3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.53-0.16 (SLES 10),
6 2.6.18-8.1.15.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs4
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: Inconsistent usage of lustre_pack_reply()
18 Details : Standardize the usage of lustre_pack_reply() such that it
19 always generate a CERROR on failure.
22 Frequency : occasional
24 Description: MDS hang or stay in waiting lock
25 Details : If client receive lock with CBPENDING flag ldlm need send lock
26 cancel as separate rpc, to avoid situation when cancel request
27 can't processed due all i/o threads stay in wait lock.
31 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
35 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
39 Description: Update to SLES9 kernel-2.6.5-7.287.3.
43 Description: Update to RHEL4 latest kernel.
45 Severity : enhancement
47 Description: Build SLES10 patchless client fails
48 Details : The configure was broken by run ./configure with
49 --with-linux-obj=.... argument for patchless client. When the
50 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
51 can't search header adequately. Use absolute path such as
52 -I($LINUX)/include instead.
56 Description: interrupt oig_wait produce painc on resend.
57 Details : brw_redo_request can be used for resend requests from ptlrpcd and
58 private set, and this produce situation when rq_ptlrpcd_data not
59 copyed to new allocated request and triggered LBUG on assert
60 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
61 ptlrpcd set if request is changed and can be safety changed to use
66 Description: LASSERT_{REQ,REP}SWAB macros are buggy
67 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
68 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
69 these macros and replace them with their difinitions instead.
73 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
74 Details : Update which_patch & target file for SLES10 latest kernel.
77 Description: Remove client patches from SLES 10 kernel.
78 Details : This causes SLES 10 clients to behave as patchless clients
79 even on a Lustre-patched (server) kernel.
81 Severity : enhancement
83 Description: self-adjustable client's lru lists
84 Details : use adaptive algorithm for managing client cached locks lru
85 lists according to current server load, other client's work
86 pattern, memory activities, etc. Both, server and client
87 side namespaces provide number of proc tunables for controlling
90 Severity : enhancement
92 Description: light-weight GSS support
93 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
94 and reduce performance overhead.
96 Severity : enhancement
98 Description: Linux keyring support
99 Details : Support using service of Linux keyring for Lustre GSS internal
100 context refresh/cache mechanism.
104 Description: Fix errors in lfs documentation
105 Details : Fixes man pages
109 Description: don't use GFP_* in generic Lustre code.
110 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
114 Description: obdclass is limited by single OBD_ALLOC(idarray)
115 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
119 Description: warnings with build patchless client with vanila 2.6.19 and up
120 Details : change old ctl_table style and replace ctl_table/ctl_table_header
121 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
125 Frequency : Only for SLES
126 Description: sanity_quota fail test_1
127 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
128 file, which makes TSTID[2] inunique.
132 Frequency : Only for SLES9
133 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
134 Details : Update target/ChangeLog/which_patch .
138 Frequency : Only for test_10 in sanity-quota.sh
139 Description: a bug in quota test code
140 Details : checking a test flag in a wrong place
144 Frequency : Only for test_13 in sanity-quota.sh
145 Description: a bug in quota test script
146 Details : shouldn't check overfull allocated quota
150 Description: jbd statistics
151 Details : Port older jbd statistics patch for sles10
153 Severity : enhancement
155 Description: Add printing inode info into message about error in writepage.
160 Description: reply_lock_interpret crash due to race with it and lock cancel.
161 Details : Do not replay locks that are being cancelled. Do not reference
162 locks by their address during replay, just by their handle.
166 Frequency : When flocks are used.
167 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
168 Details : Flock locks might destroy just granted lock if it could be merged
169 with another existing flock, this is done in completion handler,
170 so teach ldlm_cli_enquque_fini that this is a valid case for
176 Description: Oops in read and write path when failing to allocate lock.
177 Details : Check if lock allocation failed and return error back.
181 Description: lstripe command fails for valid OST index
182 Details : The stripe offset is compared to lov->desc.ld_tgt_count
183 instead of lov->desc.ld_active_tgt_count.
187 Description: sanity.sh failed test 103
188 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
193 Description: df doesn't work properly if diskfs blocksize != 4K
194 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
198 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
199 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
203 Description: obd_health_check_timeout too short
204 Details : set obd_health_check_timeout as 1.5x of obd_timeout
208 Description: llapi_file_create() does not allow some changes
209 Details : add llapi_file_open() that allows specifying the mode and
210 open flags, and also returns an open file handle.
214 Description: merge and cleanup kernel patches.
215 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
219 Description: Add journal checksum support.(Kernel part)
220 Details : The journal checksum feature adds two new flags i.e
221 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
222 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
223 indicates that the commit block contains the checksum for
224 the blocks described by the descriptor blocks. Now commit
225 record can be sent to disk without waiting for descriptor
226 blocks to be written to disk. This behavior is controlled
227 using JBD2_FEATURE_ASYNC_COMMIT flag.
231 Description: OSS needs mutliple precreate threads
232 Details : Add ability to start more than one create thread per OSS.
236 Description: Sanity.sh test_65k failed: setstripe should have succeeded
237 Details : execute lfs setstripe on client
241 Description: mds_obd_create error creating tmp object
242 Details : When the user sets quota on root, llog will be affected and can't
243 create files and write files.
246 Frequency : Always on ia64 patchless client, and possibly others.
248 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
249 Details : This allows the patchless client to be loaded on architectures
254 Description: RedHat Update kernel for RHEL5
255 Details : Add the kernel config file for RHEL5.
259 Description: RedHat Update kernel for RHEL5
260 Details : Modify the kernel config file more closer RHEL5.
264 Description: Build failure against Centos5 (RHEL5)
265 Details : Define PAGE_SIZE when it isn't present.
269 Description: test_7 on sanity-quota.sh can't be used on separate machines
270 Details : add facet to handle it
274 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
275 Details : with NFS, the anon dentry's parent was set to itself in
276 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
277 corresponding dentry other than use rec->ur_name.
279 Severity : enhancement
281 Description: lfs setstripe enhancement
282 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
286 Description: enable data checksumming by default
287 Details : enable checksum by default, allow --disable-checksum
288 configure option and "-o nochecksum" mount option. Checksums
289 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
293 Description: lustre support for RHEL5
294 Details : Add support for RHEL5.
298 Description: Client eviction due to failover config
299 Details : after a connection loss, the lustre client should attempt to
300 reconnect to the last active server first before trying the
301 other potential connections.
305 Description: when mds and osts use different quota unit(32bit and 64bit),
306 quota will be released repeatly.
307 Details : void sending multiple quota reqs to mds, which will keep the
308 status between the reqs.
312 Description: osts not allocated evenly to files
313 Details : change the condition to increase offset_idx
316 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
318 Description: Data corruption for OSTs that are formatted larger than 2TB
320 Details : When generating the bio request for lustre file writes the
321 sector number would overflow a temporary variable before being
322 used for the IO. The data reads correctly from Lustre (which
323 will overflow in a similar manner) but other file data or
324 filesystem metadata may be corrupted in some cases.
328 Description: Tune the kernel for good SCSI performance.
329 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
330 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
335 Description: rewrite ext2-derived code in obdclass/uuid.c
336 Details : rewrite inherited code (uuid parsing code from ext2 utils)
337 from scratch preserving functionality.
341 Description: "lfs find -obd UUID" prints directories
342 Details : "lfs find -obd UUID" will return all directory names instead
343 of just file names. It is incorrect because the directories
344 do not reside on the OSTs.
348 Description: use i_size_read and i_size_write in 2.6 port
349 Details : replace inode->i_size access with i_size_read/write()
352 Frequency : when removing large files
354 Description: scheduling issue during removal of large Lustre files
355 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
356 It causes scheduling issues when removing large files (17TB in the
360 Frequency : only with liblustre clients on XT3
362 Description: evictions taking too long
363 Details : allow llrd to evict clients directly on OSTs
366 Frequency : only on ppc
368 Description: /proc/fs/lustre/devices broken on ppc
369 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
370 We should make sure the type of 'index' is loff_t and then cast
371 explicitly as needed below. Do not assign an explicitly cast
375 Frequency : only for rhel5
377 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
378 Details : Modify the target file & which_kernel.
380 Severity : enhancement
382 Description: omit set fsid for export NFS
383 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
384 as FS_REQUIRES_DEV for avoid problems with generate fsid.
388 Frequency : Always, for kernels after 2.6.16
389 Description: Fix warning idr_remove called for id=.. which is not allocated.
390 Details : Last kernels save old s_dev before kill super and not allow
391 to restore from callback - restore it before call kill_anon_super.
395 Description: buffer overruns could theoretically occur
396 Details : llapi_semantic_traverse() modifies the "path" argument by
397 appending values to the end of the origin string, and a
398 overrun may occur. Adding buffer overrun check in liblustreapi.
402 Description: Fix error on 'ls .' at the top of the Lustre mount.
403 Details : Don't revalidate dentry if it is a root dentry.
407 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
408 Details : Modify vm-tunables-rhel4.patch.
412 Description: Kernel config for 2.6.18-vanilla.
413 Details : Modify targets/2.6-vanilla.target.in.
414 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
415 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
416 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
417 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
421 Description: improve handling recoverable errors
422 Details : if request processig with error which can be recoverable on server
423 request should be resend, otherwise page released from cache and
428 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
429 Details : Modify target file & which_patch.
430 A flaw was found in the IA32 system call emulation provided
431 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
432 value could be stored in the %RAX register, which could trigger an
433 out-of-bounds system call table access. An untrusted local user
434 could exploit this flaw to run code in the kernel
435 (ie a root privilege escalation). (CVE-2007-4573).
439 Description: O_DIRECT bypasses client statistics.
440 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
441 read_ahead_stats were updated. copied the stats section from
442 osc_send_oap_rpc() into async_internal().
446 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
450 Description: change order of libsysio includes
451 Details : '#include sysio.h' should always come before '#include xtio.h'
455 Description: handle "serious error: objid * already exists" more gracefully
456 Details : If LAST_ID value on disk is smaller than the objects existing in
457 the O/0/d* directories, it indicates disk corruption and causes an
458 LBUG(). If the object is 0-length, then we should use the existing
459 object. This will help to avoid a full fsck in most cases.
461 Severity : enhancement
463 Description: adapt the lustre_config script to support the upgrade case
464 Details : Add "-u" option for lustre_config script to support upgrading 1.4
465 server targets to 1.6 in parallel.
469 Description: To avoid grant space > avaible space when the disk is almost
470 full. Without this patch you might see the error "grant XXXX >
471 available" or some LBUG about grant, when the disk is almost
473 Details : In filter_check_grant, for non_grant cache write, we should
474 check the left space by if (*left > ungranted + bytes), instead
475 of (*left > ungranted), because only we are sure the left space
476 is enough for another "bytes", then the ungrant space should be
477 increase. In client, we should update cl_avail_grant only there
478 is OBD_MD_FLGRANT in the reply.
482 Description: Update RHEL 4 kernel to fix local root privilege escalation.
483 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
484 described in CVE-2007-4573. This problem could allow untrusted
485 local users to gain root access.
488 Frequency : when using O_DIRECT and quotas
490 Description: Incorrect file ownership on O_DIRECT output files
491 Details : block usage reported by 'lfs quota' does not take into account
492 files that have been written with O_DIRECT.
497 Description: touch file failed when fs is not full
498 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
499 otherwise we can get ENOSP while fs is not full.
503 Description: parallel lock callbacks
504 Details : Instead of sending blocking and completion callbacks as separated
505 requests, adding them to a set and sending in parallel.
507 --------------------------------------------------------------------------------
509 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
511 * Support for kernels:
512 2.6.5-7.283 (SLES 9),
513 2.6.9-55.EL (RHEL 4),
514 2.6.16.46-0.14 (SLES 10),
515 2.6.18.8 vanilla (kernel.org)
516 * Client support for unpatched kernels:
517 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
518 2.6.16 - 2.6.22 vanilla (kernel.org)
519 * Due to recently discovered recovery problems, we do not recommend
520 using patchless RHEL 4 clients with this or any earlier release.
521 * Recommended e2fsprogs version: 1.39.cfs8
522 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
523 * Starting with this release, the ldiskfs backing filesystem required
524 by Lustre is now in its own package, lustre-ldiskfs. This package
525 should be installed. It is versioned separately from Lustre and
526 may be released separately in future.
530 Description: block reactivating mgc import until all deactivates complete
531 Details : Fix race when failing back MDT/MGS to itself (testing)
534 Frequency : at statup only
536 Description: mds_lov_synchronize race leads to various problems
537 Details : simultaneous MDT->OST connections at startup can cause the
538 sync to abort, leaving the OSC in a bad state.
540 Severity : enhancement
542 Description: add optional extra BUILD_VERSION info
543 Details : add a new environment variable (namely LUSTRE_VERS) which allows
544 to override the lustre version.
547 Frequency : 2.6.18 servers only
549 Description: ll_kern_mount() doesn't release the module reference
550 Details : The ldiskfs module reference count never drops down to 0
551 because ll_kern_mount() doesn't release the module reference.
556 Description: server LBUG when using old ost_num_threads parameter
557 Details : Accept the old ost_num_threads parameter but warn that it
558 is deprecated, and fix an off-by-one error that caused an LBUG.
563 Description: Transient SCSI error results in persistent IO issue
564 Details : iobuf->dr_error is not reinitialized to 0 between two
568 Frequency : sometimes when underlying device returns I/O errors
570 Description: OSTs not going read-only during write failures
571 Details : OSTs are not remounted read-only when the journal commit threads
572 get I/O errors because fsfilt_ext3 calls journal_start/stop()
573 instead of the ext3 wrappers.
577 Description: poor connect scaling with increasing client count
578 Details : Don't run filter_grant_sanity_check for more than 100 exports
579 to improve scaling for large numbers of clients.
582 Frequency : SLES10 only
584 Description: sanity-quota.sh quotacheck failed: rc = -22
585 Details : Quotas cannot be enabled on SLES10.
588 Frequency : liblustre clients only
590 Description: getdirentries does not give error when run on compute nodes
591 Details : getdirentries does not fail when the size specified as an argument
592 is too small to contain at least one entry
594 Severity : enhancement
596 Description: Add LNET router traceability for debug purposes
597 Details : If a checksum failure occurs with a router as part of the
598 IO path, the NID of the last router that forwarded the bulk data
599 is printed so it can be identified.
604 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
605 Details : Due to a race condition, liblustre clients were occasionally
608 Severity : enhancement
610 Description: lfs setstripe use optional parameters instead of postional
613 Severity : enhancement
615 Description: Nanosecond timestamp support for ldiskfs
616 Details : The on-disk ldiskfs filesystem has added support for nanosecond
617 resolution timestamps. There is not yet support for this at
618 the Lustre filesystem level.
621 Frequency : during server recovery
623 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
624 Details : request with rq_no_resend flag not awake l_wait_event if they get a
628 Frequency : nfs export on patchless client
630 Description: connectathon hang when test nfs export over patchless client
631 Details : Disconnected dentry cannot be found with lookup, so we do not need
632 to unhash it or make it invalid
635 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
636 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
637 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
642 Description: open req refcounting wrong on reconnect
643 Details : If reconnect happened between getting open reply from server and
644 call to mdc_set_replay_data in ll_file_open, we will schedule
645 replay for unreferenced request that we are about to free.
646 Subsequent close will crash in variety of ways.
647 Check that request is still eligible for replay in
648 mdc_set_replay_data().
653 Description: disable writes to filesystem when reading health_check file
654 Details : the default for reading the health_check proc file has changed
655 to NOT do a journal transaction and write to disk, because this
656 can cause reads of the /proc file to hang and block HA state
657 checking on a healthy but otherwise heavily loaded system. It
658 is possible to return to the previous behaviour during configure
659 with --enable-health-write.
661 Severity : enhancement
663 Description: 64-bit inode version
664 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
665 to the inode. This will be required for version-based recovery.
670 Description: MDS fails to start if a duplicate client export is detected
671 Details : in some rare cases it was possible for a client to connect to
672 an MDS multiple times. Upon recovery the MDS would detect this
673 and fail during startup. Handle this more gracefully.
675 Severity : enhancement
677 Description: Add -o localflock option to simulate old noflock
679 Details : This will achieve local-only flock/fcntl locks
685 Description: log_commit_thread vs filter_destroy race leads to crash
686 Details : Take import reference before releasing llog record semaphore
691 Description: Wrong request locking in request set processing
692 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
693 imp_delayed_list, in this place should be used imp_lock.
696 Frequency : when reconnection
698 Description: Grant Leak when osc reconnect to OST
699 Details : When osc reconnect ost, OST(filter) should check whether it
700 should grant more space to client by comparing fed_grant and
701 cl_avail_grant, and return the granted space to client instead
702 of "new granted" space, because client will call osc_init_grant
703 to update the client grant space info.
