req() because we didn't check if the
2391 request was interrupted, and we also didn't break out of the
2392 event loop if there was no timeout
2397 Description: data loss during non-page-aligned writes to a single file from
2398 both multiple nodes and multiple threads on one node at same time
2399 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2400 inconsistency led to false short-reads, that were cached and later
2401 used by ->prepare_write() to fill in partially written page,
2402 leading to data loss.
2405 Frequency : always, if lconf --abort_recovery used
2407 Description: lconf --abort_recovery fails with 'Operation not supported'
2408 Details : lconf was attempting to abort recovery on the MDT device and not
2411 Severity : enhancement
2413 Description: remove cleanup logs
2414 Details : replace lconf-generated cleanup logs with lustre internal
2415 cleanup routines. Eliminates the need for client-cleanup and
2418 Severity : enhancement
2420 Description: add support for EAs (user and system) on lustre filesystems
2421 Details : it is now possible to store extended attributes in the Lustre
2422 client filesystem, and with the user_xattr mount option it
2423 is possible to allow users to store EAs on their files also
2425 Severity : enhancement
2427 Description: Add possibility (config option) to show minimal available OST free
2429 Details : When compiled with --enable-mindf configure option, statfs(2)
2430 (and so, df) will return least minimal free space available from
2431 all OSTs as amount of free space on FS, instead of summary of
2432 free spaces of all OSTs.
2434 Severity : enhancement
2436 Description: do not expand extent locks acquired on OST-side
2437 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2438 by server: they are not cached anyway.
2441 Frequency : when mmap is used/binaries executed from Lustre
2443 Description: Unmmap pages before throwing them away from read cache.
2444 Details : llap_shrink cache now attempts to unmap pages before discarding
2445 them (if unmapping failed - do not discard). SLES9 kernel has
2446 extra checks that trigger if this unmapping is not done first.
2451 Description: lconf didn't resolve symlinks before checking to see whether a
2452 given mountpoint was already in use
2455 Frequency : when migrating failover services
2456 Bugzilla : 6395, 9514
2457 Description: When migrating a subset of services from a node (e.g. failback
2458 from a failover service node) the remaining services would
2459 time out and evict clients.
2460 Details : lconf --force (implied by --failover) sets the global obd_timeout
2461 to 5 seconds in order to quickly disconnect, but this caused
2462 other RPCs to time out too quickly. Do not change the global
2463 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2465 Severity : enhancement
2466 Frequency : if MDS is started with down OST
2467 Bugzilla : 9439,5706
2468 Description: Allow startup/shutdown of an MDS without depending on the
2469 availability of the OSTs.
2470 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2471 Add appropriate locking and lov-osc refcounts for safe
2472 cleaning. Add osc abort_inflight calls in case the
2473 synchronize never started.
2476 Frequency : occasional (Cray XT3 only)
2478 Description: root not authorized to access files in CRAY_PORTALS environment
2479 Details : The client process capabilities were not honoured on the MDS in
2480 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2481 been accessed by an authorized user then root was able to access
2482 the file on the local client also. The root user capabilities
2483 are now allowed on the MDS, as this environment has secure UID.
2486 Frequency : occasional
2488 Description: ldiskfs "too long searching" message happens too often
2489 Details : A debugging message (otherwise harmless) prints too often on
2490 the OST console. This has been reduced to only happen when
2491 there are fragmentation problems on the filesystem.
2496 Description: Division by zero in statfs when all OSCs are inactive
2497 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2498 lov_statfs divides by value returned by lov_get_stripecnt().
2502 Bugzilla : 9489, 3273
2503 Description: First write from each client to each OST was only 4kB in size,
2504 to initialize client writeback cache, which caused sub-optimal
2505 RPCs and poor layout on disk for the first writen file.
2506 Details : Clients now request an initial cache grant at (re)connect time
2507 and so that they can start streaming writes to the cache right
2508 away and always do full-sized RPCs if there is enough data.
2509 If the OST is rebooted the client also re-establishes its grant
2510 so that client cached writes will be honoured under the grant.
2515 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2516 Details : Now I/O RPCs go to different portal number and (presumably) fast
2517 lock requests (and glimses) and other RPCs get their own service
2518 threads pool that should be able to service those RPCs
2521 Severity : enhancement
2523 Description: Ability to exchange lustre version between client and servers and
2524 issue warnings at client side if client is too old. Also for
2525 liblustre clients there is ability to refuse connection of too old
2527 Details : New 'version' field is added to connect data structure that is
2528 filled with version info. That info is later checked by server and
2532 Frequency : rare, liblustre only.
2533 Bugzilla : 9296, 9581
2534 Description: Two simultaneous writes from liblustre at offset within same page
2535 might proceed at the same time overwriting eachother with stale
2537 Details : I/O lock withing llu_file_prwv was released too early, before data
2538 actually was hitting the wire. Extended lock-holding time until
2539 server acknowledges receiving data.
2542 Frequency : extremely rare. Never observed in practice.
2544 Description: avoid generating lustre_handle cookie of 0.
2545 Details : class_handle_hash() generates handle cookies by incrementing
2546 global counter, and can hit 0 occasionaly (this is unlikely, but
2547 not impossible, because initial value of cookie counter is
2548 selected randonly). Value of 0 is used as a sentinel meaning
2549 "unassigned handle" --- avoid it. Also coalesce two critical
2550 sections in this function into one.
2552 Severity : enhancement
2554 Description: allow liblustre clients to delegate truncate locking to OST
2555 Details : To avoid overhead of locking, liblustre client instructs OST to
2556 take extent lock in ost_punch() on client's behalf. New connection
2557 flag is added to handle backward compatibility.
2559 Severity : enhancement
2560 Bugzilla : 4928, 7341, 9758
2561 Description: allow number of OST service threads to be specified
2562 Details : a module parameter allows the number of OST service threads
2563 to be specified via "options ost ost_num_threads={N}" in the
2564 OSS's /etc/modules.conf or /etc/modprobe.conf.
2568 Bugzilla : 6146, 9635, 9895
2569 Description: servers crash with bad pointer in target_handle_connect()
2570 Details : In rare cases when a client is reconnecting it was possible that
2571 the connection request was the last reference for that export.
2572 We would temporarily drop the export reference and get a new
2573 one, but this may have been the last reference and the export
2574 was just destroyed. Get new reference before dropping old one.
2576 Severity : enhancement
2577 Frequency : if client is started with failover MDS
2579 Description: Allow multiple MDS hostnames in the mount command
2580 Details : Try to read the configuration from all specified MDS
2581 hostnames during a client mount in case the "primary"
2584 Severity : enhancement
2586 Description: Stop sending data to evicted clients as soon as possible.
2587 Details : Check if the client we are about to send or are sending data to
2588 was evicted already. (Check is done every second of waiting,
2589 for which l_wait_event interface was extended to allow checking
2590 of exit condition at specified intervals).
2593 Frequency : rare, normally only when NFS exporting is done from client
2595 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2597 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2600 Frequency : always, if srand() is called before liblustre initialization
2602 Description: Liblustre uses system PRNG disturbing its usage by user application
2603 Details : Introduce internal to lustre fast and high-quality PRNG for
2604 lustre usage and make liblustre and some other places in generic
2605 lustre code to use it.
2607 Severity : enhancement
2608 Bugzilla : 9477, 9557, 9870
2609 Description: Verify that the MDS configuration logs are updated when xml is
2610 Details : Check if the .xml configuration logs are newer than the config
2611 logs stored on the MDS and report an error if this is the case.
2612 Request --write-conf, or allow starting with --old_conf.
2614 Severity : enhancement
2616 Description: Handle symlinks in the path when checking if Lustre is mounted.
2617 Details : Resolve intermediate symlinks when checking if a client has
2618 mounted a filesystem to avoid duplicate client mounts.
2623 Description: lconf can hit an error exception but still return success.
2624 Details : The lconf command catches the Command error exception at the top
2625 level script context and will exit with the associated exit
2626 status, but doesn't ensure that this exit status is non-zero.
2631 Description: failure of ptlrpc thread startup can cause oops
2632 Details : Starting a ptlrpc service thread can fail if there are a large
2633 number of threads or the server memory is very fragmented.
2634 Handle this without oopsing.
2637 Frequency : always, only if liblustre and non-default acceptor port was used
2639 Description: liblustre cannot connect to servers with non-default acceptor port
2640 Details : tcpnal_set_default_params() was not called and was therefore
2641 ignoring the environment varaible TCPNAL_PORT, as well as other
2642 TCPNAL_ environment variables
2647 Description: two objects could be created on the same OST for a single file
2648 Details : If an OST is down, in some cases it was possible to create two
2649 objects on a single OST for a single file. No problems other
2650 than potential performance impact and spurious error messages.