706 Frequency : when client reconnect to OST
708 Description: Grant Leak when osc do resend and replay bulk write
709 Details : When osc reconnect to OST, OST(filter)should clear grant info of
710 bulk write request, because the grant info while be sync between
711 OSC and OST when reconnect, and we should ignore the grant info
712 these of resend/replay write req.
717 Description: Grant space more than avaiable left space sometimes.
718 Details : When then OST is about to be full, if two bulk writing from
719 different clients came to OST. Accord the avaliable space of the
720 OST, the first req should be permitted, and the second one
721 should be denied by ENOSPC. But if the seconde arrived before
722 the first one is commited. The OST might wrongly permit second
723 writing, which will cause grant space > avaiable space.
726 Frequency : when client is evicted
728 Description: Grant might be wrongly erased when osc is evicted by OST
729 Details : when the import is evicted by server, it will fork another
730 thread ptlrpc_invalidate_import_thread to invalidate the
731 import, where the grant will be set to 0. While the original
732 thread will update the grant it got when connecting. So if
733 the former happened latter, the grant will be wrongly errased
734 because of this race.
739 Description: Checking Stale with correct fid
740 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
741 to check whether it is stale, because sometimes, we want the
742 enqueue happened anyway, and op_data.fid2 will not be initialized.
744 Severity : enhancement
746 Description: update patchless client
747 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
750 Frequency : only with 2.4 kernel
752 Description: random memory corruption
753 Details : size of struct ll_inode_info is to big for union inode.u and this
754 can be cause of random memory corruption.
759 Description: Memory leak in recovery
760 Details : Lov_mds_md was not free in an error handler in mds_create_object.
761 It should also check obd_fail before fsfilt_start, otherwise if
762 fsfilt_start return -EROFS,(failover mds during mds recovery).
763 then the req will return with repmsg->transno = 0 and rc = EROFS.
764 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
765 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
766 be freed no matter whether fsfilt_commit success or not.
769 Frequency : only with huge count clients
771 Description: Prevents from taking the superblock lock in llap_from_page for
773 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
774 for llap_from_page call in ll_removepage prevents from taking the
775 superblock lock for a soon died page.
780 Description: Not check open intent error before release open handle
781 Details : in some rare cases, the open intent error is not checked before
782 release open handle, which may cause
783 ASSERTION(open_req->rq_transno != 0), because it tries to release
784 the failed open handle.
789 Description: Set cat log bitmap only after create log success.
790 Details : in some rare cases, the cat log bitmap is set too early. and it
791 should be set only after create log success.
795 Description: Accessing a block bevice can re-enable I/O when Lustre is
796 tearing down a device.
797 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
802 Description: service threads may hog cpus when there are a lot of requests
803 Details : Insert cond_resched to give other threads a chance to use some CPU
808 Description: the cat log was not initialized in recovery
809 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
810 unlink log on mds will not be initialized until mds post
811 recovery. And also in mds post recovery, the unlink log will
812 initialization will be done asynchronausly, so there will be race
813 between add unlink log and unlink log initialization.
817 Description: brw_stats were being printed incorrectly
818 Details : brw_stats were being printed as log2 but all of them were not
819 recorded as log2. Also remove some code duplication arising from
820 filter_tally_{read,write}.
824 Frequency : rare, only in recovery.
825 Description: ASSERTION(req->rq_type != LI_POISON) failed
826 Details : imp_lock should be held while iterating over imp_sending_list for
827 prevent destroy request after get timeout in ptlrpc_queue_wait.
831 Description: replay-single.sh test 52 fails
832 Details : A lock's skiplist need to be cleanup when it being unlinked
833 from its resource list.
837 Description: Short directio read returns full requested size rather than
839 Details : Direct I/O operations should return actual amount of bytes
840 transferred rather than requested size.
842 Severity : enhancement
844 Description: metadata RPC reduction (e.g. for rm performance)
845 Details : decrease the amount of synchronous RPC between clients and servers
846 by canceling conflicing lock before the operation on the client
847 and packing thier handles into the main operation RPC to server.
849 Severity : enhancement
851 Description: Async OSC create to avoid the blocking unnecessarily.
852 Details : If a OST has no remain object, system will block on the creating
853 when need to create a new object on this OST. Now, ways use
854 pre-created objects when available, instead of blocking on an
855 empty osc while others are not empty. If we must block, we block
856 for the shortest possible period of time.
858 Severity : enhancement
860 Description: refine locking for avoid write wrong info into lov_objid file
861 Details : fix possible races with add new target and write/update data in
864 --------------------------------------------------------------------------------
866 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
871 Frequency : on some architectures
873 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
874 Details : Uninitialized flags sometimes cause configuration commands to
878 Frequency : patchless clients only
880 Description: missing __iget() symbol export
881 Details : The __iget() symbol export is missing. To avoid the need for
882 this on patchless clients the deathrow inode reaper is turned
883 off, and we depend on the VM to clean up old inodes. This
884 dependency was during via the fix for bug 12181.
886 --------------------------------------------------------------------------------
888 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
890 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
891 INTEROPERATE with older versions automatically. In many cases a
892 special upgrade step is needed. Please read the
893 user documentation before upgrading any part of a live system.
894 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
895 version will not interoperate with 1.6 betas before beta5 (1.5.95).
896 * WARNING: Lustre configuration and startup changes are required with
897 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
903 Severity : enhancement
905 Description: MountConf
906 Details : Lustre configuration is now managed via mkfs and mount
907 commands instead of lmc and lconf. New obd types (MGS, MGC)
908 are added for dynamic configuration management. See
909 https://mail.clusterfs.com/wikis/lustre/MountConf for
912 Severity : enhancement
914 Description: dynamic OST addition
915 Details : OSTs can now be added to a live filesystem
917 Severity : enhancement
919 Description: startup order invariance
920 Details : MDTs and OSTs can be started in any order. Clients only
921 require the MDT to complete startup.
923 Severity : enhancement
925 Description: parallel, asynchronous orphan cleanup
926 Details : orphan cleanup is now performed in separate threads for each
927 OST, allowing parallel non-blocking operation.
929 Severity : enhancement
931 Description: optimized stripe assignment
932 Details : stripe assignments are now made based on ost space available,
933 ost previous usage, and OSS previous usage, in order to try
934 to optimize storage space and networking resources.
936 Severity : enhancement
938 Description: Permanently set tunables
939 Details : All writable /proc/fs/lustre tunables can now be permanently
940 set on a per-server basis, at mkfs time or on a live system.
942 Severity : enhancement
944 Description: Lustre message v2
945 Details : Add lustre message format v2.
947 Severity : enhancement
949 Description: client OST exclusion list
950 Details : Clients can be started with a list of OSTs that should be
951 declared "inactive" for known non-responsive OSTs.
955 Description: SPEC SFS validation failure on NFS v2 over lustre.
956 Details : Changes the blocksize for regular files to be 2x RPC size,
957 and not depend on stripe size.
959 Severity : enhancement
961 Description: Multiple MD RPCs in flight.
962 Details : Further unserialise some read-only MDS RPCs - learn about intents.
963 To avoid overly-overloading MDS, introduce a limit on number of
964 MDS RPCs in flight for a single client and add /proc controls
965 to adjust this limit.
967 Severity : enhancement
969 Description: client read/write statistics
970 Details : Add client read/write call usage stats for performance
971 analysis of user processes.
972 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
973 file access. extents_stats shows chunk size distribution.
974 extents_stats_per_process show chunk size distribution per
977 Severity : enhancement
979 Description: mds statistics
980 Details : Add detailed mds operations statistics in
981 /proc/fs/lustre/mds/*/stats.
985 Description: Failure of copying files with lustre special EAs.
986 Details : Client side always return success for setxattr call for lustre
987 special xattr (currently only "trusted.lov").
992 Description: Refcount LNET uuids
993 Details : The global LNET uuid list grew linearly with every startup;
994 refcount repeated list entries instead of always adding to
997 Severity : enhancement
999 Description: Dynamic service threads
1000 Details : Within a small range, start extra service threads
1001 automatically when the request queue builds up.
1004 Frequency : mixed-endian client/server environments
1006 Description: mixed-endian crashes
1007 Details : The new msg_v2 system had some failures in mixed-endian
1010 Severity : enhancement
1012 Description: Easy OST removal
1013 Details : OSTs can be permanently deactivated with e.g. 'lctl
1014 conf_param lustre-OST0001.osc.active=0'
1016 Severity : enhancement
1018 Description: MGS proc entries
1019 Details : Added basic proc entries for the MGS showing what filesystems
1022 Severity : enhancement
1024 Description: provide MGS failover
1025 Details : Added config lock reacquisition after MGS server failover.
1027 Severity : enhancement
1029 Description: add Linux 2.4 support
1030 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1034 Description: a large application tries to do I/O to the same resource and dies
1035 in the middle of it.
1036 Details : Check the req->rq_arrival time after the call to
1037 ost_brw_lock_get(), but before we do anything about
1038 processing it & sending the BULK transfer request. This
1039 should help move old stale pending locks off the queue as
1040 quickly as obd_timeout.
1043 Frequency : when an incorrect nid is specified during startup
1045 Description: ptlrpc connect to non-existant node causes kernel crash
1046 Details : LNET can't be re-entered from an event callback, which
1047 happened when we expire a message after the export has been
1048 cleaned up. Instead, hand the zombie cleanup off to another
1051 Severity : enhancement
1053 Description: plain/inodebits lock performance improvement
1054 Details : Grouping plain/inodebits in granted list by their request modes
1055 and bits policy, thus improving the performance of search through
1059 Frequency : only if OST filesystem is corrupted
1061 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1062 Details : for a short time RPCs with bulk IO are in the replay list,
1063 but replay of bulk IOs is unimplemented. If the OST filesystem
1064 is corrupted due to disk cache incoherency and then replay is
1065 started it is possible to trip an assertion. Avoid putting
1066 committed RPCs into the replay list at all to avoid this issue.
1069 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1071 Description: System hang on startup
1072 Details : This bug allowed the liblustre (e.g. catamount) client to
1073 return to the app before handling all startup RPCs. This
1074 could leave the node unresponsive to lustre network traffic
1075 and manifested as a server ptllnd timeout.
1077 Severity : enhancement
1079 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1080 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1081 boolean to control whether to print peer debug info when a
1082 client's RPC times out.
1085 Frequency : only for kernels with patches from Lustre below 1.4.3
1087 Description: Remove old rdonly API
1088 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1091 Frequency : only for devices with external journals
1093 Description: Set external device read-only also
1094 Details : During a commanded failover stop, we set the disk device
1095 read-only while the server shuts down. We now also set any
1096 external journal device read-only at the same time.
1099 Frequency : when upgrading from 1.4 while trying to change parameters
1101 Description: The wrong (new) MDC name was used when setting parameters for
1102 upgraded MDT's. Also allows changing of OSC (and MDC)
1103 parameters if --writeconf is specified at tunefs upgrade time.
1106 Frequency : when setting specific ost indicies
1108 Description: QOS code breaks on skipped indicies
1109 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1110 created with --index need not be sequential.
1112 Severity : enhancement
1114 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1115 Details : The uninit_groups feature works in conjunction with the kernel
1116 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1117 up the pass1 processing of e2fsck. This is a read-only feature
1118 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1119 mount filesystems that have had this feature enabled.
1121 Severity : enhancement
1123 Description: Improve multi-block allocation algorithm to avoid fragmentation
1124 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1125 allocation locality and avoid filesystem fragmentation.
1127 ------------------------------------------------------------------------------
1129 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1131 * Support for kernels:
1132 2.4.21-47.0.1.EL (RHEL 3)
1133 2.6.5-7.283 (SLES 9)
1134 2.6.9-42.0.10.EL (RHEL 4)
1135 2.6.12.6 vanilla (kernel.org)
1136 2.6.16.27-0.9 (SLES 10)
1137 * Recommended e2fsprogs version: 1.39.cfs5
1139 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1143 Frequency : occasional, depends on client load and configuration
1144 Bugzilla : 12181, 12203
1145 Description: data loss for recently-modified files
1147 Details : In some cases it is possible that recently written or created
1148 files may not be written to disk in a timely manner (this should
1149 normally be within 30s unless client IO load is very high).
1150 The problem appears as zero-length files or files that are a
1151 multiple of 1MB in size after a client crash or client eviction
1152 that are missing data at the end of the file.
1154 This problem is more likely to be hit on clients where files are
1155 repeatedly created and unlinked in the same directory, clients
1156 have a large amount of RAM, have many CPUs, the filesystem has
1157 many OSTs, the clients are rebooted frequently, and/or the files
1158 are not accessed by other nodes after being written.
1160 The presence of the problem can be detected by looking at
1161 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1162 smaller than the second (nr_unused) then dirty files will not
1163 be flushed automatically to disk. "sync; sleep 10" should be
1164 run several times on the node before unmounting it to update
1165 Lustre (this is also safe to run on nodes without this problem).
1167 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1168 that can cause this same problem, so customers using that kernel
1169 also need to update the kernel in addition to Lustre. In order
1170 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1172 It is normal that files written just before a client crash (less
1173 than 30s) may not yet have been flushed to disk, even for local
1177 Frequency : frequent on thin XT3 nodes
1179 Description: UUID collision on thin XT3 Linux nodes
1180 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1181 randomly, since we relied on an insufficiently-seeded PRNG.
1186 Description: OSS hangs after "All ost request buffers busy"
1187 Details : A deadlock between quota and journal operations caused OSS
1188 hangs after printing "All ost request buffers busy."
1191 Frequency : always on liblustre builds
1193 Description: Cleanup compiler warnings on liblustre
1196 Frequency : always on liblustre builds on XT3
1198 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1202 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1203 > 32000 subdirectories
1204 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1205 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1206 subdirectory count crosses 32000. This will aid e2fsck to
1207 correctly handle more than 32000 subdirectories.
1210 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1212 Description: System hang on startup
1213 Details : This bug allowed the liblustre (e.g. catamount) client to
1214 return to the app before handling all startup RPCs. This
1215 could leave the node unresponsive to lustre network traffic
1216 and manifested as a server ptllnd timeout.
1218 Severity : enhancement
1220 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1221 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1222 boolean to control whether to print peer debug info when a
1223 client's RPC times out.
1228 Description: make O_SYNC working on 2.6 kernels
1229 Details : 2.6 kernels use different method for mark pages for write,
1230 so need add a code to lustre for O_SYNC work.
1235 Description: Failure to close file and release space on NFS
1236 Details : Put inode details into lock acquired in ll_intent_file_open.
1237 Use mdc_intent_lock in ll_intent_open to properly
1238 detect all kind of errors unhandled by mdc_enqueue.
1243 Description: proc file read during shutdown sometimes raced obd removal,
1245 Details : Add lock to prevent obd access after proc file removal.
1248 Frequency : Only for files larger than 4GB on 32-bit clients.
1250 Description: improperly doing page alignment of locks
1251 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1252 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1257 Description: RPCs being resent when they shouldn't be
1258 Details : Some RPCs that should not be resent are being resent. This
1259 can cause inconsistencies in the RPC state machine. Do not
1260 resend such requests.
1263 Frequency : rare, only with NFS export
1265 Description: Crash on NFS re-export node
1266 Details : under very unusual load conditions an assertion is hit in
1267 ll_intent_file_open()
1270 Frequency : only if OST filesystem is corrupted
1272 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1273 Details : for a short time RPCs with bulk IO are in the replay list,
1274 but replay of bulk IOs is unimplemented. If the OST filesystem
1275 is corrupted due to disk cache incoherency and then replay is
1276 started it is possible to trip an assertion. Avoid putting
1277 committed RPCs into the replay list at all to avoid this issue.
1282 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1283 Details : Large single O_DIRECT read and write calls can fail to allocate
1284 a sufficiently large buffer to process the request. In case of
1285 allocation failure the allocation is retried with a smaller
1286 buffer and broken into smaller requests.
1288 Severity : enhancement
1290 Description: Add -o localflock option to simulate old noflock behaviour.
1291 Details : This will achieve local-only flock/fcntl locks coherentness.
1296 Description: versioning check is incomplete
1297 Details : Checking the version difference of client vs. server, report
1298 error if the gap is too big.
1303 Description: add support PG_writeback bit
1304 Details : add support for PG_writeback bit for Lustre, for more carefull
1305 work with page cache in 2.6 kernel. This also fix some deadlocks
1306 and remove hack for work O_SYNC with 2.6 kernel.
1308 Severity : enhancement
1310 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1311 Details : The uninit_groups feature works in conjunction with the kernel
1312 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1313 up the pass1 processing of e2fsck. This is a read-only feature
1314 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1315 mount filesystems that have had this feature enabled.
1317 Severity : enhancement
1319 Description: Improve multi-block allocation algorithm to avoid fragmentation
1320 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1321 allocation locality and avoid filesystem fragmentation.
1323 ------------------------------------------------------------------------------
1325 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1327 * Support for kernels:
1328 2.6.9-42.0.3.EL (RHEL 4)
1329 2.6.5-7.276 (SLES 9)
1330 2.4.21-47.0.1.EL (RHEL 3)
1331 2.6.12.6 vanilla (kernel.org)
1332 2.6.16.21-0.8 (SLES10)
1333 * Recommended e2fsprogs version: 1.39.cfs2-0
1335 * The backwards-compatible /proc/sys/portals symlink has been removed
1336 in this release. Before upgrading, please ensure that you change
1337 any configuration scripts or /etc/sysctl.conf files that access
1338 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1339 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1340 in advance of the upgrade on any system running Lustre 1.4.6 or
1341 newer, since /proc/sys/lnet was added in that version.
1342 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1346 Frequency : only when quota is used
1348 Description: avoid scanning export list for quota master
1349 Details : Change the algorithms to avoid scanning export list in order
1350 to improve the efficiency.
1353 Frequency : MDS failover only, very rarely
1355 Description: "went back in time" messages on mds failover
1356 Details : The greatest transno may be lost when the current operation
1357 finishes with an error (transno==0) and the client's last_rcvd
1358 record is over-written. Save the greatest transno in the
1359 mds_last_transno for this case.
1362 Frequency : always for specific kernels and striping counts
1364 Description: client may get "Matching packet too big" without ACL support
1365 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1366 when trying to access files in certain configurations. The
1367 clients should in fact be denied when mounting because they do
1368 not understand ACLs.
1371 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1373 Description: many clients connecting with IO in progress causes connect timeouts
1374 Details : Avoid synchronous journal commits to avoid delays caused by many
1375 clients connecting/disconnecting when bulk IO is in progress.
1376 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1377 OST_IO_PORTAL to avoid delays behind potentially many pending
1381 Frequency : occasionally with multiple writers to a single file
1383 Description: shared writes to file may result in wrong size reported by stat()
1384 Details : Allow growing of kms when extent lock is cancelled
1387 Frequency : always with random mmap IO to multi-striped file
1389 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1390 Details : Only if the hole is at the end of OST object so that kms is too
1391 small. Fix is to increase kms accordingly in ll_nopage.
1394 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1396 Description: writes to a missing object would leak memory on the OST
1397 Details : If there is an inconsistency between the MDS and OST filesystems,
1398 such that the MDS references an object that doesn't exist, writes
1399 to that object will leak memory due to incorrect cleanup in the
1400 error handling path, eventually running out of memory on the OST.
1405 Description: Creating too long symlink causes lustre errors
1406 Details : Check symlink and name lengths before sending requests to MDS.
1409 Frequency : only if flock is enabled (not on by default)
1411 Description: posix locks not released on fd closure on 2.6.9+
1412 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1413 kernels, this caused such locks not to be released on fd close and
1414 then assertions on fs unmount about still used locks.
1417 Frequency : MDS failover only, very rarely
1419 Description: clients may get ASSERTION(granted_lock != NULL)
1420 Details : When request was taking a long time, and a client was resending
1421 a getattr by name lock request. The were multiple lock requests
1422 with the same client lock handle and
1423 mds_getattr_name->fixup_handle_for_resent_request found one of the
1424 lock handles but later failed with ASSERTION(granted_lock != NULL).
1429 Description: handle->h_buffer_credits > 0, assertion failure
1430 Details : h_buffer_credits is zero after truncate, causing assertion
1431 failure. This patch extends the transaction or creates a new
1435 Frequency : NFS re-export or patchless client
1436 Bugzilla : 11179, 10796
1437 Description: Crash on NFS re-export node (__d_move)
1438 Details : We do not want to hash the dentry if we don't have a lock.
1439 But if this dentry is later used in d_move, we'd hit uninitialised
1440 list head d_hash, so we just do this to init d_hash field but
1441 leave dentry unhashed.
1444 Frequency : NFS re-export or patchless client
1446 Description: NFS exports has problem with symbolic link
1447 Details : lustre client didn't properly install dentry when re-exported
1448 to NFS or running patchless client.
1451 Frequency : NFS re-export or patchless client
1453 Description: Various nfs/patchless fixes.
1454 Details : fixes reuse disconected alias for lookup process - this fixes
1455 warning "find_exported_dentry: npd != pd",
1456 fix permission error with open files at nfs.
1457 fix apply umask when do revalidate.
1460 Frequency : occasional
1462 Description: Crash on NFS re-export node
1463 Details : calling clear_page() on the wrong pointer triggered oops in
1464 generic_mapping_read().
1467 Frequency : rarely, using O_DIRECT IO
1469 Description: unaligned directio crashes client with LASSERT
1470 Details : check for unaligned buffers before trying any requests.
1473 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1475 Description: stale data returned from RAID cache
1476 Details : If only a small amount of IO is done to the RAID device before
1477 reading it again it is possible to get stale data from the RAID
1478 cache instead of reading it from disk.
1481 Frequency : always for sles10 kernel
1483 Description: sles10 support
1484 Details : ll_follow_link: compile fixes and using of nd_set_link
1485 under newer kernels.
1488 Frequency : depends on arch, kernel and compiler version, always on sles10
1491 Description: recursive or deep enough symlinks cause stack overflow
1492 Details : getting rid of large stack-allocated variable in
1496 Frequency : depends on hardware
1498 Description: lustre write performance loss in the SLES10 kernel
1499 Details : the performance loss is caused by using of write barriers in the
1500 ext3 code. The SLES10 kernel turns barrier support on by
1501 default. The fix is to undo that change for ldiskfs.
1503 ------------------------------------------------------------------------------
1505 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1507 * Support for kernels:
1508 2.6.9-42.0.3EL (RHEL 4)
1509 2.6.5-7.276 (SLES 9)
1510 2.4.21-47.0.1.EL (RHEL 3)
1511 2.6.12.6 vanilla (kernel.org)
1515 Frequency : quota enabled and large files being deleted
1517 Description: releasing more than 4GB of quota at once hangs OST
1518 Details : If a user deletes more than 4GB of files on a single OST it
1519 will cause the OST to spin in an infinite loop. Release
1520 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1525 Description: statfs data retrieved from /proc may be stale or zero
1526 Details : When reading per-device statfs data from /proc, in the
1527 {kbytes,files}_{total,free,avail} files, it may appear
1528 as zero or be out of date.
1531 Frequency : systems with MD RAID1 external journal devices
1533 Description: lconf's call to blkid is confused by RAID1 journal devices
1534 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1535 of returning all block devices that match the journal UUID.
1538 Frequency : always, for aggregate stripe size over 4GB
1540 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1541 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1542 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1545 Frequency : always if "lfs find" used on a local file/directory
1547 Description: "lfs find" segfaults if used on a local file/directory
1548 Details : The case where a directory component was not specified wasn't
1549 handled correctly. Handle this properly.
1552 Frequency : always on ppc64
1554 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1555 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1556 perform endianness conversion.
1559 Frequency : rarely (truncate to non-zero file size after write under load)
1560 Bugzilla : 10730, 10687
1561 Description: Files padded with zeros to next 4K multiple
1562 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1563 it is possible that files that are truncated to a non-zero size
1564 immediately after being written are filled with zero bytes beyond
1565 the truncated size. No file data is lost.
1567 Severity : enhancement
1569 Description: Allow recovery/failover for liblustre clients.
1570 Details : liblustre clients were unaware of failover configurations until
1573 Severity : enhancement
1575 Description: user file locks should fail when not mounting with flock option
1576 Details : Set up an error-returning stub in ll_file_operations.lock field
1577 to prevent incorrect behaviour when client is mounted without
1578 flock option. Also, set up properly f_op->flock field for
1582 Frequency : always on ia64
1584 Description: "lfs df" loops on printing out MDS statfs information
1585 Details : The obd_ioctl_data was not initialized and in some systems
1586 this caused a failure during the ioctl that did not return
1587 an error. Initialize the struct and return an error on failure.
1590 Frequency : SLES 9 only
1592 Description: Error of copying files with lustre special EAs as root
1593 Details : Client side always return success for setxattr call for lustre
1594 special xattr (currently only "trusted.lov").
1597 Frequency : rarely on clusters with both ia64+i386 clients
1599 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1600 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1601 clients are doing concurrent writes to the same file it is
1602 possible that the ia64 clients may LASSERT because the OST
1603 extent locks are not PAGE_SIZE aligned. Ensure that grown
1604 locks are always aligned on the request boundary.
1607 Frequency : specific use, occasional
1609 Description: Overwriting in use executable truncates on-disk binary image
1610 Details : If one node attempts to overwrite an executable in use by
1611 another node, we now correctly return ETXTBSY instead of
1612 truncating the file.
1614 Severity : enhancement
1616 Description: Async OSC create to avoid the blocking unnecessarily.
1617 Details : If a OST has no remain object, system will block on the creating
1618 when need to create a new object on this OST. Now, ways use
1619 pre-created objects when available, instead of blocking on an
1620 empty osc while others are not empty. If we must block, we block
1621 for the shortest possible period of time.
1626 Description: chmod on Lustre root is propagated to other clients
1627 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1628 invalid by the follow_mount time.
1631 Frequency : liblustre clients only
1633 Description: Race in 'instant cancel' lock handling could lead to such locks
1634 never to be granted in case of SMP MDS
1635 Details : Do not destroy not yet granted but cbpending locks in
1639 Frequency : replay/resend of open
1641 Description: non null lock assetion failure in mds_intent_policy
1642 Details : Trying to replay/resend lockless open requests resulted in
1643 mds_open() returning 0 with no lock. Now it sets a flag if
1644 a lock is going to be returned.
1646 Severity : enhancement
1648 Description: Checksum enhancements
1649 Details : New checksum enhancements allow for resending RPCs that failed
1652 Severity : enhancement
1654 Description: Tunables on number of dirty pages in cacche
1655 Details : Allow to set limit on number of dirty pages cached.
1660 Description: client crash on unmount - lock still has references
1661 Details : In some error handling cases it was possible to leak a lock
1662 reference on a client while accessing a file. This was not
1663 harmful to the client during operation, but would cause the
1664 client to crash when the filesystem is unmounted.
1667 Frequency : specific case, rare
1669 Description: ETXTBSY on mds though file not in use by client
1670 Details : ETXTBSY is no longer incorrectly returned when attempting to
1671 chmod or chown a directory that the user previously tried to
1672 execute or a currently-executing binary.
1675 Frequency : extremely rare except on liblustre-based clients
1677 Description: Lustre space not freed when files are deleted
1678 Details : Clean up open-unlinked files after client eviction. Previously
1679 the unlink was skipped and the files remained as orphans.
1684 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1685 Details : In some cases it was possible to send a blocking callback to
1686 a client doing a glimpse, even though that client didn't get
1687 a lock granted. When the glimpse lock is cancelled on the OST
1688 the freed lock is left on the waiting list and corrupted the list.
1691 Frequency : all core dumps
1693 Description: Broke core dumps to lustre
1694 Details : Negative dentry may be unhashed if parent does not have UPDATE
1695 lock, but some callers, e.g. do_coredump, expect dentry to be
1696 hashed after successful create, hash it in ll_create_it.
1698 ------------------------------------------------------------------------------
1700 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1702 * Support for kernels:
1703 2.6.9-42.0.2.EL (RHEL 4)
1704 2.6.5-7.276 (SLES 9)
1705 2.4.21-40.EL (RHEL 3)
1706 2.6.12.6 vanilla (kernel.org)
1710 Frequency : always on RHEL 3
1712 Description: Number of open files grows over time
1713 Details : The number of open files grows over time, whether or not
1714 Lustre is started. This was due to a filp leak introduced
1715 by one of our kernel patches.
1717 ------------------------------------------------------------------------------
1719 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1721 * Support for kernels:
1722 2.6.9-42.EL (RHEL 4)
1723 2.6.5-7.276 (SLES 9)
1724 2.4.21-40.EL (RHEL 3)
1725 2.6.12.6 vanilla (kernel.org)
1730 Bugzilla : 5719, 9635, 9792, 9684
1731 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1732 Details : If a server is under heavy load and cannot reply to new
1733 connection requests before the client resends the (re)connect,
1734 the connection handling code can behave badly if two service
1735 threads are concurrently handing separate (re)connections from
1736 the same client. Add better locking to the connection handling
1737 code, and ensure that only a single connection will be processed
1738 for a given client UUID, even if the lock is dropped.
1740 Severity : enhancement
1742 Description: add TCP zero-copy support to kernel
1743 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1744 sends if the hardware supports scatter-gather and checksumming.
1745 This allows socklnd to do client-write and server-read more
1746 efficiently and reduce CPU utilization from skbuf copying.
1749 Frequency : only if NFS exporting from client
1751 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1752 Details : If setting attributes on a file created under NFS that had
1753 never been opened it would be possible to oops the client
1754 if the file had no objects.
1757 Frequency : always for liblustre
1759 Description: liblustre client does MDS+OSTs setattr RPC for each write
1760 Details : When doing a write from a liblustre client, the client
1761 incorrectly issued an RPC to the MDS and each OST the file was
1762 striped over in order to update the timestamps. When writing
1763 with small chunks and many clients this could overwhelm the MDS
1764 with RPCs. In all cases it would slow down the write because
1765 these RPCs are unnecessary.
1767 Severity : enhancement
1769 Description: allow number of MDS service threads to be changed at module load
1770 Details : It is now possible to change the number of MDS service threads
1771 running. Adding "options mds mds_num_threads={N}" to the MDS's
1772 /etc/modprobe.conf will set the number of threads for the next
1773 time Lustre is restarted (assuming the "mds" module is also
1774 reloaded at that time). The default number of threads will
1775 stay the same, 32 for most systems.
1780 Description: OST crash if filesystem is unformatted or corrupt
1781 Details : If an OST is started on a device that has never been formatted
1782 or if the filesystem is corrupt and cannot even mount then the
1783 error handling cleanup routines would dereference a NULL pointer.
1788 Description: NULL pointer deref in llap_from_page.
1789 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1790 incorrect) mapping if the page was truncated/reclaimed while it was
1791 searched for. Check for this condition and skip such pages when
1792 doing readahead. Introduce extra check to llap_from_page() to
1793 verify page->mapping->host is non-NULL (so page is not anonymous).
1796 Frequency : Sometimes when using sys_sendfile
1798 Description: "page not covered by a lock" warnings from ll_readpage
1799 Details : sendfile called ll_readpage without right page locks present.
1800 Now we introduced ll_file_sendfile that does necessary locking
1801 around call to generic_file_sendfile() much like we do in
1805 Frequency : with certain MDS communication failures at client mount time
1807 Description: NULL pointer deref after failed client mount
1808 Details : a client connection request may delayed by the network layer
1809 and not be sent until after the PTLRPC layer has timed out the
1810 request. If the client fails the mount immediately it will try
1811 to clean up before the network times out the request. Add a
1812 reference from the request import to the obd device and delay
1813 the cleanup until the network drops the request.
1816 Frequency : occasionally during client (re)connect
1818 Description: assertion failure during client (re)connect
1819 Details : processing a client connection request may be delayed by the
1820 client or server longer than the client connect timeout. This
1821 causes the client to resend the connection request. If the
1822 original connection request is replied in this interval, the
1823 client may trip an assertion failure in ptlrpc_connect_interpret()
1824 which thought it would be the only running connect process.
1827 Frequency : only with obd_echo servers and clients that are rebooted
1829 Description: kernel BUG accessing uninitialized data structure
1830 Details : When running an obd_echo server it did not start the ping_evictor
1831 thread, and when a client was evicted an uninitialized data
1832 structure was accessed. Start the ping_evictor in the RPC
1833 service startup instead of the OBD startup.
1835 Severity : enhancement
1836 Bugzilla : 10193 (patchless)
1837 Description: Remove dependency on various unexported kernel interfaces.
1838 Details : No longer need reparent_to_init, exit_mm, exit_files,
1839 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1842 Frequency : rare (only users of deprecated and unsupported LDAP config)
1844 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1845 Details : LDAP apparently contains 'lustreName' attributes instead of
1846 'name'. A simple remapping of the name is sufficient.
1849 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1851 Description: waiting_locks_callback trips kernel BUG if client is evicted
1852 Details : Running with the dump_on_timeout debug flag turned on makes
1853 it possible that the waiting_locks_callback() can try to dump
1854 the Lustre kernel debug logs from an interrupt handler. Defer
1855 this log dumping to the expired_lock_main() thread.