2654 Bugzilla : 5681, 9562
2655 Description: Client may oops in ll_unhash_aliases
2656 Details : Client dcache may become inconsistent in race condition.
2657 In some cases "getcwd" can fail if the current directory is
2663 Description: Inode refcounting problems in NFS export code
2664 Details : link_raw functions used to call d_instantiate without obtaining
2665 extra inode reference first.
2669 Bugzilla : 9942, 9903
2670 Description: Referencing freed requests leading to crash, memleaks with NFS.
2671 Details : We used to require that call to ll_revalidate_it was always
2672 followed by ll_lookup_it. Also with revalidate_special() it is
2673 possible to call ll_revalidate_it() twice for the same dentry
2674 even if first occurence returned success. This fix changes semantic
2675 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2676 reference on a request referred from the intent.
2677 ll_intent_release() then releases such a request.
2680 Frequency : rare, normally benchmark loads only
2682 Description: unlinked inodes were kept in memory on the client
2683 Details : If a client is repeatedly creating and unlinking files it
2684 can accumulate a lot of stale inodes in the inode slab cache.
2685 If there is no other client load running this can cause the
2686 client node to run out of memory. Instead flush old inodes
2687 from client cache that have the same inode number as a new inode.
2690 Frequency : SLES9 2.6.5 kernel and long filenames only
2691 Bugzilla : 9969, 10379
2692 Description: utime reports stale NFS file handle
2693 Details : SLES9 uses out-of-dentry names in some cases, which confused
2694 the lustre dentry revalidation. Change it to always use the
2698 Frequency : rare, unless heavy write-truncate concurrency is continuous
2699 Bugzilla : 4180, 6984, 7171, 9963, 9331
2700 Description: OST becomes very slow and/or deadlocked during object unlink
2701 Details : filter_destroy() was holding onto the parent directory lock
2702 while truncating+unlinking objects. For very large objects this
2703 may block other threads for a long time and slow overall OST
2704 responsiveness. It may also be possible to get a lock ordering
2705 deadlock in this case, or run out of journal credits because of
2706 the combined truncate+unlink. Solution is to do object truncate
2707 first in one transaction without parent lock, and then do the
2708 final unlink in a new transaction with the parent lock. This
2709 reduces the lock hold time dramatically.
2712 Frequency : rare, 2.4 kernels only
2714 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2715 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2716 before unmounting their filesystem in order to avoid deadlock.
2717 The kernel_locked() function in 2.4 kernels only checks whether
2718 the kernel lock is held, not whether it is this process that is
2719 holding it as 2.6 kernels do.
2724 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2725 Details : The client ptlrpc code may be trying to reconnect to a down
2726 server before a previous connection attempt has timed out.
2727 Increase the reconnect interval to be longer than the connection
2728 timeout interval to avoid sending duplicate connections to
2732 Frequency : echo_client brw_test command
2734 Description: fix echo_client to work with OST preallocated code
2735 Details : OST preallocation code (5137) didn't take echo_client IO path
2736 into account: echo_client calls filter methods outside of any
2737 OST thread and, hence, there is no per-thread preallocated
2738 pages and buffers to use. Solution: hijack pga pages for IO. As
2739 a byproduct, this avoids unnecessary data copying.
2743 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2744 Description: Client can oops in mdc_commit_close() after open replay
2745 Details : It was possible for the MDS to return an open request with no
2746 transaction number in mds_finish_transno() if the client was
2747 evicted, but without actually returning an error. Clients
2748 would later try to replay that open and may trip an assertion
2749 Simplify the client close codepath, and always return an error
2750 from the MDS in case the open is not successful.
2753 Frequency : rare, 2.6 OSTs only
2755 Description: OST may deadlock under high load on fragmented files
2756 Details : If there was a heavy load and highly-fragmented OST filesystems
2757 it was possible to have all the OST threads deadlock waiting on
2758 allocation of biovecs, because the biovecs were not released
2759 until the entire RPC IO was completed. Instead, release biovecs
2760 as soon as they are complete to ensure forward IO progress.
2762 Severity : enhancement
2764 Description: Support for specifying external journal device at mount
2765 Details : If an OST or MDS device is formatted with an external journal
2766 device, this device major/minor is stored in the ext3 superblock
2767 and may not be valid for failover. Allow detecting and
2768 specifying the external journal at mount time.
2773 Description: Mounting an MDS with pending unlinked files may cause oops
2774 Details : target_finish_recovery() calls mds_postrecov() which returned
2775 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2776 considers this an error and immediately begins cleaning up the
2777 lov, just after starting the mds_lov process
2779 Severity : enhancement
2781 Description: Implement 'lfs df' to report actual free space on per-OST basis
2782 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2783 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2784 usage of objects; '-h' to report in human readable format.
2786 ------------------------------------------------------------------------------
2788 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2795 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2796 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2797 at startup during orphan unlinking. Instead, skip buddy update
2798 before setup, as it will be regenerated after recovery is complete.
2803 Description: If an OST is inactive, its locks might reference stale inodes.
2804 Details : lov_change_cbdata() must iterate over all namespaces, even if
2805 they are inactive to clear inode references from the lock.
2807 Severity : enhancement
2808 Frequency : occasional, if non-standard max_dirty_mb used
2810 Description: Client will block write RPCs if not enough grant
2811 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2812 then the client will block writes while waiting for another RPC
2813 to complete instead of consuming its dirty limit. With change
2814 we get improved performance when max_dirty_mb is small.
2816 Severity : enhancement
2817 Bugzilla : 3389, 6253
2818 Description: Add support for supplementary groups on the MDS.
2819 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2820 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2821 lookups for user supplementary groups into a cache.
2825 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2826 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2827 for liblustre O_EXCL is enforced by MDS.
2831 Description: readdir never returns NULL in liblustre.
2832 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2833 dentry in struct dirent.
2837 Frequency : liblustre only, depends on application IO pattern
2838 Description: liblustre clients evicted if not contacting servers
2839 Details : Don't put liblustre clients into the ping_evictor list, so
2840 they will not be evicted by the pinger ever.
2842 Severity : enhancement
2844 Description: Add ability to evict clients by NID from MDS.
2845 Details : By echoing "nid:$NID" string into
2846 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2847 $NID would be instantly evicted from this MDS and from all active
2848 OSTs connected to it.
2852 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2853 Details : lookup_it_finish() used to query file size from OSTs that was not
2858 Description: service threads change working directory to that of init
2859 Details : Starting lustre service threads may pin the working directory
2860 of the parent thread, making that filesystem busy. Threads
2861 now change to the working directory of init to avoid this.
2865 Frequency : during shutdown only
2866 Description: shutdown with a failed MDS or OST can cause unmount to hang
2867 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2872 Frequency : 2.6 only
2873 Description: chmod/chown may include an extra supplementary group
2874 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2875 supplementary group and if none is specified this is used.
2878 Bugzilla : 5479 (6816)
2880 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2881 Details : If lookup is in progress on a file that is unlinked we might try
2882 to revalidate the inode and fail in revalidate after lookup is
2883 complete and ll_file_open() enqueues the open again but
2884 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2887 Frequency : always, if lconf --abort_recovery used
2889 Description: lconf --abort_recovery fails with 'Operation not supported'
2890 Details : lconf was attempting to abort recovery on the MDT device and not
2893 ------------------------------------------------------------------------------
2895 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2900 Frequency : rare (only unsupported configurations with a node running as an
2902 Bugzilla : 6514, 5137
2903 Description: Mounting a Lustre file system on a node running as an OST could
2905 Details : OSTs now preallocates memory needed to write out data at
2906 startup, instead of when needed, to avoid having to
2907 allocate memory in possibly low memory situations.
2908 Specifically, if the file system is mounted on on OST,
2909 memory pressure could force it to try to write out data,
2910 which it needed to allocate memory to do. Due to the low
2911 memory, it would be unable to do so and the node would
2912 become unresponsive.
2914 Severity : enhancement
2916 Description: Addition of lconf --service command line option
2917 Details : lconf now accepts a '--service <arg>' option, which is
2918 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2920 Severity : enhancement
2922 Description: Failover mode is now the default for OSTs.
2923 Details : By default, OSTs will now run in failover mode. To return to
2924 the old behaviour, add '--failout' to the lmc line for OSTs.