1857 Severity : enhancement
1859 Description: Support NFS exporting on 2.6 kernels.
1860 Details : Implement non-rawops metadata methods for NFS server to use without
1861 changing NFS server code.
1864 Frequency : very rare (synthetic metadata workload only)
1866 Description: two racing renames might cause an MDS thread to deadlock
1867 Details : Running the "racer" program may cause one MDS thread to rename
1868 a file from being the source of a rename to being the target of
1869 a rename at exactly the same time that another thread is doing
1870 so, and the second thread has already enqueued these locks after
1871 doing a lookup of the target and is trying to relock them in
1872 order. Ensure that we don't try to re-lock the same resource.
1875 Frequency : only very large systems with liblustre clients
1877 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1878 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1879 all OSTs in parallel. This allows the eviction of stale liblustre
1880 clients to proceed much faster than if they were done in series,
1881 and also offers similar improvements for other set_info RPCs.
1886 Description: excessive CPU usage during initial read phase on client
1887 Details : During the initial read phase on a client, it would agressively
1888 retry readahead on the file, consuming too much CPU and impacting
1889 performance (since 1.4.5.8). Improve the readahead algorithm
1890 to avoid this, and also improve some other common cases (read
1891 of small files in particular, where "small" is files smaller than
1892 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1897 Description: MDS crash when receiving packet with unknown intent.
1898 Details : Do not LBUG in unknown intent case, just return -EFAULT
1900 Severity : enhancement
1901 Bugzilla : 9293, 9385
1902 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1903 Details : Do not serialize getattr (non-intent version) and statfs.
1906 Frequency : occasional, when OST network is overloaded/intermittent
1908 Description: client evicted by OST after bulk IO timeout
1909 Details : If a client sends a bulk IO request (read or write) the OST
1910 may evict the client if it is unresposive to its data GET/PUT
1911 request. This is incorrect if the network is overloaded (takes
1912 too long to transfer the RPC data) or dropped the OST GET/PUT
1913 request. There is no need to evict the client at all, since
1914 the pinger and/or lock callbacks will handle this, and the
1915 client can restart the bulk request.
1918 Frequency : Always when mmapping file with no objects
1920 Description: client crashes when mmapping file with no objects
1921 Details : Check that we actually have objects in a file before doing any
1922 operations on objects in ll_vm_open, ll_vm_close and
1928 Description: Request leak when working with deleted CWD
1929 Details : Introduce advanced request refcount tracking for requests
1930 referenced from lustre intent.
1932 Severity : Enhancement
1934 Description: Cache open file handles on client.
1935 Details : MDS now will return special lock along with openhandle, if
1936 requested and client is allowed to hold openhandle, even if unused,
1937 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1938 closing files for every read/write openration.
1940 Severity : Enhancement
1942 Description: Cache open negative dentries on client when possible.
1943 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1944 negative dentries on lock revocation.
1949 Description: Remounting a client read-only wasn't possible with a zconf mount
1950 Details : It wasn't possible to remount a client read-only with llmount.
1952 Severity : enhancement
1953 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1954 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1955 MPICH is included to add Lustre-specific ADIO interfaces.
1956 This is based closely on the UFS ADIO layer and only differs
1957 in file creation, in order to allow the OST striping to be set.
1958 This is user-contributed code and not supported by CFS.
1963 Description: extended inode attributes (immutable, append-only) work improperly
1964 when 2.4 and 2.6 kernels are used on client/server or vice versa
1965 Details : Introduce kernel-independent values for these flags.
1967 Severity : enhancement
1970 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1971 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1972 granularity. Now, Fractional MB granularity is be supported,
1973 this is very useful for low memory system.
1975 Severity : enhancement
1977 Description: Getattr by fid
1978 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1979 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1980 have locks and avoids getattr rpc requests for it on every stat.
1983 Frequency : Always, for filesystems larger than 2TB
1985 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1986 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1987 filesystems larger than 2TB to be mounted with mballoc enabled.
1990 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1992 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1993 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1994 filesystems larger than 2TB is mounted then the kernel will
1995 silently corrupt the start of the filesystem. CONFIG_LBD is
1996 enabled for all CFS-supported kernels, but the possibility of
1997 this happening with a modified kernel config exists.
1999 Severity : enhancement
2001 Description: add client O_DIRECT support for 2.6 kernels
2002 Details : It is now possible to do O_DIRECT reads and writes to files
2003 in the Lustre client mountpoint on 2.6 kernel clients.
2005 Severity : enhancement
2007 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2008 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2009 parallel, not waiting for response from every OST before sending
2010 a rpc to the next OST.
2015 Description: setattr vs write race when updating file timestamps
2016 Details : Client processes that update a file timestamp into the past
2017 right after writing to the file (e.g. tar) it is possible that
2018 the updated file modification time can be reset to the current
2019 time due to a race between processing the setattr and write RPC.
2021 Severity : enhancement
2023 Description: Bring 'lfs find' closer in line with regular Linux find.
2024 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2025 -print0 options and obtains all the needed info through the lustre
2028 Severity : enhancement
2030 Description: support up to 1024 configured devices on one node
2031 Details : change obd_dev array from statically allocated to dynamically
2032 allocated structs as they are first used to reduce memory usage
2037 Description: Flush dirty partially truncated pages during truncate
2038 Details : Immediatelly flush partially truncated pages in filter_setattr,
2039 this way we completely avoid having any pages in page cache on OST
2040 and can retire ugly workarounds during writes to flush such pages.
2045 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2046 Details : protect inode from truncation within vfs_unlink() context
2047 just take a reference before calling vfs_unlink() and release it
2048 when parent's i_sem is free.
2053 Description: last_id value checked outside lock on OST caused LASSERT failure
2054 Details : If there were multiple MDS->OST object precreate requests in
2055 flight, it was possible that the OST's last object id was checked
2056 outside a lock and incorrectly tripped an assertion. Move checks
2057 inside locks, and discard old precreate requests.
2060 Frequency : always, if extents are used on OSTs
2062 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2063 Details : OSTs using the extents format would not zero the high 16 bits of
2064 the index physical block number. This is not a problem for any
2065 OST filesystems smaller than 16TB, and no kernels support ext3
2066 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2067 new/modified files) and can be fixed for existing filesystems
2068 with e2fsprogs-1.39-cfs1.
2073 Description: import connection selection may be incorrect if timer wraps
2074 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2075 import connections to be ignored if the 32-bit jiffies counter
2076 wraps. Use a 64-bit jiffies counter.
2079 Frequency : very large clusters immediately after boot
2081 Description: LNET request buffers exhausted under heavy short-term load
2082 Details : If a large number of client requests are generated on a service
2083 that has previously never seen so many requests it is possible
2084 that the request buffer growth cannot keep up with the spike in
2085 demand. Instead of dropping incoming requests, they are held in
2086 the LND until the RPC service can accept more requests.
2089 Frequency : Sometimes during replay
2091 Description: Assertion failure in ll_local_open after replay.
2092 Details : If replay happened on an open request reply before we were able
2093 to set replay handler, reply will become not swabbed tripping the
2094 assertion in ll_local_open. Now we set the handler right after
2095 recognising of open request
2098 Frequency : very rare
2100 Description: kernel reports "badness in vsnprintf"
2101 Details : Reading from the "recovery_status" /proc file in small chunks
2102 may cause a negative length in lprocfs_obd_rd_recovery_status()
2103 call to vsnprintf() (which is otherwise harmless). Exit early
2104 if there is no more space in the output buffer.
2106 Severity : enhancement
2108 Description: clear OBD RPC statistics by writing to them
2109 Details : It is now possible to clear the OBD RPC statistics by writing
2110 to the "stats" file.
2115 Description: Client mtime is not the same on different clients after utimes
2116 Details : In some cases, the client was using the utimes() syscall on
2117 a file cached on another node. The clients now validate the
2118 ctime from the MDS + OSTs to determine which one is right.
2123 Description: Inability to activate failout mode
2124 Details : lconf script incorrectly assumed that in pythong string's numeric
2125 value is used in comparisons.
2128 Frequency : always with multiple stripes per file
2130 Description: Inefficient object allocation for mutli-stripe files
2131 Details : When selecting which OSTs to stripe files over, for files with
2132 a stripe count that divides evenly into the number of OSTs,
2133 the MDS is always picking the same starting OST for each file.
2134 Return the OST selection heuristic to the original design.
2139 Description: mount failures may take full timeout to return an error
2140 Details : Under some heavy load conditions it is possible that a
2141 failed mount can wait for the full obd_timeout interval,
2142 possibly several minutes, before reporting an error.
2143 Instead return an error as soon as the status is known.
2145 Frequency : quota enabled and large files being deleted
2147 Description: releasing more than 4GB of quota at once hangs OST
2148 Details : If a user deletes more than 4GB of files on a single OST it
2149 will cause the OST to spin in an infinite loop. Release
2150 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2155 Description: statfs data retrieved from /proc may be stale or zero
2156 Details : When reading per-device statfs data from /proc, in the
2157 {kbytes,files}_{total,free,avail} files, it may appear
2158 as zero or be out of date.
2161 Frequency : systems with MD RAID1 external journal devices
2163 Description: lconf's call to blkid is confused by RAID1 journal devices
2164 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2165 of returning all block devices that match the journal UUID.
2168 Frequency : always, for aggregate stripe size over 4GB
2170 Description: assertion fails when trying to use 4GB stripe size
2171 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2172 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2175 Frequency : always on ppc64
2177 Description: the first write on an ext3 filesystem with mballoc got stuck
2178 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2179 perform endianness conversion.
2181 ------------------------------------------------------------------------------
2183 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2185 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2186 INTEROPERATE with older versions automatically. Please read the
2187 user documentation before upgrading any part of a live system.
2188 * WARNING: Lustre networking configuration changes are required with
2189 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2192 * Support for kernels:
2193 2.6.9-22.0.2.EL (RHEL 4)
2194 2.6.5-7.244 (SLES 9)
2195 2.6.12.6 vanilla (kernel.org)
2198 Severity : enhancement
2199 Bugzilla : 7981/8208
2200 Description: Introduced Lustre Networking (LNET)
2201 Details : LNET is new networking infrastructure for Lustre, it includes
2202 a reorganized network configuration mode (see the user
2203 documentation for full details) as well as support for routing
2204 between different network fabrics. Lustre Networking Devices
2205 (LNDs) for the supported network fabrics have also been
2206 created for this new infrastructure.
2208 Severity : enhancement
2209 Description: Introduced Access control lists
2210 Details : clients can set ACLs on files and directories in order to have
2211 more fine-grained permissions than the standard Unix UGO+RWX.
2212 The MDS must be started with the "-o acl" mount option.
2214 Severity : enhancement
2215 Description: Introduced filesystem quotas
2216 Details : Administrators may now establish per-user quotas on the
2219 Severity : enhancement
2221 Description: Configuration change for the XT3
2222 The PTLLND is now used to run Lustre over Portals on the XT3
2223 The configure option(s) --with-cray-portals are no longer used.
2224 Rather --with-portals=<path-to-portals-includes> is used to
2225 enable building on the XT3. In addition to enable XT3 specific
2226 features the option --enable-cray-xt3 must be used.
2231 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2232 Details : A race between allocating a new llcd and re-getting the llcd_lock
2233 allowed another thread to grab newly-allocated llcd.
2235 Severity : enhancement
2237 Description: 2.6 OST async journal commit and locking fix to improve performance
2238 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2239 2.6 kernels are now async as they already were in 2.4 kernels so
2240 that they can commit concurrently with the network bulk transfer.
2241 For block-allocated files the filter allocation semaphore is held
2242 to avoid filesystem fragmentation during allocation. BKL lock
2243 removed for 2.6 xattr operations where it is no longer needed.
2248 Description: lconf incorrectly determined whether two IP networks could talk
2249 Details : In some more complicated routing and multiple-network
2250 configurations, lconf will avoid trying to make a network
2251 connection to a disjoint part of the IP space. It was doing the
2252 math incorrectly for one set of cases.
2257 Description: Fix for potential infinite loop processing records in an llog.
2258 Details : If an llog record is corrupted/zeroed, it is possible to loop
2259 forever in llog_process(). Validate the llog record length
2260 and skip the remainder of the block on error.
2263 Frequency : occasional (liblustre only)
2265 Description: liblustre could not open files whose last component is a symlink
2266 Details : sysio_path_walk() would incorrectly pass the open intent to
2267 intermediate path components.
2270 Frequency : rare (liblustre only with non-standard tuning)
2271 Bugzilla : 7201 (7350)
2272 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2273 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2274 after it has been released, only do this for referenced locks
2276 Severity : enhancement
2278 Description: specifying an (invalid) directory default stripe_size of -1
2279 would reset the directory default striping
2280 Details : stripe_size -1 was used internally to signal directory stripe
2281 removal, now use "all default" to signal dir stripe removal
2282 as a directory striping of "all default" is not useful
2285 Frequency : common for large clusters running liblustre clients
2287 Description: doing an ls when liblustre clients are running is slow
2288 Details : sending a glimpse AST to a liblustre client waits for every AST
2289 to time out, as liblustre clients will not respond. Since they
2290 cannot cache data we refresh the OST lock LVB from disk instead.
2292 Severity : enhancement
2294 Description: doing an ls at the same time as file IO can be slow
2295 Details : enqueue and other "small" requests can be blocked behind many
2296 large IO requests. Create a new OST IO portal for non-IO
2297 requests so they can be processed faster.
2300 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2302 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2303 created file with mode 000
2304 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2305 allow an ftruncate() as a non-root user to a file with mode 000.
2306 Lustre now respects this flag to disable mode checks when
2307 truncating a file owned by the user
2310 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2313 Description: Revoking locks from clients that went dead or catatonic might take
2315 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2316 cancellation of such locks instant on servers without waiting for
2317 any reply from clients. Clients drops these locks when cancel
2318 notification from server is received without replying.
2321 Frequency : liblustre-only, when liblustre client dies or becomes busy
2323 Description: Doing ls on Linux clients can take a long time with active
2325 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2326 granting such locks to it in the first place if possible. Locks
2327 are taken by proxy on the OST during the read or write and
2328 dropped immediately afterward. Add connect flags handling, do
2329 not grant locks to liblustre clients for glimpse ASTs.
2331 Severity : enhancement
2333 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2334 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2335 sequential reads and behaves badly with applications doing random
2336 reads. Improve it by reading ahead at least read region, and
2337 avoiding excessive large RPC for small reads.
2339 Severity : enhancement
2341 Description: Creating more than 1000 files for a single job may cause a load
2342 imbalance on the OSTs if there are also a large number of OSTs.
2343 Details : qos_prep_create() uses an OST index reseed value that is an
2344 even multiple of the number of available OSTs so that if the
2345 reseed happens in the middle of the object allocation it will
2346 still utilize the OSTs as uniformly as possible.
2351 Description: OST or MDS may oops in ping_evictor_main()
2352 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2353 but doesn't restart at beginning of obd_exports_timed list
2356 Severity : enhancement
2358 Description: improve by-nid export eviction on the MDS and OST
2359 Details : allow multiple exports with the same NID to be evicted at one
2360 time without re-searching the exports list.
2363 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2365 Description: MDS may oops in groups_free()
2366 Details : in rare race conditions a newly allocated group_info struct is
2367 freed again, and this can be NULL. The 2.4 compatibility code
2368 for groups_free() checked for a NULL pointer, but 2.6 did not.
2371 Frequency : common for liblustre clients doing little filesystem IO
2372 Bugzilla : 9352, 7313
2373 Description: server may evict liblustre clients accessing contended locks
2374 Details : if a client is granted a lock or receives a completion AST
2375 with a blocking AST already set it would not reply to the AST
2376 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2379 Frequency : lfs setstripe, only systems with more than 160 OSTs
2381 Description: unable to set striping with a starting offset beyond OST 160
2382 Details : llapi_create_file() incorrectly limited the starting stripe
2383 index to the maximum single-file stripe count.
2386 Frequency : LDAP users only
2388 Description: lconf did not handle in-kernel recovery with LDAP properly
2389 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2391 Severity : enhancement
2393 Description: bind OST threads to NUMA nodes to improve performance
2394 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2395 node and do their allocations there to localize memory access
2397 Severity : enhancement
2399 Description: llmount can determine client NID directly from Myrinet (GM)
2400 Details : the client NID code from gmnalnid was moved directly into
2401 llmount, removing the need to use this or specifying the
2402 client NID explicitly when mounting GM clients with zeroconf
2405 Frequency : if client is started with down MDS
2407 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2408 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2409 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2410 ptlrpc_import_delayed_req() because we didn't check if the
2411 request was interrupted, and we also didn't break out of the
2412 event loop if there was no timeout
2417 Description: data loss during non-page-aligned writes to a single file from
2418 both multiple nodes and multiple threads on one node at same time
2419 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2420 inconsistency led to false short-reads, that were cached and later
2421 used by ->prepare_write() to fill in partially written page,
2422 leading to data loss.
2425 Frequency : always, if lconf --abort_recovery used
2427 Description: lconf --abort_recovery fails with 'Operation not supported'
2428 Details : lconf was attempting to abort recovery on the MDT device and not
2431 Severity : enhancement
2433 Description: remove cleanup logs
2434 Details : replace lconf-generated cleanup logs with lustre internal
2435 cleanup routines. Eliminates the need for client-cleanup and
2438 Severity : enhancement
2440 Description: add support for EAs (user and system) on lustre filesystems
2441 Details : it is now possible to store extended attributes in the Lustre
2442 client filesystem, and with the user_xattr mount option it
2443 is possible to allow users to store EAs on their files also
2445 Severity : enhancement
2447 Description: Add possibility (config option) to show minimal available OST free
2449 Details : When compiled with --enable-mindf configure option, statfs(2)
2450 (and so, df) will return least minimal free space available from
2451 all OSTs as amount of free space on FS, instead of summary of
2452 free spaces of all OSTs.
2454 Severity : enhancement
2456 Description: do not expand extent locks acquired on OST-side
2457 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2458 by server: they are not cached anyway.
2461 Frequency : when mmap is used/binaries executed from Lustre
2463 Description: Unmmap pages before throwing them away from read cache.
2464 Details : llap_shrink cache now attempts to unmap pages before discarding
2465 them (if unmapping failed - do not discard). SLES9 kernel has
2466 extra checks that trigger if this unmapping is not done first.
2471 Description: lconf didn't resolve symlinks before checking to see whether a
2472 given mountpoint was already in use
2475 Frequency : when migrating failover services
2476 Bugzilla : 6395, 9514
2477 Description: When migrating a subset of services from a node (e.g. failback
2478 from a failover service node) the remaining services would
2479 time out and evict clients.
2480 Details : lconf --force (implied by --failover) sets the global obd_timeout
2481 to 5 seconds in order to quickly disconnect, but this caused
2482 other RPCs to time out too quickly. Do not change the global
2483 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2485 Severity : enhancement
2486 Frequency : if MDS is started with down OST
2487 Bugzilla : 9439,5706
2488 Description: Allow startup/shutdown of an MDS without depending on the
2489 availability of the OSTs.
2490 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2491 Add appropriate locking and lov-osc refcounts for safe
2492 cleaning. Add osc abort_inflight calls in case the
2493 synchronize never started.
2496 Frequency : occasional (Cray XT3 only)
2498 Description: root not authorized to access files in CRAY_PORTALS environment
2499 Details : The client process capabilities were not honoured on the MDS in
2500 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2501 been accessed by an authorized user then root was able to access
2502 the file on the local client also. The root user capabilities
2503 are now allowed on the MDS, as this environment has secure UID.
2506 Frequency : occasional
2508 Description: ldiskfs "too long searching" message happens too often
2509 Details : A debugging message (otherwise harmless) prints too often on
2510 the OST console. This has been reduced to only happen when
2511 there are fragmentation problems on the filesystem.
2516 Description: Division by zero in statfs when all OSCs are inactive
2517 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2518 lov_statfs divides by value returned by lov_get_stripecnt().
2522 Bugzilla : 9489, 3273
2523 Description: First write from each client to each OST was only 4kB in size,
2524 to initialize client writeback cache, which caused sub-optimal
2525 RPCs and poor layout on disk for the first writen file.
2526 Details : Clients now request an initial cache grant at (re)connect time
2527 and so that they can start streaming writes to the cache right
2528 away and always do full-sized RPCs if there is enough data.
2529 If the OST is rebooted the client also re-establishes its grant
2530 so that client cached writes will be honoured under the grant.
2535 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2536 Details : Now I/O RPCs go to different portal number and (presumably) fast
2537 lock requests (and glimses) and other RPCs get their own service
2538 threads pool that should be able to service those RPCs
2541 Severity : enhancement
2543 Description: Ability to exchange lustre version between client and servers and
2544 issue warnings at client side if client is too old. Also for
2545 liblustre clients there is ability to refuse connection of too old
2547 Details : New 'version' field is added to connect data structure that is
2548 filled with version info. That info is later checked by server and
2552 Frequency : rare, liblustre only.
2553 Bugzilla : 9296, 9581
2554 Description: Two simultaneous writes from liblustre at offset within same page
2555 might proceed at the same time overwriting eachother with stale
2557 Details : I/O lock withing llu_file_prwv was released too early, before data
2558 actually was hitting the wire. Extended lock-holding time until
2559 server acknowledges receiving data.
2562 Frequency : extremely rare. Never observed in practice.
2564 Description: avoid generating lustre_handle cookie of 0.
2565 Details : class_handle_hash() generates handle cookies by incrementing
2566 global counter, and can hit 0 occasionaly (this is unlikely, but
2567 not impossible, because initial value of cookie counter is
2568 selected randonly). Value of 0 is used as a sentinel meaning
2569 "unassigned handle" --- avoid it. Also coalesce two critical
2570 sections in this function into one.
2572 Severity : enhancement
2574 Description: allow liblustre clients to delegate truncate locking to OST
2575 Details : To avoid overhead of locking, liblustre client instructs OST to
2576 take extent lock in ost_punch() on client's behalf. New connection
2577 flag is added to handle backward compatibility.
2579 Severity : enhancement
2580 Bugzilla : 4928, 7341, 9758
2581 Description: allow number of OST service threads to be specified
2582 Details : a module parameter allows the number of OST service threads
2583 to be specified via "options ost ost_num_threads={N}" in the
2584 OSS's /etc/modules.conf or /etc/modprobe.conf.
2588 Bugzilla : 6146, 9635, 9895
2589 Description: servers crash with bad pointer in target_handle_connect()
2590 Details : In rare cases when a client is reconnecting it was possible that
2591 the connection request was the last reference for that export.
2592 We would temporarily drop the export reference and get a new
2593 one, but this may have been the last reference and the export
2594 was just destroyed. Get new reference before dropping old one.
2596 Severity : enhancement
2597 Frequency : if client is started with failover MDS
2599 Description: Allow multiple MDS hostnames in the mount command
2600 Details : Try to read the configuration from all specified MDS
2601 hostnames during a client mount in case the "primary"
2604 Severity : enhancement
2606 Description: Stop sending data to evicted clients as soon as possible.
2607 Details : Check if the client we are about to send or are sending data to
2608 was evicted already. (Check is done every second of waiting,
2609 for which l_wait_event interface was extended to allow checking
2610 of exit condition at specified intervals).
2613 Frequency : rare, normally only when NFS exporting is done from client
2615 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2617 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2620 Frequency : always, if srand() is called before liblustre initialization
2622 Description: Liblustre uses system PRNG disturbing its usage by user application
2623 Details : Introduce internal to lustre fast and high-quality PRNG for
2624 lustre usage and make liblustre and some other places in generic
2625 lustre code to use it.
2627 Severity : enhancement
2628 Bugzilla : 9477, 9557, 9870
2629 Description: Verify that the MDS configuration logs are updated when xml is
2630 Details : Check if the .xml configuration logs are newer than the config
2631 logs stored on the MDS and report an error if this is the case.
2632 Request --write-conf, or allow starting with --old_conf.
2634 Severity : enhancement
2636 Description: Handle symlinks in the path when checking if Lustre is mounted.
2637 Details : Resolve intermediate symlinks when checking if a client has
2638 mounted a filesystem to avoid duplicate client mounts.
2643 Description: lconf can hit an error exception but still return success.
2644 Details : The lconf command catches the Command error exception at the top
2645 level script context and will exit with the associated exit
2646 status, but doesn't ensure that this exit status is non-zero.
2651 Description: failure of ptlrpc thread startup can cause oops
2652 Details : Starting a ptlrpc service thread can fail if there are a large
2653 number of threads or the server memory is very fragmented.
2654 Handle this without oopsing.
2657 Frequency : always, only if liblustre and non-default acceptor port was used
2659 Description: liblustre cannot connect to servers with non-default acceptor port
2660 Details : tcpnal_set_default_params() was not called and was therefore
2661 ignoring the environment varaible TCPNAL_PORT, as well as other
2662 TCPNAL_ environment variables
2667 Description: two objects could be created on the same OST for a single file
2668 Details : If an OST is down, in some cases it was possible to create two
2669 objects on a single OST for a single file. No problems other
2670 than potential performance impact and spurious error messages.
2674 Bugzilla : 5681, 9562
2675 Description: Client may oops in ll_unhash_aliases
2676 Details : Client dcache may become inconsistent in race condition.
2677 In some cases "getcwd" can fail if the current directory is
2683 Description: Inode refcounting problems in NFS export code
2684 Details : link_raw functions used to call d_instantiate without obtaining
2685 extra inode reference first.
2689 Bugzilla : 9942, 9903
2690 Description: Referencing freed requests leading to crash, memleaks with NFS.
2691 Details : We used to require that call to ll_revalidate_it was always
2692 followed by ll_lookup_it. Also with revalidate_special() it is
2693 possible to call ll_revalidate_it() twice for the same dentry
2694 even if first occurence returned success. This fix changes semantic
2695 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2696 reference on a request referred from the intent.
2697 ll_intent_release() then releases such a request.
2700 Frequency : rare, normally benchmark loads only
2702 Description: unlinked inodes were kept in memory on the client
2703 Details : If a client is repeatedly creating and unlinking files it
2704 can accumulate a lot of stale inodes in the inode slab cache.
2705 If there is no other client load running this can cause the
2706 client node to run out of memory. Instead flush old inodes
2707 from client cache that have the same inode number as a new inode.
2710 Frequency : SLES9 2.6.5 kernel and long filenames only
2711 Bugzilla : 9969, 10379
2712 Description: utime reports stale NFS file handle
2713 Details : SLES9 uses out-of-dentry names in some cases, which confused
2714 the lustre dentry revalidation. Change it to always use the
2718 Frequency : rare, unless heavy write-truncate concurrency is continuous
2719 Bugzilla : 4180, 6984, 7171, 9963, 9331
2720 Description: OST becomes very slow and/or deadlocked during object unlink
2721 Details : filter_destroy() was holding onto the parent directory lock
2722 while truncating+unlinking objects. For very large objects this
2723 may block other threads for a long time and slow overall OST
2724 responsiveness. It may also be possible to get a lock ordering
2725 deadlock in this case, or run out of journal credits because of
2726 the combined truncate+unlink. Solution is to do object truncate
2727 first in one transaction without parent lock, and then do the
2728 final unlink in a new transaction with the parent lock. This
2729 reduces the lock hold time dramatically.
2732 Frequency : rare, 2.4 kernels only
2734 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2735 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2736 before unmounting their filesystem in order to avoid deadlock.
2737 The kernel_locked() function in 2.4 kernels only checks whether
2738 the kernel lock is held, not whether it is this process that is
2739 holding it as 2.6 kernels do.
2744 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2745 Details : The client ptlrpc code may be trying to reconnect to a down
2746 server before a previous connection attempt has timed out.
2747 Increase the reconnect interval to be longer than the connection
2748 timeout interval to avoid sending duplicate connections to
2752 Frequency : echo_client brw_test command
2754 Description: fix echo_client to work with OST preallocated code
2755 Details : OST preallocation code (5137) didn't take echo_client IO path
2756 into account: echo_client calls filter methods outside of any
2757 OST thread and, hence, there is no per-thread preallocated
2758 pages and buffers to use. Solution: hijack pga pages for IO. As
2759 a byproduct, this avoids unnecessary data copying.
2763 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2764 Description: Client can oops in mdc_commit_close() after open replay
2765 Details : It was possible for the MDS to return an open request with no
2766 transaction number in mds_finish_transno() if the client was
2767 evicted, but without actually returning an error. Clients
2768 would later try to replay that open and may trip an assertion
2769 Simplify the client close codepath, and always return an error
2770 from the MDS in case the open is not successful.
2773 Frequency : rare, 2.6 OSTs only
2775 Description: OST may deadlock under high load on fragmented files
2776 Details : If there was a heavy load and highly-fragmented OST filesystems
2777 it was possible to have all the OST threads deadlock waiting on
2778 allocation of biovecs, because the biovecs were not released
2779 until the entire RPC IO was completed. Instead, release biovecs
2780 as soon as they are complete to ensure forward IO progress.
2782 Severity : enhancement
2784 Description: Support for specifying external journal device at mount
2785 Details : If an OST or MDS device is formatted with an external journal
2786 device, this device major/minor is stored in the ext3 superblock
2787 and may not be valid for failover. Allow detecting and
2788 specifying the external journal at mount time.
2793 Description: Mounting an MDS with pending unlinked files may cause oops
2794 Details : target_finish_recovery() calls mds_postrecov() which returned
2795 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2796 considers this an error and immediately begins cleaning up the
2797 lov, just after starting the mds_lov process
2799 Severity : enhancement
2801 Description: Implement 'lfs df' to report actual free space on per-OST basis
2802 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2803 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2804 usage of objects; '-h' to report in human readable format.
2806 ------------------------------------------------------------------------------
2808 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2815 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2816 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2817 at startup during orphan unlinking. Instead, skip buddy update
2818 before setup, as it will be regenerated after recovery is complete.
2823 Description: If an OST is inactive, its locks might reference stale inodes.
2824 Details : lov_change_cbdata() must iterate over all namespaces, even if
2825 they are inactive to clear inode references from the lock.
2827 Severity : enhancement
2828 Frequency : occasional, if non-standard max_dirty_mb used
2830 Description: Client will block write RPCs if not enough grant
2831 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2832 then the client will block writes while waiting for another RPC
2833 to complete instead of consuming its dirty limit. With change
2834 we get improved performance when max_dirty_mb is small.
2836 Severity : enhancement
2837 Bugzilla : 3389, 6253
2838 Description: Add support for supplementary groups on the MDS.
2839 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2840 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2841 lookups for user supplementary groups into a cache.
2845 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2846 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2847 for liblustre O_EXCL is enforced by MDS.
2851 Description: readdir never returns NULL in liblustre.
2852 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2853 dentry in struct dirent.
2857 Frequency : liblustre only, depends on application IO pattern
2858 Description: liblustre clients evicted if not contacting servers
2859 Details : Don't put liblustre clients into the ping_evictor list, so
2860 they will not be evicted by the pinger ever.
2862 Severity : enhancement
2864 Description: Add ability to evict clients by NID from MDS.
2865 Details : By echoing "nid:$NID" string into
2866 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2867 $NID would be instantly evicted from this MDS and from all active
2868 OSTs connected to it.
2872 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2873 Details : lookup_it_finish() used to query file size from OSTs that was not
2878 Description: service threads change working directory to that of init
2879 Details : Starting lustre service threads may pin the working directory
2880 of the parent thread, making that filesystem busy. Threads
2881 now change to the working directory of init to avoid this.
2885 Frequency : during shutdown only
2886 Description: shutdown with a failed MDS or OST can cause unmount to hang
2887 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2892 Frequency : 2.6 only
2893 Description: chmod/chown may include an extra supplementary group
2894 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2895 supplementary group and if none is specified this is used.
2898 Bugzilla : 5479 (6816)
2900 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2901 Details : If lookup is in progress on a file that is unlinked we might try
2902 to revalidate the inode and fail in revalidate after lookup is
2903 complete and ll_file_open() enqueues the open again but
2904 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2907 Frequency : always, if lconf --abort_recovery used
2909 Description: lconf --abort_recovery fails with 'Operation not supported'
2910 Details : lconf was attempting to abort recovery on the MDT device and not
2913 ------------------------------------------------------------------------------
2915 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2920 Frequency : rare (only unsupported configurations with a node running as an
2922 Bugzilla : 6514, 5137
2923 Description: Mounting a Lustre file system on a node running as an OST could
2925 Details : OSTs now preallocates memory needed to write out data at
2926 startup, instead of when needed, to avoid having to
2927 allocate memory in possibly low memory situations.
2928 Specifically, if the file system is mounted on on OST,
2929 memory pressure could force it to try to write out data,
2930 which it needed to allocate memory to do. Due to the low
2931 memory, it would be unable to do so and the node would
2932 become unresponsive.
2934 Severity : enhancement
2936 Description: Addition of lconf --service command line option
2937 Details : lconf now accepts a '--service <arg>' option, which is
2938 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2940 Severity : enhancement
2942 Description: Failover mode is now the default for OSTs.
2943 Details : By default, OSTs will now run in failover mode. To return to
2944 the old behaviour, add '--failout' to the lmc line for OSTs.