2926 Severity : enhancement
2928 Description: Health checks are now provided for MDS and OSTs
2929 Details : Additional detailed health check information on MSD and OSTs
2930 is now provided through the procfs health_check value.
2933 Frequency : occasional, depends on IO load
2935 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2936 after numerous create/delete cycles
2937 Details : The ext3 inode allocation policy would not allocate new inodes
2938 very well on the OSTs because there are no new directories
2939 being created. Instead we look for groups with free space if
2940 the parent directories are nearly full.
2945 Description: Network or server problems during mount may cause partially
2946 mounted clients instead of returning an error.
2947 Details : The config llog parsing code may overwrite the error return
2948 code during mount error handling, returning success instead
2954 Description: MDS can fail to allocate large reply buffers
2955 Details : After long uptimes the MDS can fail to allocate large reply
2956 buffers (e.g. zconf client mount config records) due to memory
2957 fragmentation or consumption by the buffer cache. Preallocate
2958 some large reply buffers so that these replies can be sent even
2959 under memory pressure.
2963 Frequency : rare (liblustre)
2964 Description: fsx running with liblustre complained that using truncate() to
2965 extend the file doesn't work. This patch corrects that issue.
2966 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2967 ATTR_SIZE and lsm use in llu_setattr_raw.
2971 Frequency : rare, only 2.6 kernels
2972 Description: Unusual file access patterns on the MDS may result in inode
2973 data being lost in very rare circumstances.
2974 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2975 code in the 2.6 kernel caused the inode and/or EA data not to
2976 be read from disk, causing single-file corruption.
2980 Frequency : rare, only 2.6 filesystems using extents
2981 Description: Heavy concurrent write and delete load may cause data corruption.
2982 Details : It was possible under high-load situations to have an extent
2983 metadata block in the block device cache from a just-unlinked
2984 file overwrite a newly-allocated data block. We now unmap any
2985 metadata buffers that alias just-allocated data blocks.
2989 Frequency : filesystems with default stripe_count larger than 77
2990 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2991 Details : lconf specifies an inode size of 4096 bytes when the default
2992 stripe_count is larger than 77. This conflicts with the default
2993 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2994 this case to avoid pinning too much memory for large EAs.
2996 ------------------------------------------------------------------------------
2998 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3003 Frequency : rare (extremely heavy IO load with hundreds of clients)
3005 Description: Client is evicted, gets IO error writing to file
3006 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3007 caused clients to be evicted for AST timeouts. The fixes in
3008 bug 5192 mean we no longer need to have such short AST timeouts
3009 so ldlm_timeout has been increased.
3012 Frequency : occasional during --force or --failover shutdown under load
3013 Bugzilla : 5949, 4834
3014 Description: Server oops/LBUG if stopped with --force or --failover under load
3015 Details : a collection of import/export refcount and cleanup ordering
3016 issues fixed for safer force cleanup
3019 Frequency : only filesystems larger than 120 OSTs
3020 Bugzilla : 5990, 6223
3021 Description: lfs getstripe would oops on a very large filesystem
3022 Details : lov_getconfig used kfree on vmalloc'd memory
3025 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3026 Bugzilla : 6242, 6243
3027 Description: reading from files that had been truncated to a non-zero size
3028 but never opened returned no data
3029 Details : ll_file_read() reads zeros from no-object files to EOF
3034 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3035 Details : There's little harm in aborting MDS/OSS recovery and letting it
3036 try again, so I removed the LASSERT and return an error instead.
3038 Severity : enhancement
3040 Description: New debugging infrastructure for tracking down data corruption
3041 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3042 (b) cover more of the client-side code path, and (c) try to narrow
3043 down where problems occurred
3047 Bugzilla : 3819, 4364, 4397, 6313
3048 Description: Racing close and eviction MDS could cause assertion in mds_close
3049 Details : It was possible to get multiple mfd references during close and
3050 client eviction, leading to one thread referencing a freed mfd.
3052 Severity: : enhancement
3053 Bugzilla : 3262, 6359
3054 Description: Attempts to reconnect to servers are now more aggressive.
3055 Details : This builds on the enhanced upcall-less recovery that was added
3056 in 1.4.2. When trying to reconnect to servers, clients will
3057 now try each server in the failover group every 10 seconds. By
3058 default, clients would previously try one server every 25 seconds.
3063 Description: After recovery, certain operations trigger a failed
3064 assertion on a client.
3065 Details : Failing over an mds, using lconf -d --failover, while a
3066 client was doing a readdir() call would cause the client to
3067 LBUG after recovery completed and the readdir() was resent.
3069 Severity : enhancement
3071 Description: Default groups are now added by lconf
3072 Details : You can now run lconf --group <servicename> without having to
3073 manually add groups with lmc.
3076 Frequency : occasional
3078 Description: Nodes with an elan id of 0 trigger a failed assertion
3081 Frequency : always when accessing e.g. tty/console device nodes
3083 Description: tty and some other devices nodes cannot be used on lustre
3084 Details : file's private_data field is used by device data and lustre
3085 values in there got lost. New field was added to struct file to
3086 store fs-specific private data.
3089 Frequency : when exporting Lustre via NFS
3091 Description: NFSD failed occasionally when looking up a path component
3092 Details : NFSD is looking up ".." which was broken in ext3 directories
3093 that had grown large enough to become hashed.
3096 Frequency : Clusters with multiple interfaces not on the same subnet
3098 Description: Nodes will repeatedly try to reconnect to an interface which it
3099 cannot reach and report an error to the log.
3100 Details : Extra peer list entries will be created by lconf with some peers
3101 unreachable. lconf now validates the peer before adding it.
3104 Frequency : Only if a default stripe is set on the filesystem root.
3106 Description: Setting a default stripe on the filesystem root prevented the
3107 filesystem from being remounted.
3108 Details : The client was sending extra request flags in the root getattr
3109 request and did not allocate a reply buffer for the dir EA.
3112 Frequency : occasional, higher if lots of files are accessed by one client
3113 Bugzilla : 6159, 6097
3114 Description: Client trips assertion regarding lsm mismatch/magic
3115 Details : While revalidating inodes the VFS looks up inodes with ifind()
3116 and in rare cases can find an inode that is being freed.
3117 The ll_test_inode() code will free the lsm during ifind()
3118 when it finds an existing inode and then the VFS later attaches
3119 this free lsm to a new inode.
3123 Bugzilla : 6422, 7030
3124 Description: MDS deadlock between mkdir and client eviction
3125 Details : Creating a new file via mkdir or mknod (starting a transaction
3126 and getting the ns lock) can deadlock with client eviction
3127 (gets ns lock and trying to finish a synchronous transaction).
3130 Frequency : occasional
3131 Description: While starting a server, the fsfilt_ext3 module could not be
3133 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3134 kernels. Previously, lconf would still use the ext3 name
3135 when trying to load modules. Now, it will correctly use
3136 ext3 on 2.4 and ldiskfs on 2.6.
3138 Severity : enhancement
3139 Description: The default stripe count has been changed to 1
3140 Details : The interpretation of the default stripe count (0, to lfs
3141 or lmc) has been changed to mean striping across a single
3142 OST, rather than all available. For general usage we have
3143 found a stripe count of 1 or 2 works best.
3145 Severity : enhancement
3146 Description: Add support for compiling against Cray portals.
3147 Details : Conditional compiling for some areas that are different
3151 Frequency : occasional
3152 Bugzilla : 6409, 6834
3153 Description: Creating files with an explicit stripe count may lead to
3154 a failed assertion on the MDS
3155 Details : If some OSTs are full or unavailable, creating files may
3156 trigger a failed assertion on the MDS. Now, Lustre will
3157 try to use other servers or return an error to the
3161 Frequency : occasional
3163 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3164 caused lock timeout during readahead (since 1.4.2).
3165 Details : Processes doing readahead might match a lock that hasn't been
3166 granted yet if there are overlapping and conflicting lock
3167 requests. The readahead process waits on ungranted lock
3168 (original lock is CBPENDING), while OST waits for that process
3169 to cancel CBPENDING read lock and eventually evicts client.
3171 Severity : enhancement
3173 Description: Initial enabling of flock support for clients
3174 Details : Implements fcntl advisory locking and file status functions.
3175 This feature is provided as an optional mount flag (default
3176 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3177 locking are implemented yet, and those that are are not guaranteed
3178 to be completely correct in production environments.