2946 Severity : enhancement
2948 Description: Health checks are now provided for MDS and OSTs
2949 Details : Additional detailed health check information on MSD and OSTs
2950 is now provided through the procfs health_check value.
2953 Frequency : occasional, depends on IO load
2955 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2956 after numerous create/delete cycles
2957 Details : The ext3 inode allocation policy would not allocate new inodes
2958 very well on the OSTs because there are no new directories
2959 being created. Instead we look for groups with free space if
2960 the parent directories are nearly full.
2965 Description: Network or server problems during mount may cause partially
2966 mounted clients instead of returning an error.
2967 Details : The config llog parsing code may overwrite the error return
2968 code during mount error handling, returning success instead
2974 Description: MDS can fail to allocate large reply buffers
2975 Details : After long uptimes the MDS can fail to allocate large reply
2976 buffers (e.g. zconf client mount config records) due to memory
2977 fragmentation or consumption by the buffer cache. Preallocate
2978 some large reply buffers so that these replies can be sent even
2979 under memory pressure.
2983 Frequency : rare (liblustre)
2984 Description: fsx running with liblustre complained that using truncate() to
2985 extend the file doesn't work. This patch corrects that issue.
2986 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2987 ATTR_SIZE and lsm use in llu_setattr_raw.
2991 Frequency : rare, only 2.6 kernels
2992 Description: Unusual file access patterns on the MDS may result in inode
2993 data being lost in very rare circumstances.
2994 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2995 code in the 2.6 kernel caused the inode and/or EA data not to
2996 be read from disk, causing single-file corruption.
3000 Frequency : rare, only 2.6 filesystems using extents
3001 Description: Heavy concurrent write and delete load may cause data corruption.
3002 Details : It was possible under high-load situations to have an extent
3003 metadata block in the block device cache from a just-unlinked
3004 file overwrite a newly-allocated data block. We now unmap any
3005 metadata buffers that alias just-allocated data blocks.
3009 Frequency : filesystems with default stripe_count larger than 77
3010 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3011 Details : lconf specifies an inode size of 4096 bytes when the default
3012 stripe_count is larger than 77. This conflicts with the default
3013 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3014 this case to avoid pinning too much memory for large EAs.
3016 ------------------------------------------------------------------------------
3018 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3023 Frequency : rare (extremely heavy IO load with hundreds of clients)
3025 Description: Client is evicted, gets IO error writing to file
3026 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3027 caused clients to be evicted for AST timeouts. The fixes in
3028 bug 5192 mean we no longer need to have such short AST timeouts
3029 so ldlm_timeout has been increased.
3032 Frequency : occasional during --force or --failover shutdown under load
3033 Bugzilla : 5949, 4834
3034 Description: Server oops/LBUG if stopped with --force or --failover under load
3035 Details : a collection of import/export refcount and cleanup ordering
3036 issues fixed for safer force cleanup
3039 Frequency : only filesystems larger than 120 OSTs
3040 Bugzilla : 5990, 6223
3041 Description: lfs getstripe would oops on a very large filesystem
3042 Details : lov_getconfig used kfree on vmalloc'd memory
3045 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3046 Bugzilla : 6242, 6243
3047 Description: reading from files that had been truncated to a non-zero size
3048 but never opened returned no data
3049 Details : ll_file_read() reads zeros from no-object files to EOF
3054 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3055 Details : There's little harm in aborting MDS/OSS recovery and letting it
3056 try again, so I removed the LASSERT and return an error instead.
3058 Severity : enhancement
3060 Description: New debugging infrastructure for tracking down data corruption
3061 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3062 (b) cover more of the client-side code path, and (c) try to narrow
3063 down where problems occurred
3067 Bugzilla : 3819, 4364, 4397, 6313
3068 Description: Racing close and eviction MDS could cause assertion in mds_close
3069 Details : It was possible to get multiple mfd references during close and
3070 client eviction, leading to one thread referencing a freed mfd.
3072 Severity: : enhancement
3073 Bugzilla : 3262, 6359
3074 Description: Attempts to reconnect to servers are now more aggressive.
3075 Details : This builds on the enhanced upcall-less recovery that was added
3076 in 1.4.2. When trying to reconnect to servers, clients will
3077 now try each server in the failover group every 10 seconds. By
3078 default, clients would previously try one server every 25 seconds.
3083 Description: After recovery, certain operations trigger a failed
3084 assertion on a client.
3085 Details : Failing over an mds, using lconf -d --failover, while a
3086 client was doing a readdir() call would cause the client to
3087 LBUG after recovery completed and the readdir() was resent.
3089 Severity : enhancement
3091 Description: Default groups are now added by lconf
3092 Details : You can now run lconf --group <servicename> without having to
3093 manually add groups with lmc.
3096 Frequency : occasional
3098 Description: Nodes with an elan id of 0 trigger a failed assertion
3101 Frequency : always when accessing e.g. tty/console device nodes
3103 Description: tty and some other devices nodes cannot be used on lustre
3104 Details : file's private_data field is used by device data and lustre
3105 values in there got lost. New field was added to struct file to
3106 store fs-specific private data.
3109 Frequency : when exporting Lustre via NFS
3111 Description: NFSD failed occasionally when looking up a path component
3112 Details : NFSD is looking up ".." which was broken in ext3 directories
3113 that had grown large enough to become hashed.
3116 Frequency : Clusters with multiple interfaces not on the same subnet
3118 Description: Nodes will repeatedly try to reconnect to an interface which it
3119 cannot reach and report an error to the log.
3120 Details : Extra peer list entries will be created by lconf with some peers
3121 unreachable. lconf now validates the peer before adding it.
3124 Frequency : Only if a default stripe is set on the filesystem root.
3126 Description: Setting a default stripe on the filesystem root prevented the
3127 filesystem from being remounted.
3128 Details : The client was sending extra request flags in the root getattr
3129 request and did not allocate a reply buffer for the dir EA.
3132 Frequency : occasional, higher if lots of files are accessed by one client
3133 Bugzilla : 6159, 6097
3134 Description: Client trips assertion regarding lsm mismatch/magic
3135 Details : While revalidating inodes the VFS looks up inodes with ifind()
3136 and in rare cases can find an inode that is being freed.
3137 The ll_test_inode() code will free the lsm during ifind()
3138 when it finds an existing inode and then the VFS later attaches
3139 this free lsm to a new inode.
3143 Bugzilla : 6422, 7030
3144 Description: MDS deadlock between mkdir and client eviction
3145 Details : Creating a new file via mkdir or mknod (starting a transaction
3146 and getting the ns lock) can deadlock with client eviction
3147 (gets ns lock and trying to finish a synchronous transaction).
3150 Frequency : occasional
3151 Description: While starting a server, the fsfilt_ext3 module could not be
3153 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3154 kernels. Previously, lconf would still use the ext3 name
3155 when trying to load modules. Now, it will correctly use
3156 ext3 on 2.4 and ldiskfs on 2.6.
3158 Severity : enhancement
3159 Description: The default stripe count has been changed to 1
3160 Details : The interpretation of the default stripe count (0, to lfs
3161 or lmc) has been changed to mean striping across a single
3162 OST, rather than all available. For general usage we have
3163 found a stripe count of 1 or 2 works best.
3165 Severity : enhancement
3166 Description: Add support for compiling against Cray portals.
3167 Details : Conditional compiling for some areas that are different
3171 Frequency : occasional
3172 Bugzilla : 6409, 6834
3173 Description: Creating files with an explicit stripe count may lead to
3174 a failed assertion on the MDS
3175 Details : If some OSTs are full or unavailable, creating files may
3176 trigger a failed assertion on the MDS. Now, Lustre will
3177 try to use other servers or return an error to the
3181 Frequency : occasional
3183 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3184 caused lock timeout during readahead (since 1.4.2).
3185 Details : Processes doing readahead might match a lock that hasn't been
3186 granted yet if there are overlapping and conflicting lock
3187 requests. The readahead process waits on ungranted lock
3188 (original lock is CBPENDING), while OST waits for that process
3189 to cancel CBPENDING read lock and eventually evicts client.
3191 Severity : enhancement
3193 Description: Initial enabling of flock support for clients
3194 Details : Implements fcntl advisory locking and file status functions.
3195 This feature is provided as an optional mount flag (default
3196 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3197 locking are implemented yet, and those that are are not guaranteed
3198 to be completely correct in production environments.
3199 mount -t lustre -o [flock|noflock] ...
3202 Frequency : occasional
3204 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3205 assertion in the ext3 JBD (journaling) layer.
3206 Details : The b_committed_data struct is protected by the big kernel lock
3207 in 2.4 kernels, serializing journal_commit_transaction() and
3208 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3209 access to this struct. In 2.6 kernels there is finer grained
3210 locking to improve SMP performance of the JBD layer.
3214 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3219 Description: merge and cleanup kernel patches.
3222 -----------------------------------------------------------------------------
3224 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3226 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3227 versions of Lustre. Please update all servers and clients to
3228 version 1.4.2 or later at the same time. You must also run
3229 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3230 to update the configuration logs.
3232 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3233 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3234 - swab llog records of type '0' so we get proper header size/idx (5861)
3235 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3236 - fix rename of one directory over another leaking an inode (5953)
3237 - avoid SetPageDirty on 2.6 (5981)
3238 - don't re-add just-being-destroyed locks to the waiting list (5653)
3239 - when creating new directories, inherit the parent's custom
3240 striping settings if present parent (3048)
3241 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3242 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3243 - don't hold client locks on temporary worklist from l_lru (5666)
3244 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3245 - automatically evict dead clients (5921)
3246 - Update file size properly in create+truncate+fstat case (6196)
3247 - Do not unhash mountpoint dentries, do not allow removal of
3249 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3250 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3251 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3252 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3253 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3254 - Correctly handle reads of files with no objects (6243)
3255 - lctl recover will also mark a device active if deactivate used (5933)
3257 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3258 - allow --write-conf on an MDS with different nettype than client (5619)
3259 - don't write config llogs to MDS for mounts not from that MDS (5617)
3260 - lconf should create multiple TCP connections from a client (5201)
3261 - init scripts are now turned off by default; run chkconfig --on
3262 lustre and chkconfig --on lustrefs to use them
3263 - upcalls are no longer needed for clients to recover to failover
3265 - add --abort-recovery option to lconf to abort recovery on device
3267 - add support for an arbitrary number of OSTs (3026)
3268 - Quota support protocol changes.
3269 - forward compatibility changes to wire structs (6007)
3270 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3271 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3272 - improved "lustre status" script
3273 - initialize blocksize for non-regular files (6062)
3274 - added --disable-server and --disable-client configure options (5782)
3275 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3276 - Vanilla 2.4.29 support
3277 - increase maximum number of obd devices to 520 (6242)
3278 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3279 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3280 SLES 9 2.6.5 kernels
3281 - limit stripes per file to 160 (the maximum EA size) (6093)
3283 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3286 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3287 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3288 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3289 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3290 - small ext3 extents cleanups and fixes (5733)
3291 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3292 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3293 - avoid needless client->OST connect, fix handle mismatch (5317)
3294 - fix DLM error path that led to out-of-sync client, long delays (5779)
3295 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3296 - fix several locking issues related to i_size (5492,5624,5654,5672)
3297 - don't move pending lock onto export if it is already evicted (5683)
3298 - fix kernel oops when creating .foo in unlinked directory (5548)
3299 - fix deadlock in obdfilter statistics vs. object create (5811)
3300 - use time_{before,after} to avoid timer jiffies wrap (5882)
3301 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3302 - Do not leak request if server was not able to process it (5154)
3303 - If mds_open unable to find parent dir, make that negative lookup(5154)
3304 - don't create new directories with extent-mapping (5909, 5936)
3306 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3307 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3308 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3309 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3310 - support for RHEL4 kernel on i686 (5773)
3311 - provide error messages when incompatible logs are encountered (5898)
3313 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3314 * version 1.4.0.10 (1.4.1 release candidate 1)
3316 - don't keep a lock reference when lock is not granted (4238)
3317 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3318 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3319 - drop import inflight refcount on signal_completed_replay error (5255)
3320 - unlock page after async write error during send (3677)
3321 - handle missing objects in filter_preprw_read properly (5265)
3322 - no transno return for symlink open, don't save no-trasno open (3440)
3323 - don't try to complete elan receive that already failed (4012)
3324 - free RPC server reply state on error (5406)
3325 - clean up thread from ptlrpc_start_thread() on error (5160)
3326 - readahead could read extra page into cache that wasn't ejected (5388)
3327 - prevent races in class_attach/setup/cleanup/detach (5260)
3328 - don't dereference de->d_inode after l_dput of de (5458)
3329 - use "int" for stripe value returned from lock_to_stripe (5544)
3330 - mballoc allocation and error-checking fixes in 2.6 (5504)
3331 - block device patches to fix I/O request sizes in 2.6 (5482)
3332 - look up hostnames for IB nals (5602)
3333 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3334 - don't start multiple acceptors for the same port (5277)
3335 - fix incorrect LASSERT in mds_getattr_name (5635)
3336 - export a proc file for general "ping" checking (5628)
3337 - fix "lfs check" to not block when the MDS is down (5628)
3339 - service request history (4965)
3340 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3341 - create an "evict_client" /proc entry on OSTs, like the MDS has
3342 - fix mount usage message, return errors per mount(8) (5168)
3343 - change grep [] to grep "[]" in tests so they work in more UMLs
3344 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3345 - remove ancient LOV_MAGIC_V0 EA support (5047)
3346 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3347 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3348 - allow readahead allocations to fail when low on memory (5383)
3349 - mmap locking landed again, after considerable improvement (2828)
3350 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3352 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3355 - send OST transaction number in read/write reply to free req (4966)
3356 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3357 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3358 - the watchdog thread now runs as interruptible (5246)
3359 - flock/lockf fixes (but it's still disabled, pending 5135)
3360 - don't use EXT3 constants in llite code (5094)
3361 - memory shortage at startup could cause assertion (5176)
3363 - reorganization of lov code
3364 - single portals codebase
3366 - add extents/mballoc support (5025)
3367 - direct I/O reads in the obdfilter (4048)
3368 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3370 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3373 - send OST transaction number in read/write reply to free req (4966)
3374 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3375 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3376 - the watchdog thread now runs as interruptible (5246)
3377 - handle missing objects in filter_preprw_read properly (5265)
3378 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3379 - drop import inflight refcount on signal_completed_replay error (5255)
3380 - unlock page after async write error during send (3677)
3381 - return original error code on reconstructed replies (3761)
3382 - no transno return for symlink open, don't save no-trasno open (3440)
3384 - add pid to ldlm debugging output (4922)
3385 - bump the watchdog timeouts -- we can't handle 30sec yet
3386 - extra debugging for orphan dentry/inode bug (5259)
3388 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3391 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3392 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3393 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3394 - fix NULL dereference in /proc/sys/portals/routes (4827)
3395 - allow failed mdc_close() operations to be interrupted (4561)
3396 - stop precreate on OST before MDS would time out on it (4778)
3397 - don't send partial-page writes before EOF from client (4410)
3398 - discard client grant for sub-page writes on large-page clients (4520)
3399 - don't free dentries not owned by NFS code, check generation (4806)
3400 - fix lsm leak if mds_create_objects() fails (4801)
3401 - limit debug_daemon file size, always print CERROR messages (4789)
3402 - use transno after validating reply (3892)
3403 - process timed out requests if import state changes (3754)
3404 - update mtime on OST during writes, return in glimpse (4829)
3405 - add mkfsoptions to LDAP (4679)
3406 - use ->max_readahead method instead of zapping global ra (5039)
3407 - don't interrupt __l_wait_event() during strace
3409 - add software watchdogs to catch hung threads quickly (4941)
3410 - make lustrefs init script start after nfs is mounted
3411 - fix CWARN/ERROR duplication (4930)
3412 - return async write errors to application if possible (2248)
3413 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3414 - print NAL number in %x format (4645)
3415 - update barely-supported suse-2.4.21-171 series (4842)
3416 - support for sles 9 %post scripts
3417 - support for building 2.6 kernel-source packages
3418 - support for sles km_* packages
3420 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3423 - ignore -ENOENT errors in osc_destroy (3639)
3424 - notify osc create thread that OSC is being cleaned up (4600)
3425 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3426 - reconstruct ost_handle() like mds_handle() (4657)
3427 - create a new thread to do import eviction to avoid deadlock (3969)
3428 - let lconf resolve symlinked-to devices (4629)
3429 - don't unlink "objects" from directory with default EA (4554)
3430 - hold socknal file ref over connect in case target is down (4394)
3431 - allow more than 32000 subdirectories in a single directory (3244)
3432 - fix blocks count for O_DIRECT writes (3751)
3433 - OST returns ENOSPC from object create when no space left (4539)
3434 - don't send truncate RPC if file size isn't changing (4410)
3435 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3436 - bind to privileged port in socknal and tcpnal (3689)
3438 - rate limit CERROR/CWARN console message to avoid overload (4519)
3439 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3440 - basic mmap support (3918)
3441 - kernel patch series update from b1_4 (4711)
3443 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3446 - avoid crash during MDS cleanup with OST shut down (2775)
3447 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3448 - don't use bad inodes on the MDS (3744)
3449 - dynamic object preallocation to improve recovery speed (4236)
3450 - don't hold spinlock over lock dumping or change debug flags (4401)
3451 - don't zero obd_dev when it is force cleaned (3651)
3452 - print grants to console if they go negative (4431)
3453 - "lctl deactivate" will stop automatic recovery attempts (3406)
3454 - look for existing locks in ldlm_handle_enqueue() (3764)
3455 - don't resolve lock handle twice in recovery avoiding race (4401)
3456 - revalidate should check working dir is a directory (4134)
3458 - don't always mark "slow" obdfilter messages as errors (4418)
3460 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3463 - don't close LustreDB during write_conf until it is done (3860)
3464 - fix typo in lconf for_each_profile (3821)
3465 - allow dumping logs from multiple threads at one time (3820)
3466 - don't allow multiple threads in OSC recovery (3812)
3467 - fix debug_size parameters (3864)
3468 - fix mds_postrecov to initialize import for llog ctxt (3121)
3469 - replace config semaphore with spinlock (3306)
3470 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3471 - don't allow enqueue to complete on a destroyed export (3822)
3472 - down write_lock before checking llog header bitmap (3825)
3473 - recover from lock replay timeout (3764)
3474 - up llog sem before sending rpc (3652)
3475 - reduce ns lock hold times when setting kms (3267)
3476 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3477 - fix NULL deref and obd_dev leak on setup error (3312)
3478 - replace some LBUG about llog ops with error handling (3841)
3479 - don't match INVALID dentries from d_lookup and spin (3784)
3480 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3481 - fix invalid assertion in ptlrpc_set_wait (3880)
3483 - add libwrap support for the TCP acceptor (3996)
3484 - add /proc/sys/portals/routes for non-root route listing (3994)
3485 - allow setting MDS UUID in .xml (2580)
3486 - print the stack of a process that LBUGs (4228)
3488 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3491 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3492 - make sure to unset replay flag from failed open requests (3440)
3493 - if default stripe count is 0, use OST count for inode size (3636)
3494 - update parent mtime/ctime on client for create/unlink (2611)
3495 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3496 - free recovery state on server during a forced cleanup (3571)
3497 - unregister_reply for resent reqs (3063)
3498 - loop back devices mounting and status check on 2.6 (3563)
3499 - fix resource-creation race that can provoke i_size == 0 (3513)
3500 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3501 - more debugging for page-accounting assertion (3746)
3502 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3503 - avoid deadlock after precreation failure (3758)
3504 - fix race and lock order deadlock in orphan handling (3450, 3750)
3505 - add validity checks when grabbing inodes from l_ast_data (3599)
3507 - add /proc/.../recovery_status to obdfilter (3428)
3508 - lightweight CDEBUG infrastructure, debug daemon (3668)
3509 - change default OSC RPC parameters to be better on small clusters
3510 - turn off OST read cache for files smaller than 32MB
3511 - install man pages and include them in rpms (3100)
3512 - add new init script for (un)mounting lustre filesystems (2593)
3513 - run chkconfig in %post for init scripts (3701)
3514 - drop scimac NAL (unmaintained)
3516 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3519 - clean kiobufs before and after use (3485)
3520 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3521 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3522 - change init script to fail more clearly if not run as root (1528)
3523 - allow clients to reconnect during replay (1742)
3524 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3525 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3526 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3527 - fix cond_resched (3554)
3528 - only evict unfinished clients after recovery (3515)
3529 - allow bulk resend, prevent data loss (3570)
3530 - dynamic ptlrpc request buffer allocation (2102)
3531 - don't allow unlinking open directory if it isn't empty (2904)
3532 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3533 - remove extraneous obd dereference causing LASSERT failure (3334)
3534 - don't use get_cycles() when creating temp. files on the mds (3156)
3535 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3536 - handle EEXIST for set-stripe, set proper directory name (3336)
3538 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3539 - fix ksocknal_fmb_callback() error messages (2918)
3541 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3544 - don't copy lvb into (possibly NULL) reply on error (2983)
3545 - don't deref dentry after dput, don't free lvb on error (2922)
3546 - use the kms to determine writeback rpc length (2947)
3547 - increment oti_logcookies when osc is inactive (2948)
3548 - update client's i_blocks count via lvb messages (2543)
3549 - handle intent open/close of special files properly (1557)
3550 - mount MDS with errors=remount-ro, like obdfilter (2009)
3551 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3552 - don't use cancelling-locks' kms values (2947)
3553 - use highest lock extent for kms, not last one (2925)
3554 - don't dereference ERR_PTR() dentry in error handling path (3107)
3555 - fix thread race in portals_debug_dumplog() (3122)
3556 - create lprocfs device entries at setup instead of at attach (1519)
3557 - common AST error handler, don't evict client on completion race (3145)
3558 - zero nameidata in detach_mnt in 2.6 (3118)
3559 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3560 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3561 - initialize RPC timeout timer earlier for 2.6 (3219)
3562 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3563 - print nal/nid for unknown nid (3258)
3564 - additional checks for oscc recovery before doing precreate (3284)
3565 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3566 - don't crash in mdc_close for bad permissions on open (3285)
3567 - zero i_rdev for non-device files (3147)
3568 - clear page->private before handing to FS, better assertion (3119)
3569 - tune the read pipeline (3236)
3570 - fix incorrect decref of invalidated dentry (2350)
3571 - provide read-ahead stats and refine rpc in flight stats (3328)
3572 - don't hold journal transaction open across create RPC (3313)
3573 - update atime on MDS at close time (3265)
3574 - close LDAP connection when recovering to avoid server load (3315)
3575 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3576 - don't leak open file on MDS after open resend (3325)
3577 - serialize filter_precreate and filter_destroy_precreated (3329)
3578 - loop device shouldn't call sync_dev() for nul device (3092)
3579 - clear page cache after eviction (2766)
3580 - resynchronize MDS->OST in background (2824)
3581 - refuse to mount the same filesystem twice on same mountpoint (3394)
3582 - allow llmount to create routes for mounting behind routers (3320)
3583 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3584 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3585 - fix rare problem with rename on htree directories (3417)
3587 - allow default OST striping configuration per directory (1414)
3588 - fix compilation for qswnal for 2.6 kernels (3125)
3589 - increase maximum number of MDS request buffers for large systems
3590 - change liblustreapi to be useful for external progs like lfsck (3098)
3591 - increase local configuration timeout for slow disks (3353)
3592 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3594 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3597 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3598 - don't overwrite extent policy data in reply if lock was blocked (2901)
3599 - drop filter export grants atomically with removal from device (2663)
3600 - del obd_self_export from work_list in class_disconnect_exports (2908)
3601 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3602 - swab reply message in mdc_close, other PPC fixes (2464)
3603 - fix destroying of named logs (2325)
3604 - overwrite old logs when running lconf --write_conf (2264)
3605 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3606 - fix race in target_handle_connect (2898)
3607 - mds_reint_create() should take same inode create lock (2926)
3608 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3609 - don't close files for self_export to avoid uninitialized obd (2936)
3610 - allow MDS with the same name as client node (2939)
3611 - hold dentry reference for closed log files for unlink (2325)
3612 - reserve space for all logs during transactions (2059)
3613 - don't evict page beyond end of stripe extent (2925)
3614 - don't oops on a deleted current working directory (2399)
3615 - handle hard links to targets without a parent properly (2517)
3616 - don't dereference NULL lock when racing during eviction (2867)
3617 - don't grow lock extents when lots of conflicting locks (2919)
3619 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3622 - account for cache space usage on clients to avoid data loss (974)
3623 - lfsck support in lustre kernel code (2349)
3624 - reduce journal credits needed for BRW writes (2370)
3625 - orphan handling to avoid losing space on client/server crashes
3626 - ptlrpcd can be blocked, stopping ALL progress (2477)
3627 - use lock value blocks to assist in proper KMS, faster stat (1021)
3628 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3629 - recovery for initial connections (2355)
3630 - fixes for mds_cleanup_orphans (1934)
3631 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3632 - block all file creations until orphan recovery completes (1901)
3633 - client remove rq_connection from request struct (2423)
3634 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3635 - recovery timer race (2670)
3636 - mdc_close recovey bug (2532)
3637 - ptlrpc cleanup bug (2710)
3638 - mds timeout on local locks (2588)
3639 - namespace lock held during RPCs (2431)
3640 - handle interrupted sync write properly (2503)
3641 - don't try to handle a message that hasn't been replied to (2699)
3642 - client assert failure during cleanup after abort recovery (2701)
3643 - leak mdc device after failed mount (2712)
3644 - ptlrpc_check_set allows timedout requests to complete (2714)
3645 - wait for inflight reqs when ptlrpcd finishes (2710)
3646 - make sure unregistered services are removed from the srv_list
3647 - reset bulk XID's when resending them (caught by 1138 test)
3648 - unregister_bulk after timeout
3649 - fix lconf error (2694)
3650 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3651 - readahead locks pages, leaves pending causing memory pressure (2673)
3652 - increase OST request buffers to 4096 on large machines (2729)
3653 - fix up permission of existing directories in simple_mkdir (2661)
3654 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3655 - don't assign transno to errored transactions (2742)
3656 - don't delete objects on OST if given a bogus objid from MDS (2751)
3657 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3658 - if rq_no_resend, then timeout request after recovery (2432)
3659 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3660 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3661 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3662 - MDS recovery shouldn't delete objects that it has given out (2730)
3663 - if enqueue arrives after completion, don't clobber LVB (2819)
3664 - don't unlock pages twice when trigger_group_io returns error (2814)
3665 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3666 - don't write pages to disk if there was an error (1450)
3667 - don't ping imports that have recovery disabled (2676)
3668 - take buffered bytes into account when balancing socknal conn (2817)
3669 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3670 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3671 - remove little-endian swabbing of llog records (1987)
3672 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3673 - retry reposting request buffers if they fail (1191)
3674 - grow extent at grant time to avoid granting a revoked lock (2809)
3675 - lock revoke doesn't evict page if covered by a second lock (2765)
3676 - disable VM readahead to avoid reading outside lock extents (2805)
3678 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3679 - updated kernel patches for hp-2.4.20 kernel (2681)
3681 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3684 - fix truncated write corruption (2366)
3685 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3687 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3688 - ensure len(uuid) < 37 in lmc (1171)
3689 - fix ia64 OOPS in llog_test (2255)
3690 - zero end of page at obdfilter for partial page writes (2648)
3691 - don't leave stale dentries around after renames (bug 2428)
3692 - fix timeouts when evicting a client with a single lock held (2642)
3693 - set deadline for the initial HELLO message to drain (2634)
3694 - print out dotted-quad IP addresses in the socknal (2302)
3696 - additional debugging for MDS client eviction problem (2443)
3697 - fix mkfsoptions support for osts (2603, 2604)
3699 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3702 - add series for the vanilla 2.6.0 kernel
3703 - add series for the vanilla 2.4.24 kernel
3704 - add series for a cray x86/64 UL kernel drop
3705 - fix xattr patches for the vanilla 2.4.19 series
3707 - generate true UUIDs in lmc (1171)
3708 - have portals stack dumping break in UML (2466)
3709 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3710 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3711 - wake callback waiting threads on client eviction (2460)
3712 - Add --ptldebug and --subsystem to lmc (1719)
3713 - update assertion to allow safe interrupt allocation
3714 - set rq_no_resend for cancel requests (2432)
3715 - recalculate ptlrpcd timeout after resend (2494)
3716 - call vfs_rmdir when removing pending directories (2368)
3717 - fix renaming a file to itself (2429)
3718 - lmc creates a default one-stripe lov (2454)
3719 - expand procfs space to handle large clusters (2326)
3720 - increase UML stack to avoid overflow
3721 - update lconf's list of debug and subsystem masks
3722 - fix lfs find --obd (2510)
3723 - /proc tunable for disabling filter read caching (2591)
3724 - stop rpm packages from altering slapd.conf (2301)
3725 - disable nagle in the socknal under 0conf (2578)
3726 - choose mds inode size based on stripe count (2572)
3727 - fix kernel-source rpm problems (2516)
3729 - add --disable-doc to avoid pdf generation (2421)
3730 - update documentation, tests, type-os, comments
3731 - avoid format warnings on ia64
3732 - remove the TOE NAL
3733 - tiny code cleanups by removing unused fields
3735 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3738 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3739 - remove the most blatant lies from BUILDING file (2371)
3740 - change default debug level to reasonable production setting
3741 - reduce client side cache size to reduce cache flush time
3742 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3743 - make TCP zerocopy and pinger support enabled by default (2476)
3744 - sync writes completed after process exits caused crashes (2319)
3745 - maintain correct mount count on the MDS (2356)
3746 - backout 1557, because 2316 wasn't really fixed
3747 - better file I/O statistics gathering in /proc
3748 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3749 - another kernel patch to fix zero-copy TCP function export
3750 - don't take duplicate lock when processing re-sent getattr (2420)
3751 - lctl uses obd_self_export instead of creating new conn (2353)
3752 - MDS/OST recovery case which requires object creation asserted (2425)
3753 - move lfs from /usr/sbin to /usr/bin in packages
3754 - fix race between mds_client_add and mds_client_free (2417)
3755 - use kmalloc instead of slabs in portals (2430)
3756 - don't create duplicate records when a failover MDS is present (2442)
3757 - remove unnecessary mount age check (2332)
3758 - don't remove directory inodes from locks prematurely (2451)
3759 - don't break if MDS service name is the same as hostname (2103)
3760 - fix races in client write RPC generation when cache full (2482)
3762 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3765 - remove now-unused request->rq_obd (278)
3766 - if an allocation fails, print out how much memory we've used (1933)
3767 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3768 - add the "configurable stack size" patch to most series files (1256)
3769 - ability to write large log records, for 100+ OST configs (2306)
3770 - fix NULL deref when filter_prep fails (2314)
3771 - fix operator precedence error in filter_sync
3772 - dynamic allocation of socknal TX descriptors (2315)
3773 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3774 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3775 - ensure that CWARN messages in Portals always get to the syslog
3776 - __init/__exit are not for prototype decls (ldlm_init/exit)
3777 - x86-64 compile warning fixes
3778 - fix gateway LMC keyword conflict (2318)
3779 - fix MDS lock inversions in getattr/reint paths (1844)
3780 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3781 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3782 - DLM race condition prevented some lock evictions (2328)
3783 - ENOMEM detection and retry on socknal sends (2230)
3784 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3785 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3786 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3787 - better error messages when a client is rejected during recovery (1505)
3788 - avoid cancelling locks which were never granted, after failure (2330)
3789 - fix i_sem/journal inversion in mds_client_add (2333)
3790 - fix truncate/getattr lock cycle deadlock (2334)
3791 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3792 - fix two rare exit paths which could leak an l_lock() ref (2321)
3793 - fencepost error in MDS/OST orphan recovery (2226)
3794 - make log record alignment 8 bytes (1988)
3795 - lstripe now fails when requested offset > ost_count (2237)
3796 - ensure that all kernel series have a complete list.h (1607)
3797 - fix crashes in special-file operations (2316)
3798 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3800 - allow configurable automake binary, for testing new versions
3801 - small update to the lfs documentation
3803 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3805 * fix negative export reference count in fsfilt_sync (2312)
3807 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3808 * release candidate 0.9.1
3810 - orphans are moved into the PENDING directory for possible recovery
3811 - replayed opens now open by fid for orphan/rename safety (1042)
3812 - last close of an orphan inode generates a transno (683)
3813 - chdir() and mount() now pin the directory entry (1020)