3179 mount -t lustre -o [flock|noflock] ...
3182 Frequency : occasional
3184 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3185 assertion in the ext3 JBD (journaling) layer.
3186 Details : The b_committed_data struct is protected by the big kernel lock
3187 in 2.4 kernels, serializing journal_commit_transaction() and
3188 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3189 access to this struct. In 2.6 kernels there is finer grained
3190 locking to improve SMP performance of the JBD layer.
3194 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3199 Description: merge and cleanup kernel patches.
3202 -----------------------------------------------------------------------------
3204 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3206 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3207 versions of Lustre. Please update all servers and clients to
3208 version 1.4.2 or later at the same time. You must also run
3209 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3210 to update the configuration logs.
3212 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3213 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3214 - swab llog records of type '0' so we get proper header size/idx (5861)
3215 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3216 - fix rename of one directory over another leaking an inode (5953)
3217 - avoid SetPageDirty on 2.6 (5981)
3218 - don't re-add just-being-destroyed locks to the waiting list (5653)
3219 - when creating new directories, inherit the parent's custom
3220 striping settings if present parent (3048)
3221 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3222 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3223 - don't hold client locks on temporary worklist from l_lru (5666)
3224 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3225 - automatically evict dead clients (5921)
3226 - Update file size properly in create+truncate+fstat case (6196)
3227 - Do not unhash mountpoint dentries, do not allow removal of
3229 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3230 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3231 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3232 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3233 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3234 - Correctly handle reads of files with no objects (6243)
3235 - lctl recover will also mark a device active if deactivate used (5933)
3237 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3238 - allow --write-conf on an MDS with different nettype than client (5619)
3239 - don't write config llogs to MDS for mounts not from that MDS (5617)
3240 - lconf should create multiple TCP connections from a client (5201)
3241 - init scripts are now turned off by default; run chkconfig --on
3242 lustre and chkconfig --on lustrefs to use them
3243 - upcalls are no longer needed for clients to recover to failover
3245 - add --abort-recovery option to lconf to abort recovery on device
3247 - add support for an arbitrary number of OSTs (3026)
3248 - Quota support protocol changes.
3249 - forward compatibility changes to wire structs (6007)
3250 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3251 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3252 - improved "lustre status" script
3253 - initialize blocksize for non-regular files (6062)
3254 - added --disable-server and --disable-client configure options (5782)
3255 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3256 - Vanilla 2.4.29 support
3257 - increase maximum number of obd devices to 520 (6242)
3258 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3259 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3260 SLES 9 2.6.5 kernels
3261 - limit stripes per file to 160 (the maximum EA size) (6093)
3263 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3266 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3267 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3268 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3269 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3270 - small ext3 extents cleanups and fixes (5733)
3271 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3272 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3273 - avoid needless client->OST connect, fix handle mismatch (5317)
3274 - fix DLM error path that led to out-of-sync client, long delays (5779)
3275 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3276 - fix several locking issues related to i_size (5492,5624,5654,5672)
3277 - don't move pending lock onto export if it is already evicted (5683)
3278 - fix kernel oops when creating .foo in unlinked directory (5548)
3279 - fix deadlock in obdfilter statistics vs. object create (5811)
3280 - use time_{before,after} to avoid timer jiffies wrap (5882)
3281 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3282 - Do not leak request if server was not able to process it (5154)
3283 - If mds_open unable to find parent dir, make that negative lookup(5154)
3284 - don't create new directories with extent-mapping (5909, 5936)
3286 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3287 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3288 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3289 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3290 - support for RHEL4 kernel on i686 (5773)
3291 - provide error messages when incompatible logs are encountered (5898)
3293 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3294 * version 1.4.0.10 (1.4.1 release candidate 1)
3296 - don't keep a lock reference when lock is not granted (4238)
3297 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3298 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3299 - drop import inflight refcount on signal_completed_replay error (5255)
3300 - unlock page after async write error during send (3677)
3301 - handle missing objects in filter_preprw_read properly (5265)
3302 - no transno return for symlink open, don't save no-trasno open (3440)
3303 - don't try to complete elan receive that already failed (4012)
3304 - free RPC server reply state on error (5406)
3305 - clean up thread from ptlrpc_start_thread() on error (5160)
3306 - readahead could read extra page into cache that wasn't ejected (5388)
3307 - prevent races in class_attach/setup/cleanup/detach (5260)
3308 - don't dereference de->d_inode after l_dput of de (5458)
3309 - use "int" for stripe value returned from lock_to_stripe (5544)
3310 - mballoc allocation and error-checking fixes in 2.6 (5504)
3311 - block device patches to fix I/O request sizes in 2.6 (5482)
3312 - look up hostnames for IB nals (5602)
3313 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3314 - don't start multiple acceptors for the same port (5277)
3315 - fix incorrect LASSERT in mds_getattr_name (5635)
3316 - export a proc file for general "ping" checking (5628)
3317 - fix "lfs check" to not block when the MDS is down (5628)
3319 - service request history (4965)
3320 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3321 - create an "evict_client" /proc entry on OSTs, like the MDS has
3322 - fix mount usage message, return errors per mount(8) (5168)
3323 - change grep [] to grep "[]" in tests so they work in more UMLs
3324 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3325 - remove ancient LOV_MAGIC_V0 EA support (5047)
3326 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3327 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3328 - allow readahead allocations to fail when low on memory (5383)
3329 - mmap locking landed again, after considerable improvement (2828)
3330 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3332 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3335 - send OST transaction number in read/write reply to free req (4966)
3336 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3337 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3338 - the watchdog thread now runs as interruptible (5246)
3339 - flock/lockf fixes (but it's still disabled, pending 5135)
3340 - don't use EXT3 constants in llite code (5094)
3341 - memory shortage at startup could cause assertion (5176)
3343 - reorganization of lov code
3344 - single portals codebase
3346 - add extents/mballoc support (5025)
3347 - direct I/O reads in the obdfilter (4048)
3348 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3350 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3353 - send OST transaction number in read/write reply to free req (4966)
3354 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3355 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3356 - the watchdog thread now runs as interruptible (5246)
3357 - handle missing objects in filter_preprw_read properly (5265)
3358 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3359 - drop import inflight refcount on signal_completed_replay error (5255)
3360 - unlock page after async write error during send (3677)
3361 - return original error code on reconstructed replies (3761)
3362 - no transno return for symlink open, don't save no-trasno open (3440)
3364 - add pid to ldlm debugging output (4922)
3365 - bump the watchdog timeouts -- we can't handle 30sec yet
3366 - extra debugging for orphan dentry/inode bug (5259)
3368 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3371 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3372 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3373 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3374 - fix NULL dereference in /proc/sys/portals/routes (4827)
3375 - allow failed mdc_close() operations to be interrupted (4561)
3376 - stop precreate on OST before MDS would time out on it (4778)
3377 - don't send partial-page writes before EOF from client (4410)
3378 - discard client grant for sub-page writes on large-page clients (4520)
3379 - don't free dentries not owned by NFS code, check generation (4806)
3380 - fix lsm leak if mds_create_objects() fails (4801)
3381 - limit debug_daemon file size, always print CERROR messages (4789)
3382 - use transno after validating reply (3892)
3383 - process timed out requests if import state changes (3754)
3384 - update mtime on OST during writes, return in glimpse (4829)
3385 - add mkfsoptions to LDAP (4679)
3386 - use ->max_readahead method instead of zapping global ra (5039)
3387 - don't interrupt __l_wait_event() during strace
3389 - add software watchdogs to catch hung threads quickly (4941)
3390 - make lustrefs init script start after nfs is mounted
3391 - fix CWARN/ERROR duplication (4930)
3392 - return async write errors to application if possible (2248)
3393 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3394 - print NAL number in %x format (4645)
3395 - update barely-supported suse-2.4.21-171 series (4842)
3396 - support for sles 9 %post scripts
3397 - support for building 2.6 kernel-source packages
3398 - support for sles km_* packages
3400 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3403 - ignore -ENOENT errors in osc_destroy (3639)
3404 - notify osc create thread that OSC is being cleaned up (4600)
3405 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3406 - reconstruct ost_handle() like mds_handle() (4657)
3407 - create a new thread to do import eviction to avoid deadlock (3969)
3408 - let lconf resolve symlinked-to devices (4629)
3409 - don't unlink "objects" from directory with default EA (4554)
3410 - hold socknal file ref over connect in case target is down (4394)
3411 - allow more than 32000 subdirectories in a single directory (3244)
3412 - fix blocks count for O_DIRECT writes (3751)
3413 - OST returns ENOSPC from object create when no space left (4539)
3414 - don't send truncate RPC if file size isn't changing (4410)
3415 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3416 - bind to privileged port in socknal and tcpnal (3689)
3418 - rate limit CERROR/CWARN console message to avoid overload (4519)
3419 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3420 - basic mmap support (3918)
3421 - kernel patch series update from b1_4 (4711)
3423 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3426 - avoid crash during MDS cleanup with OST shut down (2775)
3427 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3428 - don't use bad inodes on the MDS (3744)
3429 - dynamic object preallocation to improve recovery speed (4236)
3430 - don't hold spinlock over lock dumping or change debug flags (4401)
3431 - don't zero obd_dev when it is force cleaned (3651)
3432 - print grants to console if they go negative (4431)
3433 - "lctl deactivate" will stop automatic recovery attempts (3406)
3434 - look for existing locks in ldlm_handle_enqueue() (3764)
3435 - don't resolve lock handle twice in recovery avoiding race (4401)
3436 - revalidate should check working dir is a directory (4134)
3438 - don't always mark "slow" obdfilter messages as errors (4418)
3440 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3443 - don't close LustreDB during write_conf until it is done (3860)
3444 - fix typo in lconf for_each_profile (3821)
3445 - allow dumping logs from multiple threads at one time (3820)
3446 - don't allow multiple threads in OSC recovery (3812)
3447 - fix debug_size parameters (3864)
3448 - fix mds_postrecov to initialize import for llog ctxt (3121)
3449 - replace config semaphore with spinlock (3306)
3450 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3451 - don't allow enqueue to complete on a destroyed export (3822)
3452 - down write_lock before checking llog header bitmap (3825)
3453 - recover from lock replay timeout (3764)
3454 - up llog sem before sending rpc (3652)
3455 - reduce ns lock hold times when setting kms (3267)
3456 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3457 - fix NULL deref and obd_dev leak on setup error (3312)
3458 - replace some LBUG about llog ops with error handling (3841)
3459 - don't match INVALID dentries from d_lookup and spin (3784)
3460 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3461 - fix invalid assertion in ptlrpc_set_wait (3880)
3463 - add libwrap support for the TCP acceptor (3996)
3464 - add /proc/sys/portals/routes for non-root route listing (3994)
3465 - allow setting MDS UUID in .xml (2580)
3466 - print the stack of a process that LBUGs (4228)
3468 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3471 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3472 - make sure to unset replay flag from failed open requests (3440)
3473 - if default stripe count is 0, use OST count for inode size (3636)
3474 - update parent mtime/ctime on client for create/unlink (2611)
3475 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3476 - free recovery state on server during a forced cleanup (3571)
3477 - unregister_reply for resent reqs (3063)
3478 - loop back devices mounting and status check on 2.6 (3563)
3479 - fix resource-creation race that can provoke i_size == 0 (3513)
3480 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3481 - more debugging for page-accounting assertion (3746)
3482 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3483 - avoid deadlock after precreation failure (3758)
3484 - fix race and lock order deadlock in orphan handling (3450, 3750)
3485 - add validity checks when grabbing inodes from l_ast_data (3599)
3487 - add /proc/.../recovery_status to obdfilter (3428)
3488 - lightweight CDEBUG infrastructure, debug daemon (3668)
3489 - change default OSC RPC parameters to be better on small clusters
3490 - turn off OST read cache for files smaller than 32MB
3491 - install man pages and include them in rpms (3100)
3492 - add new init script for (un)mounting lustre filesystems (2593)
3493 - run chkconfig in %post for init scripts (3701)
3494 - drop scimac NAL (unmaintained)
3496 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3499 - clean kiobufs before and after use (3485)
3500 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3501 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3502 - change init script to fail more clearly if not run as root (1528)
3503 - allow clients to reconnect during replay (1742)
3504 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3505 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3506 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3507 - fix cond_resched (3554)
3508 - only evict unfinished clients after recovery (3515)
3509 - allow bulk resend, prevent data loss (3570)
3510 - dynamic ptlrpc request buffer allocation (2102)
3511 - don't allow unlinking open directory if it isn't empty (2904)
3512 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3513 - remove extraneous obd dereference causing LASSERT failure (3334)
3514 - don't use get_cycles() when creating temp. files on the mds (3156)
3515 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3516 - handle EEXIST for set-stripe, set proper directory name (3336)
3518 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3519 - fix ksocknal_fmb_callback() error messages (2918)
3521 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3524 - don't copy lvb into (possibly NULL) reply on error (2983)
3525 - don't deref dentry after dput, don't free lvb on error (2922)
3526 - use the kms to determine writeback rpc length (2947)
3527 - increment oti_logcookies when osc is inactive (2948)
3528 - update client's i_blocks count via lvb messages (2543)
3529 - handle intent open/close of special files properly (1557)
3530 - mount MDS with errors=remount-ro, like obdfilter (2009)
3531 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3532 - don't use cancelling-locks' kms values (2947)
3533 - use highest lock extent for kms, not last one (2925)
3534 - don't dereference ERR_PTR() dentry in error handling path (3107)
3535 - fix thread race in portals_debug_dumplog() (3122)
3536 - create lprocfs device entries at setup instead of at attach (1519)
3537 - common AST error handler, don't evict client on completion race (3145)
3538 - zero nameidata in detach_mnt in 2.6 (3118)
3539 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3540 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3541 - initialize RPC timeout timer earlier for 2.6 (3219)
3542 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3543 - print nal/nid for unknown nid (3258)
3544 - additional checks for oscc recovery before doing precreate (3284)
3545 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3546 - don't crash in mdc_close for bad permissions on open (3285)
3547 - zero i_rdev for non-device files (3147)
3548 - clear page->private before handing to FS, better assertion (3119)
3549 - tune the read pipeline (3236)
3550 - fix incorrect decref of invalidated dentry (2350)
3551 - provide read-ahead stats and refine rpc in flight stats (3328)
3552 - don't hold journal transaction open across create RPC (3313)
3553 - update atime on MDS at close time (3265)
3554 - close LDAP connection when recovering to avoid server load (3315)
3555 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3556 - don't leak open file on MDS after open resend (3325)
3557 - serialize filter_precreate and filter_destroy_precreated (3329)
3558 - loop device shouldn't call sync_dev() for nul device (3092)
3559 - clear page cache after eviction (2766)
3560 - resynchronize MDS->OST in background (2824)
3561 - refuse to mount the same filesystem twice on same mountpoint (3394)
3562 - allow llmount to create routes for mounting behind routers (3320)
3563 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3564 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3565 - fix rare problem with rename on htree directories (3417)
3567 - allow default OST striping configuration per directory (1414)
3568 - fix compilation for qswnal for 2.6 kernels (3125)
3569 - increase maximum number of MDS request buffers for large systems
3570 - change liblustreapi to be useful for external progs like lfsck (3098)
3571 - increase local configuration timeout for slow disks (3353)
3572 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3574 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3577 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3578 - don't overwrite extent policy data in reply if lock was blocked (2901)
3579 - drop filter export grants atomically with removal from device (2663)
3580 - del obd_self_export from work_list in class_disconnect_exports (2908)
3581 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3582 - swab reply message in mdc_close, other PPC fixes (2464)
3583 - fix destroying of named logs (2325)
3584 - overwrite old logs when running lconf --write_conf (2264)
3585 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3586 - fix race in target_handle_connect (2898)
3587 - mds_reint_create() should take same inode create lock (2926)
3588 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3589 - don't close files for self_export to avoid uninitialized obd (2936)
3590 - allow MDS with the same name as client node (2939)
3591 - hold dentry reference for closed log files for unlink (2325)
3592 - reserve space for all logs during transactions (2059)
3593 - don't evict page beyond end of stripe extent (2925)
3594 - don't oops on a deleted current working directory (2399)
3595 - handle hard links to targets without a parent properly (2517)
3596 - don't dereference NULL lock when racing during eviction (2867)
3597 - don't grow lock extents when lots of conflicting locks (2919)
3599 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3602 - account for cache space usage on clients to avoid data loss (974)
3603 - lfsck support in lustre kernel code (2349)
3604 - reduce journal credits needed for BRW writes (2370)
3605 - orphan handling to avoid losing space on client/server crashes
3606 - ptlrpcd can be blocked, stopping ALL progress (2477)
3607 - use lock value blocks to assist in proper KMS, faster stat (1021)
3608 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3609 - recovery for initial connections (2355)
3610 - fixes for mds_cleanup_orphans (1934)
3611 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3612 - block all file creations until orphan recovery completes (1901)
3613 - client remove rq_connection from request struct (2423)
3614 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3615 - recovery timer race (2670)
3616 - mdc_close recovey bug (2532)
3617 - ptlrpc cleanup bug (2710)
3618 - mds timeout on local locks (2588)
3619 - namespace lock held during RPCs (2431)
3620 - handle interrupted sync write properly (2503)
3621 - don't try to handle a message that hasn't been replied to (2699)
3622 - client assert failure during cleanup after abort recovery (2701)
3623 - leak mdc device after failed mount (2712)
3624 - ptlrpc_check_set allows timedout requests to complete (2714)
3625 - wait for inflight reqs when ptlrpcd finishes (2710)
3626 - make sure unregistered services are removed from the srv_list
3627 - reset bulk XID's when resending them (caught by 1138 test)
3628 - unregister_bulk after timeout
3629 - fix lconf error (2694)
3630 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3631 - readahead locks pages, leaves pending causing memory pressure (2673)
3632 - increase OST request buffers to 4096 on large machines (2729)
3633 - fix up permission of existing directories in simple_mkdir (2661)
3634 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3635 - don't assign transno to errored transactions (2742)
3636 - don't delete objects on OST if given a bogus objid from MDS (2751)
3637 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3638 - if rq_no_resend, then timeout request after recovery (2432)
3639 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3640 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3641 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3642 - MDS recovery shouldn't delete objects that it has given out (2730)
3643 - if enqueue arrives after completion, don't clobber LVB (2819)
3644 - don't unlock pages twice when trigger_group_io returns error (2814)
3645 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3646 - don't write pages to disk if there was an error (1450)
3647 - don't ping imports that have recovery disabled (2676)
3648 - take buffered bytes into account when balancing socknal conn (2817)
3649 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3650 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3651 - remove little-endian swabbing of llog records (1987)
3652 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3653 - retry reposting request buffers if they fail (1191)
3654 - grow extent at grant time to avoid granting a revoked lock (2809)
3655 - lock revoke doesn't evict page if covered by a second lock (2765)
3656 - disable VM readahead to avoid reading outside lock extents (2805)
3658 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3659 - updated kernel patches for hp-2.4.20 kernel (2681)
3661 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3664 - fix truncated write corruption (2366)
3665 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3667 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3668 - ensure len(uuid) < 37 in lmc (1171)
3669 - fix ia64 OOPS in llog_test (2255)
3670 - zero end of page at obdfilter for partial page writes (2648)
3671 - don't leave stale dentries around after renames (bug 2428)
3672 - fix timeouts when evicting a client with a single lock held (2642)
3673 - set deadline for the initial HELLO message to drain (2634)
3674 - print out dotted-quad IP addresses in the socknal (2302)
3676 - additional debugging for MDS client eviction problem (2443)
3677 - fix mkfsoptions support for osts (2603, 2604)
3679 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3682 - add series for the vanilla 2.6.0 kernel
3683 - add series for the vanilla 2.4.24 kernel
3684 - add series for a cray x86/64 UL kernel drop
3685 - fix xattr patches for the vanilla 2.4.19 series
3687 - generate true UUIDs in lmc (1171)
3688 - have portals stack dumping break in UML (2466)
3689 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3690 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3691 - wake callback waiting threads on client eviction (2460)
3692 - Add --ptldebug and --subsystem to lmc (1719)
3693 - update assertion to allow safe interrupt allocation
3694 - set rq_no_resend for cancel requests (2432)
3695 - recalculate ptlrpcd timeout after resend (2494)
3696 - call vfs_rmdir when removing pending directories (2368)
3697 - fix renaming a file to itself (2429)
3698 - lmc creates a default one-stripe lov (2454)
3699 - expand procfs space to handle large clusters (2326)
3700 - increase UML stack to avoid overflow
3701 - update lconf's list of debug and subsystem masks
3702 - fix lfs find --obd (2510)
3703 - /proc tunable for disabling filter read caching (2591)
3704 - stop rpm packages from altering slapd.conf (2301)
3705 - disable nagle in the socknal under 0conf (2578)
3706 - choose mds inode size based on stripe count (2572)
3707 - fix kernel-source rpm problems (2516)
3709 - add --disable-doc to avoid pdf generation (2421)
3710 - update documentation, tests, type-os, comments
3711 - avoid format warnings on ia64
3712 - remove the TOE NAL
3713 - tiny code cleanups by removing unused fields
3715 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3718 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3719 - remove the most blatant lies from BUILDING file (2371)
3720 - change default debug level to reasonable production setting
3721 - reduce client side cache size to reduce cache flush time
3722 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3723 - make TCP zerocopy and pinger support enabled by default (2476)
3724 - sync writes completed after process exits caused crashes (2319)
3725 - maintain correct mount count on the MDS (2356)
3726 - backout 1557, because 2316 wasn't really fixed
3727 - better file I/O statistics gathering in /proc
3728 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3729 - another kernel patch to fix zero-copy TCP function export
3730 - don't take duplicate lock when processing re-sent getattr (2420)
3731 - lctl uses obd_self_export instead of creating new conn (2353)
3732 - MDS/OST recovery case which requires object creation asserted (2425)
3733 - move lfs from /usr/sbin to /usr/bin in packages
3734 - fix race between mds_client_add and mds_client_free (2417)
3735 - use kmalloc instead of slabs in portals (2430)
3736 - don't create duplicate records when a failover MDS is present (2442)
3737 - remove unnecessary mount age check (2332)
3738 - don't remove directory inodes from locks prematurely (2451)
3739 - don't break if MDS service name is the same as hostname (2103)
3740 - fix races in client write RPC generation when cache full (2482)
3742 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3745 - remove now-unused request->rq_obd (278)
3746 - if an allocation fails, print out how much memory we've used (1933)
3747 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3748 - add the "configurable stack size" patch to most series files (1256)
3749 - ability to write large log records, for 100+ OST configs (2306)
3750 - fix NULL deref when filter_prep fails (2314)
3751 - fix operator precedence error in filter_sync
3752 - dynamic allocation of socknal TX descriptors (2315)
3753 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3754 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3755 - ensure that CWARN messages in Portals always get to the syslog
3756 - __init/__exit are not for prototype decls (ldlm_init/exit)
3757 - x86-64 compile warning fixes
3758 - fix gateway LMC keyword conflict (2318)
3759 - fix MDS lock inversions in getattr/reint paths (1844)
3760 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3761 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3762 - DLM race condition prevented some lock evictions (2328)
3763 - ENOMEM detection and retry on socknal sends (2230)
3764 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3765 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3766 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3767 - better error messages when a client is rejected during recovery (1505)
3768 - avoid cancelling locks which were never granted, after failure (2330)
3769 - fix i_sem/journal inversion in mds_client_add (2333)
3770 - fix truncate/getattr lock cycle deadlock (2334)
3771 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3772 - fix two rare exit paths which could leak an l_lock() ref (2321)
3773 - fencepost error in MDS/OST orphan recovery (2226)
3774 - make log record alignment 8 bytes (1988)
3775 - lstripe now fails when requested offset > ost_count (2237)
3776 - ensure that all kernel series have a complete list.h (1607)
3777 - fix crashes in special-file operations (2316)
3778 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3780 - allow configurable automake binary, for testing new versions
3781 - small update to the lfs documentation
3783 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3785 * fix negative export reference count in fsfilt_sync (2312)
3787 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3788 * release candidate 0.9.1
3790 - orphans are moved into the PENDING directory for possible recovery
3791 - replayed opens now open by fid for orphan/rename safety (1042)
3792 - last close of an orphan inode generates a transno (683)
3793 - chdir() and mount() now pin the directory entry (1020)
3794 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3795 - discard very old requests without processing them (1502)
3796 - remove some common, well-understood CERRORs (1505)
3797 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3798 - clear "grant" flags in OST replies until OST grant code lands (1644)
3799 - fix read performance by not clobbering i_blksize on client (1598)
3800 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3801 - make LRU size a /proc tunable, clears locks when reduced (707)
3802 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3803 - server locks take references on exports now (1558)
3804 - build fixes for 2.4.20-rh trees (1663)
3805 - return an error from lov_create if all OSCs are inactive (1751)
3806 - fix import levels when a reconnect happens without a timeout (1597)
3807 - exit early from mds_open if we get a lookup error (1749)
3808 - partial page read at EOF wouldn't wait for disk before sending (1642)
3809 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3810 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3811 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3812 - write/truncate lock inversion (1639)
3813 - Don't auto-load obdclass, portals modules during cleanup (1495)
3814 - fix timestamps from jumping to "now" (1763)
3815 - extra journal assertions (1648)
3816 - add an extra multiunlink test (1771)
3817 - fix read_record/write_record API (1776)
3818 - fix leak of offset_extent, possible incorrect i_size later (1772)
3819 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3820 - Debugging for the kqswnal_get_idle_tx problems (1820)
3821 - Allow recovery to be attempted multiple times (1536)
3822 - Write out MDS last_rcvd file after it is first created (1600)
3823 - Fix tx_descriptor leak in failed transmit situations (1827)
3824 - ext3 journaling fixes for assertion failure after IO error (1871)
3825 - class_export_put() on freed export after completion AST error (1896)
3826 - Fix revalidate looping in VFS (1322)
3827 - Don't access a freed export during MDS_REINT timeout (1521)
3828 - Add open-unlink recovery support on the MDS (1673,1764)
3829 - Return an error if no MDS data was read from last_rcvd (1946)
3830 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3831 - Don't setup a disk device that doesn't match exported UUID (317)
3832 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3833 - avoid committing NULL handle in force close
3834 - local.sh is now a one-stripe LOV configuration
3835 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3836 - fix ext3 htree duplicate directory entry corruption (1516)
3837 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3838 - update to new LOV EA format (2097)
3839 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3841 2003-06-15 Phil Schwan <phil@clusterfs.com>
3844 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3845 - per-import/export recovery handling (958, 931, 959)
3846 - multiple last-rcvd slots, for serving multiple FSes (949)
3847 - connections are again shared between multiple imp/exports (963, 964)
3848 - "umount -f" would hang if any requests needed to be sent (393, 978)
3849 - avoid pinning large req buffer by copying for queued messages (989)
3850 - add "uuid" to "lctl device" command to help upcalls (991)
3851 - "open" RPCs with transnos would confuse recovery counters (1037)
3852 - do proper endian conversion of all wire messages (288, 340, 891)
3853 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3854 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3855 - fix ptlrpc_connection leak in target_handle_connect (1174)
3856 - fix import refcounting bug in OST and MDS cleanup (1134)
3857 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3858 - fix very unlikely obd_types race condition (501)
3859 - remove osc_open hack for echo_client (1187)
3860 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3861 - a failure in read_inode2 leads to deadlock (1139)
3862 - cancel ack-locks as soon as transaction is committed (1072)
3863 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3864 - make sure to commitrw after any preprw to avoid deadlock (1162)
3865 - failing to execute a file in a lustre FS would lock inode (1203)
3866 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3867 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3868 - obd_uuid2tgt would walk off the end of the list (1255)
3869 - on IA64 the osc would give portals incorrect bulk size (1258)
3870 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3871 - fix lock inversion caused by new llite matching code (1282)
3872 - limit the number of dirty pages on a client to 10MB (1286)
3873 - timed out locks were not being corrected cancelled (1289)
3874 - fix O_DIRECT above 4GB on IA-32 (1292)
3875 * major user-visible changes
3876 - fail out/fail over policy now controlled by the upcall (993)
3878 - add OBD_PING to check server availability and failure (954)
3879 - lustre messages are now sent in sending host order (288, 340, 891)
3880 - add eadatalen to MDS getattr reply (340)
3881 - OST read replies may contain second buffer, with per-page status (593)
3883 2003-03-11 Phil Schwan <phil@clusterfs.com>
3886 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3887 - failed open()s could cause deadlock; fixed (867, 869)
3888 - stop cancelling OST locks when files are closed (481)
3889 - overlapping XID spaces caused network corruption (851, 853)
3890 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3891 - setattr_raw added, to do single-RPC, server-side setattrs
3892 - lmc/lconf syntax change for OST UUIDs
3893 - fix crashy race condition between ptlrpc_free_req and osc_close
3894 - don't use request in mdc_enqueue if we hit a timeout (889)
3895 - don't set the inode i_size for regular files from the MDS (896)
3896 - handle out of order completion AST (842)
3897 - don't LBUG if a lock request times out after receiving AST (913)
3898 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3899 - if a bad lock AST arrives, send an error instead of dropping entirely
3900 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3901 - fix leak in bulk IO when only partially completed (899, 900, 926)
3902 - fix O_DIRECT for ia64 (55)
3903 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3904 - C-z after timeout could hang a process forever; fixed (977)
3906 - client-side I/O cache (678, 924, 929, 941, 970)
3908 - READPAGE and SETATTRs which don't take server-side locks get
3911 2003-02-11 Phil Schwan <phil@clusterfs.com>
3914 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3915 - Fix fsfilt_extN_readpage() to read a full page of directory
3916 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3917 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3918 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3919 - Most intent operations were removed, in favour of a new RPC mode
3920 that does a single RPC to the server and bypasses most of the VFS
3921 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3922 - Aggressively cancel local locks on DLM servers
3923 - mds_reint_unlink sends EA to the client if it's the last nlink.
3924 client uses that EA to unlink OST objects.
3925 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3926 - recursive symlinks were fixed (439)
3927 - fixed NULL deref in DEBUG_REQ
3928 - filter_update_lastobjid no longer calls sync, which annoyed extN
3929 - fixed multi-client small-writes to a single file problem (445)
3930 - fixed mtime updates during file writes (607)
3931 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3932 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3933 - fixed Linux OST size reporting problem (444, 656)
3934 - OST now updates object mtime with writes or setattr (607, 619)
3935 - client verifies file size before zeroing page past EOF (445)
3936 - OST now writes last allocated objid to disk with allocation (108)
3937 - LOV on echo now works (409)
3939 - mds_reint_unlink sends a new buffer, with the EA included. this
3940 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3941 set if a regular file was being unlinked, and it was the last link
3942 - use PtlGet from the target for bulk writes (315)
3943 - OST now updates object mtime with writes or setattr (607, 619)
3944 - LDLM now has a grant-time callback to revalidate locked items, if
3946 - Many MDS operations were reorganized to combat race conditions
3948 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3949 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3951 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3954 - Fully reactivate OST imports after reconnection (512, others)
3955 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3956 - More graceful error handling for truncating on dead OST (515)
3957 - Don't error out unless we're actually accessing dead stripes (474)
3958 - Fix garbage sizes when stripes are missing (410)
3959 - LRU counters were broken, causing constant lock purge (433, 432)
3960 - garbage on read from stripes with failed OSTs (441)
3961 - mark OSCs as active before reconnecting during recovery (438)
3962 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3963 - lfind did not preserve OST order in output (443)
3964 - symlinks cause hung clients, incorrect data (439)
3965 - stop dereferencing request after dropping refcount (457)
3966 - don't LASSERT(spin_is_locked) on non-SMP (455)
3967 - fixes for many rename() bugs
3968 - fstat didn't correctly synchronize attributes (399)
3969 - server must handle lock cancellation during blocking AST prep (487)
3970 - bulk descriptors were free()d too soon (511)
3971 - fix paths in lconf, which would load incorrect modules (451, 507)
3972 - fix confusing lconf 'host not found' error message (386)
3973 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3974 - fix race condition in mdc_blocking_ast() for inode access (526)
3975 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3976 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3977 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3978 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3979 - fix lstripe in several ways (kernel side) (527)
3980 - fix request leak in ldlm_cli_enqueue (262)
3981 - incorrect OSC was marked inactive after OST failure
3982 - call mds_fs_cleanup before unmounting filesystem (524)
3983 - fix races between taking ns_lock and ldlm_lock_change_resource
3984 - fix races updating LOV export open file list
3985 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3986 - fix recovery NULL deref in ldlm_cli_cancel_unused
3987 - fix some DLM races by using new hash table for lock handles (419)
3988 - permit the client to specify desired inodes, at replay
3989 - duplicate requests when we queue them for replay reintegration
3990 - fix last_rcvd offset calculation
3991 - sync after each recovered transaction, so we always make progress
3992 - never, not always, ERESTART requests without transnos
3993 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3995 - skip replay if the MDS says that the client is already connected
3996 - don't check for a recovery-enabled export to match lctl's UUID
3997 - don't INC_USE_COUNT for phantom exports
3998 - don't crash when cleaning up phantom exports (567)
3999 - don't double-finish or set replay data for errored mdc_open requests
4000 - abort requests when they time out, so we don't get old replies
4001 - send/receive replies for AST messages again
4002 - if the client says that it doesn't have the lock, cancel it on the
4004 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4005 mark it as destroyed, it will all work out when decref is called
4006 - fix module use counts (22, 581)
4008 - ASTs now expect a reply (server cancels lock on error reply)
4010 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4013 - fix many simultaneous client startup (392)
4014 - fix dentry->d_it clobbering
4015 - credentials weren't being shipped for readdir/getattr operations
4016 - remove invalid assertions triggered during some concurrent MD
4018 - proper Lustre versions added (336, 389)
4019 - fix memory leak for create error case (398)
4020 - fix LOV locking bug that would get cli/srv out of sync
4021 - fix echo client over LOV (409)
4022 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4023 - fix double-O_EXCL intent crash (424)
4024 - avoid sending multiple lock CANCELs (352)
4026 - MDS can do multi-client recovery (modulo bugs in new code)
4028 - many updates, edits, cleanups
4030 2002-11-18 Phil Schwan <phil@clusterfs.com>
4033 - fix null d_it dereference (346)
4034 - fix full OST/dbench hang (333)
4035 - fix permission problem with file removal (286)
4036 - fix removal of OSCs from LOV when they fail
4037 - fix NULL deref during bulk timeout (214)
4038 - fix problems related to multiple filesystems on one MDS (241)
4039 - fixed serious subtle metadata locking bugs
4040 - free locks on clients when inodes are removed due to memory
4042 - fix inode pointer in lock data (285)
4043 - partial support for multiple MDS on a single host (241)
4044 - data locks weren't cancelled at clear_inode time (290, 311)
4045 - intent locks could lead to unbounded lock growth (205)
4046 - added a maximum lock count, an LRU list, and a flusher
4047 - fix multiple rename (365)
4048 - properly abstracted the echo client
4049 - OSC locked 1 byte too many; fixed
4050 - rewrote brw callback code:
4051 - fixed recovery bugs related to LOVs (306)
4052 - fixed too-many-pages-in-one-write crash (191)
4053 - fixed (again) crash in sync_io_timeout (214)
4054 - probably fixed callback-related race (385)
4056 - Add capability to MDS protocol
4057 - LDLM cancellations and callbacks on different portals
4059 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4062 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4063 - defer open object destruction to close time (601981, 138)
4064 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4065 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4066 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4067 - don't call obd_setattr in ll_file_release for destroyed objects
4068 * protocol change to lustre_msg: move |version| and add |flags|
4069 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4070 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4071 * added replay of create, unlink, link and rename operations during
4072 MDS failover; recovery should be much more robust now
4073 * remove failed OSCs from LOVs (only lov_create uses this so far)
4074 * the lustre-HOWTO was brought (more) up to date (582544)
4076 2002-10-23 Phil Schwan <phil@clusterfs.com>
4079 - in-use dentries weren't being reused properly (617851)
4080 - prevent multiple LDLM setup (599178)
4081 - fix LOV size calculations for truncate (617853)
4082 - fix client handling of MDS intent errors (POSIX)
4083 - fix permission bug in lovstripe.c test (624321)
4084 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4085 - truncate past end of file could corrupt data
4086 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4087 - a race in recovery could return ETIMEDOUT to apps (623947)
4088 - building outside the source directory was fixed
4089 * the lustre-HOWTO was brought (more) up to date (582544)
4090 * major progress was made on recovery functionality
4092 2002-10-10 Phil Schwan <phil@clusterfs.com>
4095 - recovery deadlock fix
4096 - rm -rf causes LBUG fix (617817)
4097 - file open by multiple tasks fix (618962)
4098 - directory permissions bugs (602707 and 620007)
4099 - journal_stop fixed with locking (611313)
4100 - O_APPEND failures resolved (618273, perhaps 614459)
4101 - lconf PATH fix (619770)
4102 - IA64 build fix (621450)
4103 - RPC buffer sizes scale with amount of memory
4105 2002-10-01 Phil Schwan <phil@clusterfs.com>
4108 - locks would be cancelled without throwing away data pages,
4109 resulting in inconsistent data (605627)
4110 - inode attributes were not always being refreshed (605627, 612449)
4111 - lconf now continues to cleanup after lctl reports an error
4112 - MDS now enforces user permissions (602707)
4113 - lprocfs cleanup fixed, but not yet enabled (614157)
4114 - fixed infinite server hang, should a client not respond to an AST
4115 - avoid going into recovery if user calls readlink() with a buffer
4116 that's too small (613941)
4117 - AST RPCs no longer require replies (614867) -- this may be changed
4118 - don't crash server if client sends an IOV that's too big (611336)
4119 - fixed lock conversion deadlock (611892)
4120 - fixed the following of symlinks (614622)
4121 * recovery: the server can remove locks from a client that dies, other
4122 clients can make progress
4123 * more extN patch fixes
4124 * compile-time configurable ptlrpc buffer allocations
4126 - collaborative read cache document
4127 - Lustre Lite Performance CDR document-in-progress
4129 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4132 - fix typo in patch-2.4.18
4134 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4137 - clear ptlrpc request each time in handle_incoming_request()
4138 - unlink of files now destroys the object on the OST
4140 2002-09-19 Peter Braam <braam@clusterfs.com>
4142 * add hard link support
4143 * change obdfile creation method
4144 * kernel patch changed
4146 2002-09-19 Peter Braam <braam@clusterfs.com>
4149 - stack overflow bug in extN fixed
4151 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4153 * documentation updates
4154 - add man pages for config tools
4155 - update tests/README to describe testing with new config tools
4156 - finish metadata API descriptions
4157 * bug fixes and cleanups
4158 - statfs workaround for 16TB limit
4159 - LOV stripe allocation improved, can stripe on subset of OSTs
4160 - LOV file size/IO offset was wrong for files > 4GB in size
4161 - object EA data was being dropped, caused files to be unreadable
4162 - memory overflow with non-LOV OST caused memory corruption
4163 - fixed regression tests to work with new config tools, obdfilter
4164 - fixed bug when directory size became larger than 1 block
4165 - fixed bug (for single client case) when PWD was deleted
4166 - invalidate local directory pages when doing intent-based ops
4167 - avoid LDLM oops when lock callback contained bad data
4169 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4171 * documentation updates
4172 * bug fixes and cleanups
4173 - configuration tools
4176 - 64-bit compile warnings
4177 - 64-bit internal statfs data
4179 * test_brw on persistent OST devices
4181 * lprocfs (disabled)
4183 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4185 * documentation updates
4186 * bug fixes and cleanups
4187 * configuration tools
4189 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4192 * many small fixes to 0_5_4
4193 * io/network handling
4194 * thinkos in MDS operations
4196 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4199 * crucial basic fixes to 0.5.3
4200 * IOR, Iozone work over Elan
4203 2002-08-07 Phil Schwan <phil@clusterfs.com>
4204 * version 0_5_3, our first alpha
4205 * we use the new Portals iovs
4206 * documentation updates
4207 * bug fixes and cleanups
4208 * small changes in the DLM wire protocol
4210 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4211 * version 0_5_1 with some initial stability,
4212 * locking on MD and file I/O.
4213 * documentation updates
4214 * several bug fixes since 0.5.0
4215 * small changes in wire protocol
4217 2002-07-18 Phil Schwan <phil@clusterfs.com>
4219 * delivered as Lustre Light Alpha
4220 * fixed a crash after handling invalid MDS requests
4221 * fixed directory pages for architectures with non-4k pages sizes
4223 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4224 * release version v0_4_4
4225 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4227 * Fixes a number of interruption problems with OST operations.
4228 * Update documentation for portals header changes
4229 * Move all wire protocol structs/defines to lustre_idl.h
4230 * Fixes symlink length bug.
4231 * Add tcpdump to repository.
4233 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4234 * release version v0_4_3
4235 * Fixes statfs for inodes on extN.
4236 * Fixes bug in runtests which would delete /etc/hosts.
4237 * Use 64-bit object IDs wherever possible (not into VFS though)
4238 Remove ost_get_info, which is unused by lustre, and out of date.
4240 2002-07-03 Peter Braam <braam@clusterfs.com>
4241 * release version v0_4_2 Fixes a lookup error (type not passed)
4242 * move forward to head of Portals
4243 * move forward to latest Lustre kernel
4245 2002-06-25 Peter Braam <braam@clusterfs.com>
4246 * release version v0_4_1. Hopefully stable on single node use.