3814 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3815 - discard very old requests without processing them (1502)
3816 - remove some common, well-understood CERRORs (1505)
3817 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3818 - clear "grant" flags in OST replies until OST grant code lands (1644)
3819 - fix read performance by not clobbering i_blksize on client (1598)
3820 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3821 - make LRU size a /proc tunable, clears locks when reduced (707)
3822 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3823 - server locks take references on exports now (1558)
3824 - build fixes for 2.4.20-rh trees (1663)
3825 - return an error from lov_create if all OSCs are inactive (1751)
3826 - fix import levels when a reconnect happens without a timeout (1597)
3827 - exit early from mds_open if we get a lookup error (1749)
3828 - partial page read at EOF wouldn't wait for disk before sending (1642)
3829 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3830 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3831 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3832 - write/truncate lock inversion (1639)
3833 - Don't auto-load obdclass, portals modules during cleanup (1495)
3834 - fix timestamps from jumping to "now" (1763)
3835 - extra journal assertions (1648)
3836 - add an extra multiunlink test (1771)
3837 - fix read_record/write_record API (1776)
3838 - fix leak of offset_extent, possible incorrect i_size later (1772)
3839 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3840 - Debugging for the kqswnal_get_idle_tx problems (1820)
3841 - Allow recovery to be attempted multiple times (1536)
3842 - Write out MDS last_rcvd file after it is first created (1600)
3843 - Fix tx_descriptor leak in failed transmit situations (1827)
3844 - ext3 journaling fixes for assertion failure after IO error (1871)
3845 - class_export_put() on freed export after completion AST error (1896)
3846 - Fix revalidate looping in VFS (1322)
3847 - Don't access a freed export during MDS_REINT timeout (1521)
3848 - Add open-unlink recovery support on the MDS (1673,1764)
3849 - Return an error if no MDS data was read from last_rcvd (1946)
3850 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3851 - Don't setup a disk device that doesn't match exported UUID (317)
3852 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3853 - avoid committing NULL handle in force close
3854 - local.sh is now a one-stripe LOV configuration
3855 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3856 - fix ext3 htree duplicate directory entry corruption (1516)
3857 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3858 - update to new LOV EA format (2097)
3859 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3861 2003-06-15 Phil Schwan <phil@clusterfs.com>
3864 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3865 - per-import/export recovery handling (958, 931, 959)
3866 - multiple last-rcvd slots, for serving multiple FSes (949)
3867 - connections are again shared between multiple imp/exports (963, 964)
3868 - "umount -f" would hang if any requests needed to be sent (393, 978)
3869 - avoid pinning large req buffer by copying for queued messages (989)
3870 - add "uuid" to "lctl device" command to help upcalls (991)
3871 - "open" RPCs with transnos would confuse recovery counters (1037)
3872 - do proper endian conversion of all wire messages (288, 340, 891)
3873 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3874 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3875 - fix ptlrpc_connection leak in target_handle_connect (1174)
3876 - fix import refcounting bug in OST and MDS cleanup (1134)
3877 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3878 - fix very unlikely obd_types race condition (501)
3879 - remove osc_open hack for echo_client (1187)
3880 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3881 - a failure in read_inode2 leads to deadlock (1139)
3882 - cancel ack-locks as soon as transaction is committed (1072)
3883 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3884 - make sure to commitrw after any preprw to avoid deadlock (1162)
3885 - failing to execute a file in a lustre FS would lock inode (1203)
3886 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3887 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3888 - obd_uuid2tgt would walk off the end of the list (1255)
3889 - on IA64 the osc would give portals incorrect bulk size (1258)
3890 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3891 - fix lock inversion caused by new llite matching code (1282)
3892 - limit the number of dirty pages on a client to 10MB (1286)
3893 - timed out locks were not being corrected cancelled (1289)
3894 - fix O_DIRECT above 4GB on IA-32 (1292)
3895 * major user-visible changes
3896 - fail out/fail over policy now controlled by the upcall (993)
3898 - add OBD_PING to check server availability and failure (954)
3899 - lustre messages are now sent in sending host order (288, 340, 891)
3900 - add eadatalen to MDS getattr reply (340)
3901 - OST read replies may contain second buffer, with per-page status (593)
3903 2003-03-11 Phil Schwan <phil@clusterfs.com>
3906 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3907 - failed open()s could cause deadlock; fixed (867, 869)
3908 - stop cancelling OST locks when files are closed (481)
3909 - overlapping XID spaces caused network corruption (851, 853)
3910 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3911 - setattr_raw added, to do single-RPC, server-side setattrs
3912 - lmc/lconf syntax change for OST UUIDs
3913 - fix crashy race condition between ptlrpc_free_req and osc_close
3914 - don't use request in mdc_enqueue if we hit a timeout (889)
3915 - don't set the inode i_size for regular files from the MDS (896)
3916 - handle out of order completion AST (842)
3917 - don't LBUG if a lock request times out after receiving AST (913)
3918 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3919 - if a bad lock AST arrives, send an error instead of dropping entirely
3920 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3921 - fix leak in bulk IO when only partially completed (899, 900, 926)
3922 - fix O_DIRECT for ia64 (55)
3923 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3924 - C-z after timeout could hang a process forever; fixed (977)
3926 - client-side I/O cache (678, 924, 929, 941, 970)
3928 - READPAGE and SETATTRs which don't take server-side locks get
3931 2003-02-11 Phil Schwan <phil@clusterfs.com>
3934 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3935 - Fix fsfilt_extN_readpage() to read a full page of directory
3936 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3937 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3938 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3939 - Most intent operations were removed, in favour of a new RPC mode
3940 that does a single RPC to the server and bypasses most of the VFS
3941 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3942 - Aggressively cancel local locks on DLM servers
3943 - mds_reint_unlink sends EA to the client if it's the last nlink.
3944 client uses that EA to unlink OST objects.
3945 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3946 - recursive symlinks were fixed (439)
3947 - fixed NULL deref in DEBUG_REQ
3948 - filter_update_lastobjid no longer calls sync, which annoyed extN
3949 - fixed multi-client small-writes to a single file problem (445)
3950 - fixed mtime updates during file writes (607)
3951 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3952 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3953 - fixed Linux OST size reporting problem (444, 656)
3954 - OST now updates object mtime with writes or setattr (607, 619)
3955 - client verifies file size before zeroing page past EOF (445)
3956 - OST now writes last allocated objid to disk with allocation (108)
3957 - LOV on echo now works (409)
3959 - mds_reint_unlink sends a new buffer, with the EA included. this
3960 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3961 set if a regular file was being unlinked, and it was the last link
3962 - use PtlGet from the target for bulk writes (315)
3963 - OST now updates object mtime with writes or setattr (607, 619)
3964 - LDLM now has a grant-time callback to revalidate locked items, if
3966 - Many MDS operations were reorganized to combat race conditions
3968 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3969 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3971 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3974 - Fully reactivate OST imports after reconnection (512, others)
3975 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3976 - More graceful error handling for truncating on dead OST (515)
3977 - Don't error out unless we're actually accessing dead stripes (474)
3978 - Fix garbage sizes when stripes are missing (410)
3979 - LRU counters were broken, causing constant lock purge (433, 432)
3980 - garbage on read from stripes with failed OSTs (441)
3981 - mark OSCs as active before reconnecting during recovery (438)
3982 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3983 - lfind did not preserve OST order in output (443)
3984 - symlinks cause hung clients, incorrect data (439)
3985 - stop dereferencing request after dropping refcount (457)
3986 - don't LASSERT(spin_is_locked) on non-SMP (455)
3987 - fixes for many rename() bugs
3988 - fstat didn't correctly synchronize attributes (399)
3989 - server must handle lock cancellation during blocking AST prep (487)
3990 - bulk descriptors were free()d too soon (511)
3991 - fix paths in lconf, which would load incorrect modules (451, 507)
3992 - fix confusing lconf 'host not found' error message (386)
3993 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3994 - fix race condition in mdc_blocking_ast() for inode access (526)
3995 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3996 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3997 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3998 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3999 - fix lstripe in several ways (kernel side) (527)
4000 - fix request leak in ldlm_cli_enqueue (262)
4001 - incorrect OSC was marked inactive after OST failure
4002 - call mds_fs_cleanup before unmounting filesystem (524)
4003 - fix races between taking ns_lock and ldlm_lock_change_resource
4004 - fix races updating LOV export open file list
4005 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4006 - fix recovery NULL deref in ldlm_cli_cancel_unused
4007 - fix some DLM races by using new hash table for lock handles (419)
4008 - permit the client to specify desired inodes, at replay
4009 - duplicate requests when we queue them for replay reintegration
4010 - fix last_rcvd offset calculation
4011 - sync after each recovered transaction, so we always make progress
4012 - never, not always, ERESTART requests without transnos
4013 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4015 - skip replay if the MDS says that the client is already connected
4016 - don't check for a recovery-enabled export to match lctl's UUID
4017 - don't INC_USE_COUNT for phantom exports
4018 - don't crash when cleaning up phantom exports (567)
4019 - don't double-finish or set replay data for errored mdc_open requests
4020 - abort requests when they time out, so we don't get old replies
4021 - send/receive replies for AST messages again
4022 - if the client says that it doesn't have the lock, cancel it on the
4024 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4025 mark it as destroyed, it will all work out when decref is called
4026 - fix module use counts (22, 581)
4028 - ASTs now expect a reply (server cancels lock on error reply)
4030 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4033 - fix many simultaneous client startup (392)
4034 - fix dentry->d_it clobbering
4035 - credentials weren't being shipped for readdir/getattr operations
4036 - remove invalid assertions triggered during some concurrent MD
4038 - proper Lustre versions added (336, 389)
4039 - fix memory leak for create error case (398)
4040 - fix LOV locking bug that would get cli/srv out of sync
4041 - fix echo client over LOV (409)
4042 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4043 - fix double-O_EXCL intent crash (424)
4044 - avoid sending multiple lock CANCELs (352)
4046 - MDS can do multi-client recovery (modulo bugs in new code)
4048 - many updates, edits, cleanups
4050 2002-11-18 Phil Schwan <phil@clusterfs.com>
4053 - fix null d_it dereference (346)
4054 - fix full OST/dbench hang (333)
4055 - fix permission problem with file removal (286)
4056 - fix removal of OSCs from LOV when they fail
4057 - fix NULL deref during bulk timeout (214)
4058 - fix problems related to multiple filesystems on one MDS (241)
4059 - fixed serious subtle metadata locking bugs
4060 - free locks on clients when inodes are removed due to memory
4062 - fix inode pointer in lock data (285)
4063 - partial support for multiple MDS on a single host (241)
4064 - data locks weren't cancelled at clear_inode time (290, 311)
4065 - intent locks could lead to unbounded lock growth (205)
4066 - added a maximum lock count, an LRU list, and a flusher
4067 - fix multiple rename (365)
4068 - properly abstracted the echo client
4069 - OSC locked 1 byte too many; fixed
4070 - rewrote brw callback code:
4071 - fixed recovery bugs related to LOVs (306)
4072 - fixed too-many-pages-in-one-write crash (191)
4073 - fixed (again) crash in sync_io_timeout (214)
4074 - probably fixed callback-related race (385)
4076 - Add capability to MDS protocol
4077 - LDLM cancellations and callbacks on different portals
4079 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4082 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4083 - defer open object destruction to close time (601981, 138)
4084 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4085 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4086 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4087 - don't call obd_setattr in ll_file_release for destroyed objects
4088 * protocol change to lustre_msg: move |version| and add |flags|
4089 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4090 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4091 * added replay of create, unlink, link and rename operations during
4092 MDS failover; recovery should be much more robust now
4093 * remove failed OSCs from LOVs (only lov_create uses this so far)
4094 * the lustre-HOWTO was brought (more) up to date (582544)
4096 2002-10-23 Phil Schwan <phil@clusterfs.com>
4099 - in-use dentries weren't being reused properly (617851)
4100 - prevent multiple LDLM setup (599178)
4101 - fix LOV size calculations for truncate (617853)
4102 - fix client handling of MDS intent errors (POSIX)
4103 - fix permission bug in lovstripe.c test (624321)
4104 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4105 - truncate past end of file could corrupt data
4106 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4107 - a race in recovery could return ETIMEDOUT to apps (623947)
4108 - building outside the source directory was fixed
4109 * the lustre-HOWTO was brought (more) up to date (582544)
4110 * major progress was made on recovery functionality
4112 2002-10-10 Phil Schwan <phil@clusterfs.com>
4115 - recovery deadlock fix
4116 - rm -rf causes LBUG fix (617817)
4117 - file open by multiple tasks fix (618962)
4118 - directory permissions bugs (602707 and 620007)
4119 - journal_stop fixed with locking (611313)
4120 - O_APPEND failures resolved (618273, perhaps 614459)
4121 - lconf PATH fix (619770)
4122 - IA64 build fix (621450)
4123 - RPC buffer sizes scale with amount of memory
4125 2002-10-01 Phil Schwan <phil@clusterfs.com>
4128 - locks would be cancelled without throwing away data pages,
4129 resulting in inconsistent data (605627)
4130 - inode attributes were not always being refreshed (605627, 612449)
4131 - lconf now continues to cleanup after lctl reports an error
4132 - MDS now enforces user permissions (602707)
4133 - lprocfs cleanup fixed, but not yet enabled (614157)
4134 - fixed infinite server hang, should a client not respond to an AST
4135 - avoid going into recovery if user calls readlink() with a buffer
4136 that's too small (613941)
4137 - AST RPCs no longer require replies (614867) -- this may be changed
4138 - don't crash server if client sends an IOV that's too big (611336)
4139 - fixed lock conversion deadlock (611892)
4140 - fixed the following of symlinks (614622)
4141 * recovery: the server can remove locks from a client that dies, other
4142 clients can make progress
4143 * more extN patch fixes
4144 * compile-time configurable ptlrpc buffer allocations
4146 - collaborative read cache document
4147 - Lustre Lite Performance CDR document-in-progress
4149 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4152 - fix typo in patch-2.4.18
4154 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4157 - clear ptlrpc request each time in handle_incoming_request()
4158 - unlink of files now destroys the object on the OST
4160 2002-09-19 Peter Braam <braam@clusterfs.com>
4162 * add hard link support
4163 * change obdfile creation method
4164 * kernel patch changed
4166 2002-09-19 Peter Braam <braam@clusterfs.com>
4169 - stack overflow bug in extN fixed
4171 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4173 * documentation updates
4174 - add man pages for config tools
4175 - update tests/README to describe testing with new config tools
4176 - finish metadata API descriptions
4177 * bug fixes and cleanups
4178 - statfs workaround for 16TB limit
4179 - LOV stripe allocation improved, can stripe on subset of OSTs
4180 - LOV file size/IO offset was wrong for files > 4GB in size
4181 - object EA data was being dropped, caused files to be unreadable
4182 - memory overflow with non-LOV OST caused memory corruption
4183 - fixed regression tests to work with new config tools, obdfilter
4184 - fixed bug when directory size became larger than 1 block
4185 - fixed bug (for single client case) when PWD was deleted
4186 - invalidate local directory pages when doing intent-based ops
4187 - avoid LDLM oops when lock callback contained bad data
4189 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4191 * documentation updates
4192 * bug fixes and cleanups
4193 - configuration tools
4196 - 64-bit compile warnings
4197 - 64-bit internal statfs data
4199 * test_brw on persistent OST devices
4201 * lprocfs (disabled)
4203 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4205 * documentation updates
4206 * bug fixes and cleanups
4207 * configuration tools
4209 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4212 * many small fixes to 0_5_4
4213 * io/network handling
4214 * thinkos in MDS operations
4216 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4219 * crucial basic fixes to 0.5.3
4220 * IOR, Iozone work over Elan
4223 2002-08-07 Phil Schwan <phil@clusterfs.com>
4224 * version 0_5_3, our first alpha
4225 * we use the new Portals iovs
4226 * documentation updates
4227 * bug fixes and cleanups
4228 * small changes in the DLM wire protocol
4230 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4231 * version 0_5_1 with some initial stability,
4232 * locking on MD and file I/O.
4233 * documentation updates
4234 * several bug fixes since 0.5.0
4235 * small changes in wire protocol
4237 2002-07-18 Phil Schwan <phil@clusterfs.com>
4239 * delivered as Lustre Light Alpha
4240 * fixed a crash after handling invalid MDS requests
4241 * fixed directory pages for architectures with non-4k pages sizes
4243 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4244 * release version v0_4_4
4245 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4247 * Fixes a number of interruption problems with OST operations.
4248 * Update documentation for portals header changes
4249 * Move all wire protocol structs/defines to lustre_idl.h
4250 * Fixes symlink length bug.
4251 * Add tcpdump to repository.
4253 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4254 * release version v0_4_3
4255 * Fixes statfs for inodes on extN.
4256 * Fixes bug in runtests which would delete /etc/hosts.
4257 * Use 64-bit object IDs wherever possible (not into VFS though)
4258 Remove ost_get_info, which is unused by lustre, and out of date.
4260 2002-07-03 Peter Braam <braam@clusterfs.com>
4261 * release version v0_4_2 Fixes a lookup error (type not passed)
4262 * move forward to head of Portals
4263 * move forward to latest Lustre kernel
4265 2002-06-25 Peter Braam <braam@clusterfs.com>
4266 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob