3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.53-0.16 (SLES 10),
6 2.6.18-8.1.15.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs4
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Frequency : only at startup
18 Description: not alloc memory with spinlock held.
19 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
22 Severity : enhancement
24 Description: make lustre randomly fail allocating memory
25 Details : Make lustre randomly failed allocating memory for testing purpose.
27 Severity : enhancement
29 Description: lost problems with lov objid file
30 Details : Fixes some scability and access to not inited memory problems
31 in work with lov objdid file.
36 Description: lfs find does not continue on file error
37 Details : Continue other files processing when a file/dir is absent.
41 Description: Inconsistent usage of lustre_pack_reply()
42 Details : Standardize the usage of lustre_pack_reply() such that it
43 always generate a CERROR on failure.
46 Frequency : occasional
48 Description: MDS hang or stay in waiting lock
49 Details : If client receive lock with CBPENDING flag ldlm need send lock
50 cancel as separate rpc, to avoid situation when cancel request
51 can't processed due all i/o threads stay in wait lock.
55 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
59 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
63 Description: Update to SLES9 kernel-2.6.5-7.287.3.
67 Description: Update to RHEL4 latest kernel.
69 Severity : enhancement
71 Description: Build SLES10 patchless client fails
72 Details : The configure was broken by run ./configure with
73 --with-linux-obj=.... argument for patchless client. When the
74 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
75 can't search header adequately. Use absolute path such as
76 -I($LINUX)/include instead.
80 Description: interrupt oig_wait produce painc on resend.
81 Details : brw_redo_request can be used for resend requests from ptlrpcd and
82 private set, and this produce situation when rq_ptlrpcd_data not
83 copyed to new allocated request and triggered LBUG on assert
84 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
85 ptlrpcd set if request is changed and can be safety changed to use
90 Description: LASSERT_{REQ,REP}SWAB macros are buggy
91 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
92 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
93 these macros and replace them with their difinitions instead.
97 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
98 Details : Update which_patch & target file for SLES10 latest kernel.
101 Description: Remove client patches from SLES 10 kernel.
102 Details : This causes SLES 10 clients to behave as patchless clients
103 even on a Lustre-patched (server) kernel.
105 Severity : enhancement
107 Description: self-adjustable client's lru lists
108 Details : use adaptive algorithm for managing client cached locks lru
109 lists according to current server load, other client's work
110 pattern, memory activities, etc. Both, server and client
111 side namespaces provide number of proc tunables for controlling
114 Severity : enhancement
116 Description: light-weight GSS support
117 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
118 and reduce performance overhead.
120 Severity : enhancement
122 Description: Linux keyring support
123 Details : Support using service of Linux keyring for Lustre GSS internal
124 context refresh/cache mechanism.
128 Description: Fix errors in lfs documentation
129 Details : Fixes man pages
133 Description: don't use GFP_* in generic Lustre code.
134 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
138 Description: obdclass is limited by single OBD_ALLOC(idarray)
139 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
143 Description: warnings with build patchless client with vanila 2.6.19 and up
144 Details : change old ctl_table style and replace ctl_table/ctl_table_header
145 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
149 Frequency : Only for SLES
150 Description: sanity_quota fail test_1
151 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
152 file, which makes TSTID[2] inunique.
156 Frequency : Only for SLES9
157 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
158 Details : Update target/ChangeLog/which_patch .
162 Frequency : Only for test_10 in sanity-quota.sh
163 Description: a bug in quota test code
164 Details : checking a test flag in a wrong place
168 Frequency : Only for test_13 in sanity-quota.sh
169 Description: a bug in quota test script
170 Details : shouldn't check overfull allocated quota
174 Description: jbd statistics
175 Details : Port older jbd statistics patch for sles10
177 Severity : enhancement
179 Description: Add printing inode info into message about error in writepage.
184 Description: reply_lock_interpret crash due to race with it and lock cancel.
185 Details : Do not replay locks that are being cancelled. Do not reference
186 locks by their address during replay, just by their handle.
190 Frequency : When flocks are used.
191 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
192 Details : Flock locks might destroy just granted lock if it could be merged
193 with another existing flock, this is done in completion handler,
194 so teach ldlm_cli_enquque_fini that this is a valid case for
200 Description: Oops in read and write path when failing to allocate lock.
201 Details : Check if lock allocation failed and return error back.
205 Description: lstripe command fails for valid OST index
206 Details : The stripe offset is compared to lov->desc.ld_tgt_count
207 instead of lov->desc.ld_active_tgt_count.
211 Description: sanity.sh failed test 103
212 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
217 Description: df doesn't work properly if diskfs blocksize != 4K
218 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
222 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
223 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
227 Description: obd_health_check_timeout too short
228 Details : set obd_health_check_timeout as 1.5x of obd_timeout
232 Description: llapi_file_create() does not allow some changes
233 Details : add llapi_file_open() that allows specifying the mode and
234 open flags, and also returns an open file handle.
238 Description: merge and cleanup kernel patches.
239 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
243 Description: Add journal checksum support.(Kernel part)
244 Details : The journal checksum feature adds two new flags i.e
245 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
246 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
247 indicates that the commit block contains the checksum for
248 the blocks described by the descriptor blocks. Now commit
249 record can be sent to disk without waiting for descriptor
250 blocks to be written to disk. This behavior is controlled
251 using JBD2_FEATURE_ASYNC_COMMIT flag.
255 Description: OSS needs mutliple precreate threads
256 Details : Add ability to start more than one create thread per OSS.
260 Description: Sanity.sh test_65k failed: setstripe should have succeeded
261 Details : execute lfs setstripe on client
265 Description: mds_obd_create error creating tmp object
266 Details : When the user sets quota on root, llog will be affected and can't
267 create files and write files.
270 Frequency : Always on ia64 patchless client, and possibly others.
272 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
273 Details : This allows the patchless client to be loaded on architectures
278 Description: RedHat Update kernel for RHEL5
279 Details : Add the kernel config file for RHEL5.
283 Description: RedHat Update kernel for RHEL5
284 Details : Modify the kernel config file more closer RHEL5.
288 Description: Build failure against Centos5 (RHEL5)
289 Details : Define PAGE_SIZE when it isn't present.
293 Description: test_7 on sanity-quota.sh can't be used on separate machines
294 Details : add facet to handle it
298 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
299 Details : with NFS, the anon dentry's parent was set to itself in
300 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
301 corresponding dentry other than use rec->ur_name.
303 Severity : enhancement
305 Description: lfs setstripe enhancement
306 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
310 Description: enable data checksumming by default
311 Details : enable checksum by default, allow --disable-checksum
312 configure option and "-o nochecksum" mount option. Checksums
313 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
317 Description: lustre support for RHEL5
318 Details : Add support for RHEL5.
322 Description: Client eviction due to failover config
323 Details : after a connection loss, the lustre client should attempt to
324 reconnect to the last active server first before trying the
325 other potential connections.
329 Description: when mds and osts use different quota unit(32bit and 64bit),
330 quota will be released repeatly.
331 Details : void sending multiple quota reqs to mds, which will keep the
332 status between the reqs.
336 Description: osts not allocated evenly to files
337 Details : change the condition to increase offset_idx
340 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
342 Description: Data corruption for OSTs that are formatted larger than 2TB
344 Details : When generating the bio request for lustre file writes the
345 sector number would overflow a temporary variable before being
346 used for the IO. The data reads correctly from Lustre (which
347 will overflow in a similar manner) but other file data or
348 filesystem metadata may be corrupted in some cases.
352 Description: Tune the kernel for good SCSI performance.
353 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
354 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
359 Description: rewrite ext2-derived code in obdclass/uuid.c
360 Details : rewrite inherited code (uuid parsing code from ext2 utils)
361 from scratch preserving functionality.
365 Description: "lfs find -obd UUID" prints directories
366 Details : "lfs find -obd UUID" will return all directory names instead
367 of just file names. It is incorrect because the directories
368 do not reside on the OSTs.
372 Description: use i_size_read and i_size_write in 2.6 port
373 Details : replace inode->i_size access with i_size_read/write()
376 Frequency : when removing large files
378 Description: scheduling issue during removal of large Lustre files
379 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
380 It causes scheduling issues when removing large files (17TB in the
384 Frequency : only with liblustre clients on XT3
386 Description: evictions taking too long
387 Details : allow llrd to evict clients directly on OSTs
390 Frequency : only on ppc
392 Description: /proc/fs/lustre/devices broken on ppc
393 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
394 We should make sure the type of 'index' is loff_t and then cast
395 explicitly as needed below. Do not assign an explicitly cast
399 Frequency : only for rhel5
401 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
402 Details : Modify the target file & which_kernel.
404 Severity : enhancement
406 Description: omit set fsid for export NFS
407 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
408 as FS_REQUIRES_DEV for avoid problems with generate fsid.
412 Frequency : Always, for kernels after 2.6.16
413 Description: Fix warning idr_remove called for id=.. which is not allocated.
414 Details : Last kernels save old s_dev before kill super and not allow
415 to restore from callback - restore it before call kill_anon_super.
419 Description: buffer overruns could theoretically occur
420 Details : llapi_semantic_traverse() modifies the "path" argument by
421 appending values to the end of the origin string, and a
422 overrun may occur. Adding buffer overrun check in liblustreapi.
426 Description: Fix error on 'ls .' at the top of the Lustre mount.
427 Details : Don't revalidate dentry if it is a root dentry.
431 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
432 Details : Modify vm-tunables-rhel4.patch.
436 Description: Kernel config for 2.6.18-vanilla.
437 Details : Modify targets/2.6-vanilla.target.in.
438 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
439 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
440 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
441 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
445 Description: improve handling recoverable errors
446 Details : if request processig with error which can be recoverable on server
447 request should be resend, otherwise page released from cache and
452 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
453 Details : Modify target file & which_patch.
454 A flaw was found in the IA32 system call emulation provided
455 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
456 value could be stored in the %RAX register, which could trigger an
457 out-of-bounds system call table access. An untrusted local user
458 could exploit this flaw to run code in the kernel
459 (ie a root privilege escalation). (CVE-2007-4573).
463 Description: O_DIRECT bypasses client statistics.
464 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
465 read_ahead_stats were updated. copied the stats section from
466 osc_send_oap_rpc() into async_internal().
470 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
474 Description: change order of libsysio includes
475 Details : '#include sysio.h' should always come before '#include xtio.h'
479 Description: handle "serious error: objid * already exists" more gracefully
480 Details : If LAST_ID value on disk is smaller than the objects existing in
481 the O/0/d* directories, it indicates disk corruption and causes an
482 LBUG(). If the object is 0-length, then we should use the existing
483 object. This will help to avoid a full fsck in most cases.
485 Severity : enhancement
487 Description: adapt the lustre_config script to support the upgrade case
488 Details : Add "-u" option for lustre_config script to support upgrading 1.4
489 server targets to 1.6 in parallel.
493 Description: To avoid grant space > avaible space when the disk is almost
494 full. Without this patch you might see the error "grant XXXX >
495 available" or some LBUG about grant, when the disk is almost
497 Details : In filter_check_grant, for non_grant cache write, we should
498 check the left space by if (*left > ungranted + bytes), instead
499 of (*left > ungranted), because only we are sure the left space
500 is enough for another "bytes", then the ungrant space should be
501 increase. In client, we should update cl_avail_grant only there
502 is OBD_MD_FLGRANT in the reply.
506 Description: Update RHEL 4 kernel to fix local root privilege escalation.
507 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
508 described in CVE-2007-4573. This problem could allow untrusted
509 local users to gain root access.
512 Frequency : when using O_DIRECT and quotas
514 Description: Incorrect file ownership on O_DIRECT output files
515 Details : block usage reported by 'lfs quota' does not take into account
516 files that have been written with O_DIRECT.
521 Description: touch file failed when fs is not full
522 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
523 otherwise we can get ENOSP while fs is not full.
527 Description: parallel lock callbacks
528 Details : Instead of sending blocking and completion callbacks as separated
529 requests, adding them to a set and sending in parallel.
532 Frequency : only for Cray XT3
533 Bugzilla : 12829/13455
534 Description: Changing primary group doesn't change the group lustre assigns to
536 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
537 overridden with the primary group provided by the group upcall,
538 whereas the supplied fsgid can be trusted if it is in the list of
539 supplementary groups returned by the group upcall.
541 --------------------------------------------------------------------------------
543 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
545 * Support for kernels:
546 2.6.5-7.283 (SLES 9),
547 2.6.9-55.EL (RHEL 4),
548 2.6.16.46-0.14 (SLES 10),
549 2.6.18.8 vanilla (kernel.org)
550 * Client support for unpatched kernels:
551 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
552 2.6.16 - 2.6.22 vanilla (kernel.org)
553 * Due to recently discovered recovery problems, we do not recommend
554 using patchless RHEL 4 clients with this or any earlier release.
555 * Recommended e2fsprogs version: 1.39.cfs8
556 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
557 * Starting with this release, the ldiskfs backing filesystem required
558 by Lustre is now in its own package, lustre-ldiskfs. This package
559 should be installed. It is versioned separately from Lustre and
560 may be released separately in future.
564 Description: block reactivating mgc import until all deactivates complete
565 Details : Fix race when failing back MDT/MGS to itself (testing)
568 Frequency : at statup only
570 Description: mds_lov_synchronize race leads to various problems
571 Details : simultaneous MDT->OST connections at startup can cause the
572 sync to abort, leaving the OSC in a bad state.
574 Severity : enhancement
576 Description: add optional extra BUILD_VERSION info
577 Details : add a new environment variable (namely LUSTRE_VERS) which allows
578 to override the lustre version.
581 Frequency : 2.6.18 servers only
583 Description: ll_kern_mount() doesn't release the module reference
584 Details : The ldiskfs module reference count never drops down to 0
585 because ll_kern_mount() doesn't release the module reference.
590 Description: server LBUG when using old ost_num_threads parameter
591 Details : Accept the old ost_num_threads parameter but warn that it
592 is deprecated, and fix an off-by-one error that caused an LBUG.
597 Description: Transient SCSI error results in persistent IO issue
598 Details : iobuf->dr_error is not reinitialized to 0 between two
602 Frequency : sometimes when underlying device returns I/O errors
604 Description: OSTs not going read-only during write failures
605 Details : OSTs are not remounted read-only when the journal commit threads
606 get I/O errors because fsfilt_ext3 calls journal_start/stop()
607 instead of the ext3 wrappers.
611 Description: poor connect scaling with increasing client count
612 Details : Don't run filter_grant_sanity_check for more than 100 exports
613 to improve scaling for large numbers of clients.
616 Frequency : SLES10 only
618 Description: sanity-quota.sh quotacheck failed: rc = -22
619 Details : Quotas cannot be enabled on SLES10.
622 Frequency : liblustre clients only
624 Description: getdirentries does not give error when run on compute nodes
625 Details : getdirentries does not fail when the size specified as an argument
626 is too small to contain at least one entry
628 Severity : enhancement
630 Description: Add LNET router traceability for debug purposes
631 Details : If a checksum failure occurs with a router as part of the
632 IO path, the NID of the last router that forwarded the bulk data
633 is printed so it can be identified.
638 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
639 Details : Due to a race condition, liblustre clients were occasionally
642 Severity : enhancement
644 Description: lfs setstripe use optional parameters instead of postional
647 Severity : enhancement
649 Description: Nanosecond timestamp support for ldiskfs
650 Details : The on-disk ldiskfs filesystem has added support for nanosecond
651 resolution timestamps. There is not yet support for this at
652 the Lustre filesystem level.
655 Frequency : during server recovery
657 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
658 Details : request with rq_no_resend flag not awake l_wait_event if they get a
662 Frequency : nfs export on patchless client
664 Description: connectathon hang when test nfs export over patchless client
665 Details : Disconnected dentry cannot be found with lookup, so we do not need
666 to unhash it or make it invalid
669 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
670 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
671 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
676 Description: open req refcounting wrong on reconnect
677 Details : If reconnect happened between getting open reply from server and
678 call to mdc_set_replay_data in ll_file_open, we will schedule
679 replay for unreferenced request that we are about to free.
680 Subsequent close will crash in variety of ways.
681 Check that request is still eligible for replay in
682 mdc_set_replay_data().
687 Description: disable writes to filesystem when reading health_check file
688 Details : the default for reading the health_check proc file has changed
689 to NOT do a journal transaction and write to disk, because this
690 can cause reads of the /proc file to hang and block HA state
691 checking on a healthy but otherwise heavily loaded system. It
692 is possible to return to the previous behaviour during configure
693 with --enable-health-write.
695 Severity : enhancement
697 Description: 64-bit inode version
698 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
699 to the inode. This will be required for version-based recovery.
704 Description: MDS fails to start if a duplicate client export is detected
705 Details : in some rare cases it was possible for a client to connect to
706 an MDS multiple times. Upon recovery the MDS would detect this
707 and fail during startup. Handle this more gracefully.
709 Severity : enhancement
711 Description: Add -o localflock option to simulate old noflock
713 Details : This will achieve local-only flock/fcntl locks
719 Description: log_commit_thread vs filter_destroy race leads to crash
720 Details : Take import reference before releasing llog record semaphore
725 Description: Wrong request locking in request set processing
726 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
727 imp_delayed_list, in this place should be used imp_lock.
730 Frequency : when reconnection
732 Description: Grant Leak when osc reconnect to OST
733 Details : When osc reconnect ost, OST(filter) should check whether it
734 should grant more space to client by comparing fed_grant and
735 cl_avail_grant, and return the granted space to client instead
736 of "new granted" space, because client will call osc_init_grant
737 to update the client grant space info.
740 Frequency : when client reconnect to OST
742 Description: Grant Leak when osc do resend and replay bulk write
743 Details : When osc reconnect to OST, OST(filter)should clear grant info of
744 bulk write request, because the grant info while be sync between
745 OSC and OST when reconnect, and we should ignore the grant info
746 these of resend/replay write req.
751 Description: Grant space more than avaiable left space sometimes.
752 Details : When then OST is about to be full, if two bulk writing from
753 different clients came to OST. Accord the avaliable space of the
754 OST, the first req should be permitted, and the second one
755 should be denied by ENOSPC. But if the seconde arrived before
756 the first one is commited. The OST might wrongly permit second
757 writing, which will cause grant space > avaiable space.
760 Frequency : when client is evicted
762 Description: Grant might be wrongly erased when osc is evicted by OST
763 Details : when the import is evicted by server, it will fork another
764 thread ptlrpc_invalidate_import_thread to invalidate the
765 import, where the grant will be set to 0. While the original
766 thread will update the grant it got when connecting. So if
767 the former happened latter, the grant will be wrongly errased
768 because of this race.
773 Description: Checking Stale with correct fid
774 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
775 to check whether it is stale, because sometimes, we want the
776 enqueue happened anyway, and op_data.fid2 will not be initialized.
778 Severity : enhancement
780 Description: update patchless client
781 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
784 Frequency : only with 2.4 kernel
786 Description: random memory corruption
787 Details : size of struct ll_inode_info is to big for union inode.u and this
788 can be cause of random memory corruption.
793 Description: Memory leak in recovery
794 Details : Lov_mds_md was not free in an error handler in mds_create_object.
795 It should also check obd_fail before fsfilt_start, otherwise if
796 fsfilt_start return -EROFS,(failover mds during mds recovery).
797 then the req will return with repmsg->transno = 0 and rc = EROFS.
798 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
799 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
800 be freed no matter whether fsfilt_commit success or not.
803 Frequency : only with huge count clients
805 Description: Prevents from taking the superblock lock in llap_from_page for
807 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
808 for llap_from_page call in ll_removepage prevents from taking the
809 superblock lock for a soon died page.
814 Description: Not check open intent error before release open handle
815 Details : in some rare cases, the open intent error is not checked before
816 release open handle, which may cause
817 ASSERTION(open_req->rq_transno != 0), because it tries to release
818 the failed open handle.
823 Description: Set cat log bitmap only after create log success.
824 Details : in some rare cases, the cat log bitmap is set too early. and it
825 should be set only after create log success.
829 Description: Accessing a block bevice can re-enable I/O when Lustre is
830 tearing down a device.
831 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
836 Description: service threads may hog cpus when there are a lot of requests
837 Details : Insert cond_resched to give other threads a chance to use some CPU
842 Description: the cat log was not initialized in recovery
843 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
844 unlink log on mds will not be initialized until mds post
845 recovery. And also in mds post recovery, the unlink log will
846 initialization will be done asynchronausly, so there will be race
847 between add unlink log and unlink log initialization.
851 Description: brw_stats were being printed incorrectly
852 Details : brw_stats were being printed as log2 but all of them were not
853 recorded as log2. Also remove some code duplication arising from
854 filter_tally_{read,write}.
858 Frequency : rare, only in recovery.
859 Description: ASSERTION(req->rq_type != LI_POISON) failed
860 Details : imp_lock should be held while iterating over imp_sending_list for
861 prevent destroy request after get timeout in ptlrpc_queue_wait.
865 Description: replay-single.sh test 52 fails
866 Details : A lock's skiplist need to be cleanup when it being unlinked
867 from its resource list.
871 Description: Short directio read returns full requested size rather than
873 Details : Direct I/O operations should return actual amount of bytes
874 transferred rather than requested size.
876 Severity : enhancement
878 Description: metadata RPC reduction (e.g. for rm performance)
879 Details : decrease the amount of synchronous RPC between clients and servers
880 by canceling conflicing lock before the operation on the client
881 and packing thier handles into the main operation RPC to server.
883 Severity : enhancement
885 Description: Async OSC create to avoid the blocking unnecessarily.
886 Details : If a OST has no remain object, system will block on the creating
887 when need to create a new object on this OST. Now, ways use
888 pre-created objects when available, instead of blocking on an
889 empty osc while others are not empty. If we must block, we block
890 for the shortest possible period of time.
892 Severity : enhancement
894 Description: refine locking for avoid write wrong info into lov_objid file
895 Details : fix possible races with add new target and write/update data in
898 --------------------------------------------------------------------------------
900 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
905 Frequency : on some architectures
907 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
908 Details : Uninitialized flags sometimes cause configuration commands to
912 Frequency : patchless clients only
914 Description: missing __iget() symbol export
915 Details : The __iget() symbol export is missing. To avoid the need for
916 this on patchless clients the deathrow inode reaper is turned
917 off, and we depend on the VM to clean up old inodes. This
918 dependency was during via the fix for bug 12181.
920 --------------------------------------------------------------------------------
922 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
924 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
925 INTEROPERATE with older versions automatically. In many cases a
926 special upgrade step is needed. Please read the
927 user documentation before upgrading any part of a live system.
928 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
929 version will not interoperate with 1.6 betas before beta5 (1.5.95).
930 * WARNING: Lustre configuration and startup changes are required with
931 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
937 Severity : enhancement
939 Description: MountConf
940 Details : Lustre configuration is now managed via mkfs and mount
941 commands instead of lmc and lconf. New obd types (MGS, MGC)
942 are added for dynamic configuration management. See
943 https://mail.clusterfs.com/wikis/lustre/MountConf for
946 Severity : enhancement
948 Description: dynamic OST addition
949 Details : OSTs can now be added to a live filesystem
951 Severity : enhancement
953 Description: startup order invariance
954 Details : MDTs and OSTs can be started in any order. Clients only
955 require the MDT to complete startup.
957 Severity : enhancement
959 Description: parallel, asynchronous orphan cleanup
960 Details : orphan cleanup is now performed in separate threads for each
961 OST, allowing parallel non-blocking operation.
963 Severity : enhancement
965 Description: optimized stripe assignment
966 Details : stripe assignments are now made based on ost space available,
967 ost previous usage, and OSS previous usage, in order to try
968 to optimize storage space and networking resources.
970 Severity : enhancement
972 Description: Permanently set tunables
973 Details : All writable /proc/fs/lustre tunables can now be permanently
974 set on a per-server basis, at mkfs time or on a live system.
976 Severity : enhancement
978 Description: Lustre message v2
979 Details : Add lustre message format v2.
981 Severity : enhancement
983 Description: client OST exclusion list
984 Details : Clients can be started with a list of OSTs that should be
985 declared "inactive" for known non-responsive OSTs.
989 Description: SPEC SFS validation failure on NFS v2 over lustre.
990 Details : Changes the blocksize for regular files to be 2x RPC size,
991 and not depend on stripe size.
993 Severity : enhancement
995 Description: Multiple MD RPCs in flight.
996 Details : Further unserialise some read-only MDS RPCs - learn about intents.
997 To avoid overly-overloading MDS, introduce a limit on number of
998 MDS RPCs in flight for a single client and add /proc controls
999 to adjust this limit.
1001 Severity : enhancement
1003 Description: client read/write statistics
1004 Details : Add client read/write call usage stats for performance
1005 analysis of user processes.
1006 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1007 file access. extents_stats shows chunk size distribution.
1008 extents_stats_per_process show chunk size distribution per
1011 Severity : enhancement
1013 Description: mds statistics
1014 Details : Add detailed mds operations statistics in
1015 /proc/fs/lustre/mds/*/stats.
1019 Description: Failure of copying files with lustre special EAs.
1020 Details : Client side always return success for setxattr call for lustre
1021 special xattr (currently only "trusted.lov").
1026 Description: Refcount LNET uuids
1027 Details : The global LNET uuid list grew linearly with every startup;
1028 refcount repeated list entries instead of always adding to
1031 Severity : enhancement
1033 Description: Dynamic service threads
1034 Details : Within a small range, start extra service threads
1035 automatically when the request queue builds up.
1038 Frequency : mixed-endian client/server environments
1040 Description: mixed-endian crashes
1041 Details : The new msg_v2 system had some failures in mixed-endian
1044 Severity : enhancement
1046 Description: Easy OST removal
1047 Details : OSTs can be permanently deactivated with e.g. 'lctl
1048 conf_param lustre-OST0001.osc.active=0'
1050 Severity : enhancement
1052 Description: MGS proc entries
1053 Details : Added basic proc entries for the MGS showing what filesystems
1056 Severity : enhancement
1058 Description: provide MGS failover
1059 Details : Added config lock reacquisition after MGS server failover.
1061 Severity : enhancement
1063 Description: add Linux 2.4 support
1064 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1068 Description: a large application tries to do I/O to the same resource and dies
1069 in the middle of it.
1070 Details : Check the req->rq_arrival time after the call to
1071 ost_brw_lock_get(), but before we do anything about
1072 processing it & sending the BULK transfer request. This
1073 should help move old stale pending locks off the queue as
1074 quickly as obd_timeout.
1077 Frequency : when an incorrect nid is specified during startup
1079 Description: ptlrpc connect to non-existant node causes kernel crash
1080 Details : LNET can't be re-entered from an event callback, which
1081 happened when we expire a message after the export has been
1082 cleaned up. Instead, hand the zombie cleanup off to another
1085 Severity : enhancement
1087 Description: plain/inodebits lock performance improvement
1088 Details : Grouping plain/inodebits in granted list by their request modes
1089 and bits policy, thus improving the performance of search through
1093 Frequency : only if OST filesystem is corrupted
1095 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1096 Details : for a short time RPCs with bulk IO are in the replay list,
1097 but replay of bulk IOs is unimplemented. If the OST filesystem
1098 is corrupted due to disk cache incoherency and then replay is
1099 started it is possible to trip an assertion. Avoid putting
1100 committed RPCs into the replay list at all to avoid this issue.
1103 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1105 Description: System hang on startup
1106 Details : This bug allowed the liblustre (e.g. catamount) client to
1107 return to the app before handling all startup RPCs. This
1108 could leave the node unresponsive to lustre network traffic
1109 and manifested as a server ptllnd timeout.
1111 Severity : enhancement
1113 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1114 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1115 boolean to control whether to print peer debug info when a
1116 client's RPC times out.
1119 Frequency : only for kernels with patches from Lustre below 1.4.3
1121 Description: Remove old rdonly API
1122 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1125 Frequency : only for devices with external journals
1127 Description: Set external device read-only also
1128 Details : During a commanded failover stop, we set the disk device
1129 read-only while the server shuts down. We now also set any
1130 external journal device read-only at the same time.
1133 Frequency : when upgrading from 1.4 while trying to change parameters
1135 Description: The wrong (new) MDC name was used when setting parameters for
1136 upgraded MDT's. Also allows changing of OSC (and MDC)
1137 parameters if --writeconf is specified at tunefs upgrade time.
1140 Frequency : when setting specific ost indicies
1142 Description: QOS code breaks on skipped indicies
1143 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1144 created with --index need not be sequential.
1146 Severity : enhancement
1148 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1149 Details : The uninit_groups feature works in conjunction with the kernel
1150 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1151 up the pass1 processing of e2fsck. This is a read-only feature
1152 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1153 mount filesystems that have had this feature enabled.
1155 Severity : enhancement
1157 Description: Improve multi-block allocation algorithm to avoid fragmentation
1158 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1159 allocation locality and avoid filesystem fragmentation.
1161 ------------------------------------------------------------------------------
1163 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1165 * Support for kernels:
1166 2.4.21-47.0.1.EL (RHEL 3)
1167 2.6.5-7.283 (SLES 9)
1168 2.6.9-42.0.10.EL (RHEL 4)
1169 2.6.12.6 vanilla (kernel.org)
1170 2.6.16.27-0.9 (SLES 10)
1171 * Recommended e2fsprogs version: 1.39.cfs5
1173 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1177 Frequency : occasional, depends on client load and configuration
1178 Bugzilla : 12181, 12203
1179 Description: data loss for recently-modified files
1181 Details : In some cases it is possible that recently written or created
1182 files may not be written to disk in a timely manner (this should
1183 normally be within 30s unless client IO load is very high).
1184 The problem appears as zero-length files or files that are a
1185 multiple of 1MB in size after a client crash or client eviction
1186 that are missing data at the end of the file.
1188 This problem is more likely to be hit on clients where files are
1189 repeatedly created and unlinked in the same directory, clients
1190 have a large amount of RAM, have many CPUs, the filesystem has
1191 many OSTs, the clients are rebooted frequently, and/or the files
1192 are not accessed by other nodes after being written.
1194 The presence of the problem can be detected by looking at
1195 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1196 smaller than the second (nr_unused) then dirty files will not
1197 be flushed automatically to disk. "sync; sleep 10" should be
1198 run several times on the node before unmounting it to update
1199 Lustre (this is also safe to run on nodes without this problem).
1201 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1202 that can cause this same problem, so customers using that kernel
1203 also need to update the kernel in addition to Lustre. In order
1204 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1206 It is normal that files written just before a client crash (less
1207 than 30s) may not yet have been flushed to disk, even for local
1211 Frequency : frequent on thin XT3 nodes
1213 Description: UUID collision on thin XT3 Linux nodes
1214 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1215 randomly, since we relied on an insufficiently-seeded PRNG.
1220 Description: OSS hangs after "All ost request buffers busy"
1221 Details : A deadlock between quota and journal operations caused OSS
1222 hangs after printing "All ost request buffers busy."
1225 Frequency : always on liblustre builds
1227 Description: Cleanup compiler warnings on liblustre
1230 Frequency : always on liblustre builds on XT3
1232 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1236 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1237 > 32000 subdirectories
1238 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1239 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1240 subdirectory count crosses 32000. This will aid e2fsck to
1241 correctly handle more than 32000 subdirectories.
1244 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1246 Description: System hang on startup
1247 Details : This bug allowed the liblustre (e.g. catamount) client to
1248 return to the app before handling all startup RPCs. This
1249 could leave the node unresponsive to lustre network traffic
1250 and manifested as a server ptllnd timeout.
1252 Severity : enhancement
1254 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1255 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1256 boolean to control whether to print peer debug info when a
1257 client's RPC times out.
1262 Description: make O_SYNC working on 2.6 kernels
1263 Details : 2.6 kernels use different method for mark pages for write,
1264 so need add a code to lustre for O_SYNC work.
1269 Description: Failure to close file and release space on NFS
1270 Details : Put inode details into lock acquired in ll_intent_file_open.
1271 Use mdc_intent_lock in ll_intent_open to properly
1272 detect all kind of errors unhandled by mdc_enqueue.
1277 Description: proc file read during shutdown sometimes raced obd removal,
1279 Details : Add lock to prevent obd access after proc file removal.
1282 Frequency : Only for files larger than 4GB on 32-bit clients.
1284 Description: improperly doing page alignment of locks
1285 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1286 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1291 Description: RPCs being resent when they shouldn't be
1292 Details : Some RPCs that should not be resent are being resent. This
1293 can cause inconsistencies in the RPC state machine. Do not
1294 resend such requests.
1297 Frequency : rare, only with NFS export
1299 Description: Crash on NFS re-export node
1300 Details : under very unusual load conditions an assertion is hit in
1301 ll_intent_file_open()
1304 Frequency : only if OST filesystem is corrupted
1306 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1307 Details : for a short time RPCs with bulk IO are in the replay list,
1308 but replay of bulk IOs is unimplemented. If the OST filesystem
1309 is corrupted due to disk cache incoherency and then replay is
1310 started it is possible to trip an assertion. Avoid putting
1311 committed RPCs into the replay list at all to avoid this issue.
1316 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1317 Details : Large single O_DIRECT read and write calls can fail to allocate
1318 a sufficiently large buffer to process the request. In case of
1319 allocation failure the allocation is retried with a smaller
1320 buffer and broken into smaller requests.
1322 Severity : enhancement
1324 Description: Add -o localflock option to simulate old noflock behaviour.
1325 Details : This will achieve local-only flock/fcntl locks coherentness.
1330 Description: versioning check is incomplete
1331 Details : Checking the version difference of client vs. server, report
1332 error if the gap is too big.
1337 Description: add support PG_writeback bit
1338 Details : add support for PG_writeback bit for Lustre, for more carefull
1339 work with page cache in 2.6 kernel. This also fix some deadlocks
1340 and remove hack for work O_SYNC with 2.6 kernel.
1342 Severity : enhancement
1344 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1345 Details : The uninit_groups feature works in conjunction with the kernel
1346 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1347 up the pass1 processing of e2fsck. This is a read-only feature
1348 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1349 mount filesystems that have had this feature enabled.
1351 Severity : enhancement
1353 Description: Improve multi-block allocation algorithm to avoid fragmentation
1354 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1355 allocation locality and avoid filesystem fragmentation.
1357 ------------------------------------------------------------------------------
1359 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1361 * Support for kernels:
1362 2.6.9-42.0.3.EL (RHEL 4)
1363 2.6.5-7.276 (SLES 9)
1364 2.4.21-47.0.1.EL (RHEL 3)
1365 2.6.12.6 vanilla (kernel.org)
1366 2.6.16.21-0.8 (SLES10)
1367 * Recommended e2fsprogs version: 1.39.cfs2-0
1369 * The backwards-compatible /proc/sys/portals symlink has been removed
1370 in this release. Before upgrading, please ensure that you change
1371 any configuration scripts or /etc/sysctl.conf files that access
1372 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1373 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1374 in advance of the upgrade on any system running Lustre 1.4.6 or
1375 newer, since /proc/sys/lnet was added in that version.
1376 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1380 Frequency : only when quota is used
1382 Description: avoid scanning export list for quota master
1383 Details : Change the algorithms to avoid scanning export list in order
1384 to improve the efficiency.
1387 Frequency : MDS failover only, very rarely
1389 Description: "went back in time" messages on mds failover
1390 Details : The greatest transno may be lost when the current operation
1391 finishes with an error (transno==0) and the client's last_rcvd
1392 record is over-written. Save the greatest transno in the
1393 mds_last_transno for this case.
1396 Frequency : always for specific kernels and striping counts
1398 Description: client may get "Matching packet too big" without ACL support
1399 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1400 when trying to access files in certain configurations. The
1401 clients should in fact be denied when mounting because they do
1402 not understand ACLs.
1405 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1407 Description: many clients connecting with IO in progress causes connect timeouts
1408 Details : Avoid synchronous journal commits to avoid delays caused by many
1409 clients connecting/disconnecting when bulk IO is in progress.
1410 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1411 OST_IO_PORTAL to avoid delays behind potentially many pending
1415 Frequency : occasionally with multiple writers to a single file
1417 Description: shared writes to file may result in wrong size reported by stat()
1418 Details : Allow growing of kms when extent lock is cancelled
1421 Frequency : always with random mmap IO to multi-striped file
1423 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1424 Details : Only if the hole is at the end of OST object so that kms is too
1425 small. Fix is to increase kms accordingly in ll_nopage.
1428 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1430 Description: writes to a missing object would leak memory on the OST
1431 Details : If there is an inconsistency between the MDS and OST filesystems,
1432 such that the MDS references an object that doesn't exist, writes
1433 to that object will leak memory due to incorrect cleanup in the
1434 error handling path, eventually running out of memory on the OST.
1439 Description: Creating too long symlink causes lustre errors
1440 Details : Check symlink and name lengths before sending requests to MDS.
1443 Frequency : only if flock is enabled (not on by default)
1445 Description: posix locks not released on fd closure on 2.6.9+
1446 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1447 kernels, this caused such locks not to be released on fd close and
1448 then assertions on fs unmount about still used locks.
1451 Frequency : MDS failover only, very rarely
1453 Description: clients may get ASSERTION(granted_lock != NULL)
1454 Details : When request was taking a long time, and a client was resending
1455 a getattr by name lock request. The were multiple lock requests
1456 with the same client lock handle and
1457 mds_getattr_name->fixup_handle_for_resent_request found one of the
1458 lock handles but later failed with ASSERTION(granted_lock != NULL).
1463 Description: handle->h_buffer_credits > 0, assertion failure
1464 Details : h_buffer_credits is zero after truncate, causing assertion
1465 failure. This patch extends the transaction or creates a new
1469 Frequency : NFS re-export or patchless client
1470 Bugzilla : 11179, 10796
1471 Description: Crash on NFS re-export node (__d_move)
1472 Details : We do not want to hash the dentry if we don't have a lock.
1473 But if this dentry is later used in d_move, we'd hit uninitialised
1474 list head d_hash, so we just do this to init d_hash field but
1475 leave dentry unhashed.
1478 Frequency : NFS re-export or patchless client
1480 Description: NFS exports has problem with symbolic link
1481 Details : lustre client didn't properly install dentry when re-exported
1482 to NFS or running patchless client.
1485 Frequency : NFS re-export or patchless client
1487 Description: Various nfs/patchless fixes.
1488 Details : fixes reuse disconected alias for lookup process - this fixes
1489 warning "find_exported_dentry: npd != pd",
1490 fix permission error with open files at nfs.
1491 fix apply umask when do revalidate.
1494 Frequency : occasional
1496 Description: Crash on NFS re-export node
1497 Details : calling clear_page() on the wrong pointer triggered oops in
1498 generic_mapping_read().
1501 Frequency : rarely, using O_DIRECT IO
1503 Description: unaligned directio crashes client with LASSERT
1504 Details : check for unaligned buffers before trying any requests.
1507 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1509 Description: stale data returned from RAID cache
1510 Details : If only a small amount of IO is done to the RAID device before
1511 reading it again it is possible to get stale data from the RAID
1512 cache instead of reading it from disk.
1515 Frequency : always for sles10 kernel
1517 Description: sles10 support
1518 Details : ll_follow_link: compile fixes and using of nd_set_link
1519 under newer kernels.
1522 Frequency : depends on arch, kernel and compiler version, always on sles10
1525 Description: recursive or deep enough symlinks cause stack overflow
1526 Details : getting rid of large stack-allocated variable in
1530 Frequency : depends on hardware
1532 Description: lustre write performance loss in the SLES10 kernel
1533 Details : the performance loss is caused by using of write barriers in the
1534 ext3 code. The SLES10 kernel turns barrier support on by
1535 default. The fix is to undo that change for ldiskfs.
1537 ------------------------------------------------------------------------------
1539 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1541 * Support for kernels:
1542 2.6.9-42.0.3EL (RHEL 4)
1543 2.6.5-7.276 (SLES 9)
1544 2.4.21-47.0.1.EL (RHEL 3)
1545 2.6.12.6 vanilla (kernel.org)
1549 Frequency : quota enabled and large files being deleted
1551 Description: releasing more than 4GB of quota at once hangs OST
1552 Details : If a user deletes more than 4GB of files on a single OST it
1553 will cause the OST to spin in an infinite loop. Release
1554 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1559 Description: statfs data retrieved from /proc may be stale or zero
1560 Details : When reading per-device statfs data from /proc, in the
1561 {kbytes,files}_{total,free,avail} files, it may appear
1562 as zero or be out of date.
1565 Frequency : systems with MD RAID1 external journal devices
1567 Description: lconf's call to blkid is confused by RAID1 journal devices
1568 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1569 of returning all block devices that match the journal UUID.
1572 Frequency : always, for aggregate stripe size over 4GB
1574 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1575 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1576 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1579 Frequency : always if "lfs find" used on a local file/directory
1581 Description: "lfs find" segfaults if used on a local file/directory
1582 Details : The case where a directory component was not specified wasn't
1583 handled correctly. Handle this properly.
1586 Frequency : always on ppc64
1588 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1589 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1590 perform endianness conversion.
1593 Frequency : rarely (truncate to non-zero file size after write under load)
1594 Bugzilla : 10730, 10687
1595 Description: Files padded with zeros to next 4K multiple
1596 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1597 it is possible that files that are truncated to a non-zero size
1598 immediately after being written are filled with zero bytes beyond
1599 the truncated size. No file data is lost.
1601 Severity : enhancement
1603 Description: Allow recovery/failover for liblustre clients.
1604 Details : liblustre clients were unaware of failover configurations until
1607 Severity : enhancement
1609 Description: user file locks should fail when not mounting with flock option
1610 Details : Set up an error-returning stub in ll_file_operations.lock field
1611 to prevent incorrect behaviour when client is mounted without
1612 flock option. Also, set up properly f_op->flock field for
1616 Frequency : always on ia64
1618 Description: "lfs df" loops on printing out MDS statfs information
1619 Details : The obd_ioctl_data was not initialized and in some systems
1620 this caused a failure during the ioctl that did not return
1621 an error. Initialize the struct and return an error on failure.
1624 Frequency : SLES 9 only
1626 Description: Error of copying files with lustre special EAs as root
1627 Details : Client side always return success for setxattr call for lustre
1628 special xattr (currently only "trusted.lov").
1631 Frequency : rarely on clusters with both ia64+i386 clients
1633 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1634 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1635 clients are doing concurrent writes to the same file it is
1636 possible that the ia64 clients may LASSERT because the OST
1637 extent locks are not PAGE_SIZE aligned. Ensure that grown
1638 locks are always aligned on the request boundary.
1641 Frequency : specific use, occasional
1643 Description: Overwriting in use executable truncates on-disk binary image
1644 Details : If one node attempts to overwrite an executable in use by
1645 another node, we now correctly return ETXTBSY instead of
1646 truncating the file.
1648 Severity : enhancement
1650 Description: Async OSC create to avoid the blocking unnecessarily.
1651 Details : If a OST has no remain object, system will block on the creating
1652 when need to create a new object on this OST. Now, ways use
1653 pre-created objects when available, instead of blocking on an
1654 empty osc while others are not empty. If we must block, we block
1655 for the shortest possible period of time.
1660 Description: chmod on Lustre root is propagated to other clients
1661 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1662 invalid by the follow_mount time.
1665 Frequency : liblustre clients only
1667 Description: Race in 'instant cancel' lock handling could lead to such locks
1668 never to be granted in case of SMP MDS
1669 Details : Do not destroy not yet granted but cbpending locks in
1673 Frequency : replay/resend of open
1675 Description: non null lock assetion failure in mds_intent_policy
1676 Details : Trying to replay/resend lockless open requests resulted in
1677 mds_open() returning 0 with no lock. Now it sets a flag if
1678 a lock is going to be returned.
1680 Severity : enhancement
1682 Description: Checksum enhancements
1683 Details : New checksum enhancements allow for resending RPCs that failed
1686 Severity : enhancement
1688 Description: Tunables on number of dirty pages in cacche
1689 Details : Allow to set limit on number of dirty pages cached.
1694 Description: client crash on unmount - lock still has references
1695 Details : In some error handling cases it was possible to leak a lock
1696 reference on a client while accessing a file. This was not
1697 harmful to the client during operation, but would cause the
1698 client to crash when the filesystem is unmounted.
1701 Frequency : specific case, rare
1703 Description: ETXTBSY on mds though file not in use by client
1704 Details : ETXTBSY is no longer incorrectly returned when attempting to
1705 chmod or chown a directory that the user previously tried to
1706 execute or a currently-executing binary.
1709 Frequency : extremely rare except on liblustre-based clients
1711 Description: Lustre space not freed when files are deleted
1712 Details : Clean up open-unlinked files after client eviction. Previously
1713 the unlink was skipped and the files remained as orphans.
1718 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1719 Details : In some cases it was possible to send a blocking callback to
1720 a client doing a glimpse, even though that client didn't get
1721 a lock granted. When the glimpse lock is cancelled on the OST
1722 the freed lock is left on the waiting list and corrupted the list.
1725 Frequency : all core dumps
1727 Description: Broke core dumps to lustre
1728 Details : Negative dentry may be unhashed if parent does not have UPDATE
1729 lock, but some callers, e.g. do_coredump, expect dentry to be
1730 hashed after successful create, hash it in ll_create_it.
1732 ------------------------------------------------------------------------------
1734 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1736 * Support for kernels:
1737 2.6.9-42.0.2.EL (RHEL 4)
1738 2.6.5-7.276 (SLES 9)
1739 2.4.21-40.EL (RHEL 3)
1740 2.6.12.6 vanilla (kernel.org)
1744 Frequency : always on RHEL 3
1746 Description: Number of open files grows over time
1747 Details : The number of open files grows over time, whether or not
1748 Lustre is started. This was due to a filp leak introduced
1749 by one of our kernel patches.
1751 ------------------------------------------------------------------------------
1753 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1755 * Support for kernels:
1756 2.6.9-42.EL (RHEL 4)
1757 2.6.5-7.276 (SLES 9)
1758 2.4.21-40.EL (RHEL 3)
1759 2.6.12.6 vanilla (kernel.org)
1764 Bugzilla : 5719, 9635, 9792, 9684
1765 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1766 Details : If a server is under heavy load and cannot reply to new
1767 connection requests before the client resends the (re)connect,
1768 the connection handling code can behave badly if two service
1769 threads are concurrently handing separate (re)connections from
1770 the same client. Add better locking to the connection handling
1771 code, and ensure that only a single connection will be processed
1772 for a given client UUID, even if the lock is dropped.
1774 Severity : enhancement
1776 Description: add TCP zero-copy support to kernel
1777 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1778 sends if the hardware supports scatter-gather and checksumming.
1779 This allows socklnd to do client-write and server-read more
1780 efficiently and reduce CPU utilization from skbuf copying.
1783 Frequency : only if NFS exporting from client
1785 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1786 Details : If setting attributes on a file created under NFS that had
1787 never been opened it would be possible to oops the client
1788 if the file had no objects.
1791 Frequency : always for liblustre
1793 Description: liblustre client does MDS+OSTs setattr RPC for each write
1794 Details : When doing a write from a liblustre client, the client
1795 incorrectly issued an RPC to the MDS and each OST the file was
1796 striped over in order to update the timestamps. When writing
1797 with small chunks and many clients this could overwhelm the MDS
1798 with RPCs. In all cases it would slow down the write because
1799 these RPCs are unnecessary.
1801 Severity : enhancement
1803 Description: allow number of MDS service threads to be changed at module load
1804 Details : It is now possible to change the number of MDS service threads
1805 running. Adding "options mds mds_num_threads={N}" to the MDS's
1806 /etc/modprobe.conf will set the number of threads for the next
1807 time Lustre is restarted (assuming the "mds" module is also
1808 reloaded at that time). The default number of threads will
1809 stay the same, 32 for most systems.
1814 Description: OST crash if filesystem is unformatted or corrupt
1815 Details : If an OST is started on a device that has never been formatted
1816 or if the filesystem is corrupt and cannot even mount then the
1817 error handling cleanup routines would dereference a NULL pointer.
1822 Description: NULL pointer deref in llap_from_page.
1823 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1824 incorrect) mapping if the page was truncated/reclaimed while it was
1825 searched for. Check for this condition and skip such pages when
1826 doing readahead. Introduce extra check to llap_from_page() to
1827 verify page->mapping->host is non-NULL (so page is not anonymous).
1830 Frequency : Sometimes when using sys_sendfile
1832 Description: "page not covered by a lock" warnings from ll_readpage
1833 Details : sendfile called ll_readpage without right page locks present.
1834 Now we introduced ll_file_sendfile that does necessary locking
1835 around call to generic_file_sendfile() much like we do in
1839 Frequency : with certain MDS communication failures at client mount time
1841 Description: NULL pointer deref after failed client mount
1842 Details : a client connection request may delayed by the network layer
1843 and not be sent until after the PTLRPC layer has timed out the
1844 request. If the client fails the mount immediately it will try
1845 to clean up before the network times out the request. Add a
1846 reference from the request import to the obd device and delay
1847 the cleanup until the network drops the request.
1850 Frequency : occasionally during client (re)connect
1852 Description: assertion failure during client (re)connect
1853 Details : processing a client connection request may be delayed by the
1854 client or server longer than the client connect timeout. This
1855 causes the client to resend the connection request. If the
1856 original connection request is replied in this interval, the
1857 client may trip an assertion failure in ptlrpc_connect_interpret()
1858 which thought it would be the only running connect process.
1861 Frequency : only with obd_echo servers and clients that are rebooted
1863 Description: kernel BUG accessing uninitialized data structure
1864 Details : When running an obd_echo server it did not start the ping_evictor
1865 thread, and when a client was evicted an uninitialized data
1866 structure was accessed. Start the ping_evictor in the RPC
1867 service startup instead of the OBD startup.
1869 Severity : enhancement
1870 Bugzilla : 10193 (patchless)
1871 Description: Remove dependency on various unexported kernel interfaces.
1872 Details : No longer need reparent_to_init, exit_mm, exit_files,
1873 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1876 Frequency : rare (only users of deprecated and unsupported LDAP config)
1878 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1879 Details : LDAP apparently contains 'lustreName' attributes instead of
1880 'name'. A simple remapping of the name is sufficient.
1883 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1885 Description: waiting_locks_callback trips kernel BUG if client is evicted
1886 Details : Running with the dump_on_timeout debug flag turned on makes
1887 it possible that the waiting_locks_callback() can try to dump
1888 the Lustre kernel debug logs from an interrupt handler. Defer
1889 this log dumping to the expired_lock_main() thread.
1891 Severity : enhancement
1893 Description: Support NFS exporting on 2.6 kernels.
1894 Details : Implement non-rawops metadata methods for NFS server to use without
1895 changing NFS server code.
1898 Frequency : very rare (synthetic metadata workload only)
1900 Description: two racing renames might cause an MDS thread to deadlock
1901 Details : Running the "racer" program may cause one MDS thread to rename
1902 a file from being the source of a rename to being the target of
1903 a rename at exactly the same time that another thread is doing
1904 so, and the second thread has already enqueued these locks after
1905 doing a lookup of the target and is trying to relock them in
1906 order. Ensure that we don't try to re-lock the same resource.
1909 Frequency : only very large systems with liblustre clients
1911 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1912 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1913 all OSTs in parallel. This allows the eviction of stale liblustre
1914 clients to proceed much faster than if they were done in series,
1915 and also offers similar improvements for other set_info RPCs.
1920 Description: excessive CPU usage during initial read phase on client
1921 Details : During the initial read phase on a client, it would agressively
1922 retry readahead on the file, consuming too much CPU and impacting
1923 performance (since 1.4.5.8). Improve the readahead algorithm
1924 to avoid this, and also improve some other common cases (read
1925 of small files in particular, where "small" is files smaller than
1926 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1931 Description: MDS crash when receiving packet with unknown intent.
1932 Details : Do not LBUG in unknown intent case, just return -EFAULT
1934 Severity : enhancement
1935 Bugzilla : 9293, 9385
1936 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1937 Details : Do not serialize getattr (non-intent version) and statfs.
1940 Frequency : occasional, when OST network is overloaded/intermittent
1942 Description: client evicted by OST after bulk IO timeout
1943 Details : If a client sends a bulk IO request (read or write) the OST
1944 may evict the client if it is unresposive to its data GET/PUT
1945 request. This is incorrect if the network is overloaded (takes
1946 too long to transfer the RPC data) or dropped the OST GET/PUT
1947 request. There is no need to evict the client at all, since
1948 the pinger and/or lock callbacks will handle this, and the
1949 client can restart the bulk request.
1952 Frequency : Always when mmapping file with no objects
1954 Description: client crashes when mmapping file with no objects
1955 Details : Check that we actually have objects in a file before doing any
1956 operations on objects in ll_vm_open, ll_vm_close and
1962 Description: Request leak when working with deleted CWD
1963 Details : Introduce advanced request refcount tracking for requests
1964 referenced from lustre intent.
1966 Severity : Enhancement
1968 Description: Cache open file handles on client.
1969 Details : MDS now will return special lock along with openhandle, if
1970 requested and client is allowed to hold openhandle, even if unused,
1971 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1972 closing files for every read/write openration.
1974 Severity : Enhancement
1976 Description: Cache open negative dentries on client when possible.
1977 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1978 negative dentries on lock revocation.
1983 Description: Remounting a client read-only wasn't possible with a zconf mount
1984 Details : It wasn't possible to remount a client read-only with llmount.
1986 Severity : enhancement
1987 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1988 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1989 MPICH is included to add Lustre-specific ADIO interfaces.
1990 This is based closely on the UFS ADIO layer and only differs
1991 in file creation, in order to allow the OST striping to be set.
1992 This is user-contributed code and not supported by CFS.
1997 Description: extended inode attributes (immutable, append-only) work improperly
1998 when 2.4 and 2.6 kernels are used on client/server or vice versa
1999 Details : Introduce kernel-independent values for these flags.
2001 Severity : enhancement
2004 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2005 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2006 granularity. Now, Fractional MB granularity is be supported,
2007 this is very useful for low memory system.
2009 Severity : enhancement
2011 Description: Getattr by fid
2012 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2013 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2014 have locks and avoids getattr rpc requests for it on every stat.
2017 Frequency : Always, for filesystems larger than 2TB
2019 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2020 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2021 filesystems larger than 2TB to be mounted with mballoc enabled.
2024 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2026 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2027 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2028 filesystems larger than 2TB is mounted then the kernel will
2029 silently corrupt the start of the filesystem. CONFIG_LBD is
2030 enabled for all CFS-supported kernels, but the possibility of
2031 this happening with a modified kernel config exists.
2033 Severity : enhancement
2035 Description: add client O_DIRECT support for 2.6 kernels
2036 Details : It is now possible to do O_DIRECT reads and writes to files
2037 in the Lustre client mountpoint on 2.6 kernel clients.
2039 Severity : enhancement
2041 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2042 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2043 parallel, not waiting for response from every OST before sending
2044 a rpc to the next OST.
2049 Description: setattr vs write race when updating file timestamps
2050 Details : Client processes that update a file timestamp into the past
2051 right after writing to the file (e.g. tar) it is possible that
2052 the updated file modification time can be reset to the current
2053 time due to a race between processing the setattr and write RPC.
2055 Severity : enhancement
2057 Description: Bring 'lfs find' closer in line with regular Linux find.
2058 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2059 -print0 options and obtains all the needed info through the lustre
2062 Severity : enhancement
2064 Description: support up to 1024 configured devices on one node
2065 Details : change obd_dev array from statically allocated to dynamically
2066 allocated structs as they are first used to reduce memory usage
2071 Description: Flush dirty partially truncated pages during truncate
2072 Details : Immediatelly flush partially truncated pages in filter_setattr,
2073 this way we completely avoid having any pages in page cache on OST
2074 and can retire ugly workarounds during writes to flush such pages.
2079 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2080 Details : protect inode from truncation within vfs_unlink() context
2081 just take a reference before calling vfs_unlink() and release it
2082 when parent's i_sem is free.
2087 Description: last_id value checked outside lock on OST caused LASSERT failure
2088 Details : If there were multiple MDS->OST object precreate requests in
2089 flight, it was possible that the OST's last object id was checked
2090 outside a lock and incorrectly tripped an assertion. Move checks
2091 inside locks, and discard old precreate requests.
2094 Frequency : always, if extents are used on OSTs
2096 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2097 Details : OSTs using the extents format would not zero the high 16 bits of
2098 the index physical block number. This is not a problem for any
2099 OST filesystems smaller than 16TB, and no kernels support ext3
2100 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2101 new/modified files) and can be fixed for existing filesystems
2102 with e2fsprogs-1.39-cfs1.
2107 Description: import connection selection may be incorrect if timer wraps
2108 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2109 import connections to be ignored if the 32-bit jiffies counter
2110 wraps. Use a 64-bit jiffies counter.
2113 Frequency : very large clusters immediately after boot
2115 Description: LNET request buffers exhausted under heavy short-term load
2116 Details : If a large number of client requests are generated on a service
2117 that has previously never seen so many requests it is possible
2118 that the request buffer growth cannot keep up with the spike in
2119 demand. Instead of dropping incoming requests, they are held in
2120 the LND until the RPC service can accept more requests.
2123 Frequency : Sometimes during replay
2125 Description: Assertion failure in ll_local_open after replay.
2126 Details : If replay happened on an open request reply before we were able
2127 to set replay handler, reply will become not swabbed tripping the
2128 assertion in ll_local_open. Now we set the handler right after
2129 recognising of open request
2132 Frequency : very rare
2134 Description: kernel reports "badness in vsnprintf"
2135 Details : Reading from the "recovery_status" /proc file in small chunks
2136 may cause a negative length in lprocfs_obd_rd_recovery_status()
2137 call to vsnprintf() (which is otherwise harmless). Exit early
2138 if there is no more space in the output buffer.
2140 Severity : enhancement
2142 Description: clear OBD RPC statistics by writing to them
2143 Details : It is now possible to clear the OBD RPC statistics by writing
2144 to the "stats" file.
2149 Description: Client mtime is not the same on different clients after utimes
2150 Details : In some cases, the client was using the utimes() syscall on
2151 a file cached on another node. The clients now validate the
2152 ctime from the MDS + OSTs to determine which one is right.
2157 Description: Inability to activate failout mode
2158 Details : lconf script incorrectly assumed that in pythong string's numeric
2159 value is used in comparisons.
2162 Frequency : always with multiple stripes per file
2164 Description: Inefficient object allocation for mutli-stripe files
2165 Details : When selecting which OSTs to stripe files over, for files with
2166 a stripe count that divides evenly into the number of OSTs,
2167 the MDS is always picking the same starting OST for each file.
2168 Return the OST selection heuristic to the original design.
2173 Description: mount failures may take full timeout to return an error
2174 Details : Under some heavy load conditions it is possible that a
2175 failed mount can wait for the full obd_timeout interval,
2176 possibly several minutes, before reporting an error.
2177 Instead return an error as soon as the status is known.
2179 Frequency : quota enabled and large files being deleted
2181 Description: releasing more than 4GB of quota at once hangs OST
2182 Details : If a user deletes more than 4GB of files on a single OST it
2183 will cause the OST to spin in an infinite loop. Release
2184 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2189 Description: statfs data retrieved from /proc may be stale or zero
2190 Details : When reading per-device statfs data from /proc, in the
2191 {kbytes,files}_{total,free,avail} files, it may appear
2192 as zero or be out of date.
2195 Frequency : systems with MD RAID1 external journal devices
2197 Description: lconf's call to blkid is confused by RAID1 journal devices
2198 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2199 of returning all block devices that match the journal UUID.
2202 Frequency : always, for aggregate stripe size over 4GB
2204 Description: assertion fails when trying to use 4GB stripe size
2205 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2206 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2209 Frequency : always on ppc64
2211 Description: the first write on an ext3 filesystem with mballoc got stuck
2212 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2213 perform endianness conversion.
2215 ------------------------------------------------------------------------------
2217 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2219 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2220 INTEROPERATE with older versions automatically. Please read the
2221 user documentation before upgrading any part of a live system.
2222 * WARNING: Lustre networking configuration changes are required with
2223 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2226 * Support for kernels:
2227 2.6.9-22.0.2.EL (RHEL 4)
2228 2.6.5-7.244 (SLES 9)
2229 2.6.12.6 vanilla (kernel.org)
2232 Severity : enhancement
2233 Bugzilla : 7981/8208
2234 Description: Introduced Lustre Networking (LNET)
2235 Details : LNET is new networking infrastructure for Lustre, it includes
2236 a reorganized network configuration mode (see the user
2237 documentation for full details) as well as support for routing
2238 between different network fabrics. Lustre Networking Devices
2239 (LNDs) for the supported network fabrics have also been
2240 created for this new infrastructure.
2242 Severity : enhancement
2243 Description: Introduced Access control lists
2244 Details : clients can set ACLs on files and directories in order to have
2245 more fine-grained permissions than the standard Unix UGO+RWX.
2246 The MDS must be started with the "-o acl" mount option.
2248 Severity : enhancement
2249 Description: Introduced filesystem quotas
2250 Details : Administrators may now establish per-user quotas on the
2253 Severity : enhancement
2255 Description: Configuration change for the XT3
2256 The PTLLND is now used to run Lustre over Portals on the XT3
2257 The configure option(s) --with-cray-portals are no longer used.
2258 Rather --with-portals=<path-to-portals-includes> is used to
2259 enable building on the XT3. In addition to enable XT3 specific
2260 features the option --enable-cray-xt3 must be used.
2265 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2266 Details : A race between allocating a new llcd and re-getting the llcd_lock
2267 allowed another thread to grab newly-allocated llcd.
2269 Severity : enhancement
2271 Description: 2.6 OST async journal commit and locking fix to improve performance
2272 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2273 2.6 kernels are now async as they already were in 2.4 kernels so
2274 that they can commit concurrently with the network bulk transfer.
2275 For block-allocated files the filter allocation semaphore is held
2276 to avoid filesystem fragmentation during allocation. BKL lock
2277 removed for 2.6 xattr operations where it is no longer needed.
2282 Description: lconf incorrectly determined whether two IP networks could talk
2283 Details : In some more complicated routing and multiple-network
2284 configurations, lconf will avoid trying to make a network
2285 connection to a disjoint part of the IP space. It was doing the
2286 math incorrectly for one set of cases.
2291 Description: Fix for potential infinite loop processing records in an llog.
2292 Details : If an llog record is corrupted/zeroed, it is possible to loop
2293 forever in llog_process(). Validate the llog record length
2294 and skip the remainder of the block on error.
2297 Frequency : occasional (liblustre only)
2299 Description: liblustre could not open files whose last component is a symlink
2300 Details : sysio_path_walk() would incorrectly pass the open intent to
2301 intermediate path components.
2304 Frequency : rare (liblustre only with non-standard tuning)
2305 Bugzilla : 7201 (7350)
2306 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2307 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2308 after it has been released, only do this for referenced locks
2310 Severity : enhancement
2312 Description: specifying an (invalid) directory default stripe_size of -1
2313 would reset the directory default striping
2314 Details : stripe_size -1 was used internally to signal directory stripe
2315 removal, now use "all default" to signal dir stripe removal
2316 as a directory striping of "all default" is not useful
2319 Frequency : common for large clusters running liblustre clients
2321 Description: doing an ls when liblustre clients are running is slow
2322 Details : sending a glimpse AST to a liblustre client waits for every AST
2323 to time out, as liblustre clients will not respond. Since they
2324 cannot cache data we refresh the OST lock LVB from disk instead.
2326 Severity : enhancement
2328 Description: doing an ls at the same time as file IO can be slow
2329 Details : enqueue and other "small" requests can be blocked behind many
2330 large IO requests. Create a new OST IO portal for non-IO
2331 requests so they can be processed faster.
2334 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2336 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2337 created file with mode 000
2338 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2339 allow an ftruncate() as a non-root user to a file with mode 000.
2340 Lustre now respects this flag to disable mode checks when
2341 truncating a file owned by the user
2344 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2347 Description: Revoking locks from clients that went dead or catatonic might take
2349 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2350 cancellation of such locks instant on servers without waiting for
2351 any reply from clients. Clients drops these locks when cancel
2352 notification from server is received without replying.
2355 Frequency : liblustre-only, when liblustre client dies or becomes busy
2357 Description: Doing ls on Linux clients can take a long time with active
2359 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2360 granting such locks to it in the first place if possible. Locks
2361 are taken by proxy on the OST during the read or write and
2362 dropped immediately afterward. Add connect flags handling, do
2363 not grant locks to liblustre clients for glimpse ASTs.
2365 Severity : enhancement
2367 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2368 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2369 sequential reads and behaves badly with applications doing random
2370 reads. Improve it by reading ahead at least read region, and
2371 avoiding excessive large RPC for small reads.
2373 Severity : enhancement
2375 Description: Creating more than 1000 files for a single job may cause a load
2376 imbalance on the OSTs if there are also a large number of OSTs.
2377 Details : qos_prep_create() uses an OST index reseed value that is an
2378 even multiple of the number of available OSTs so that if the
2379 reseed happens in the middle of the object allocation it will
2380 still utilize the OSTs as uniformly as possible.
2385 Description: OST or MDS may oops in ping_evictor_main()
2386 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2387 but doesn't restart at beginning of obd_exports_timed list
2390 Severity : enhancement
2392 Description: improve by-nid export eviction on the MDS and OST
2393 Details : allow multiple exports with the same NID to be evicted at one
2394 time without re-searching the exports list.
2397 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2399 Description: MDS may oops in groups_free()
2400 Details : in rare race conditions a newly allocated group_info struct is
2401 freed again, and this can be NULL. The 2.4 compatibility code
2402 for groups_free() checked for a NULL pointer, but 2.6 did not.
2405 Frequency : common for liblustre clients doing little filesystem IO
2406 Bugzilla : 9352, 7313
2407 Description: server may evict liblustre clients accessing contended locks
2408 Details : if a client is granted a lock or receives a completion AST
2409 with a blocking AST already set it would not reply to the AST
2410 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2413 Frequency : lfs setstripe, only systems with more than 160 OSTs
2415 Description: unable to set striping with a starting offset beyond OST 160
2416 Details : llapi_create_file() incorrectly limited the starting stripe
2417 index to the maximum single-file stripe count.
2420 Frequency : LDAP users only
2422 Description: lconf did not handle in-kernel recovery with LDAP properly
2423 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2425 Severity : enhancement
2427 Description: bind OST threads to NUMA nodes to improve performance
2428 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2429 node and do their allocations there to localize memory access
2431 Severity : enhancement
2433 Description: llmount can determine client NID directly from Myrinet (GM)
2434 Details : the client NID code from gmnalnid was moved directly into
2435 llmount, removing the need to use this or specifying the
2436 client NID explicitly when mounting GM clients with zeroconf
2439 Frequency : if client is started with down MDS
2441 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2442 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2443 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2444 ptlrpc_import_delayed_req() because we didn't check if the
2445 request was interrupted, and we also didn't break out of the
2446 event loop if there was no timeout
2451 Description: data loss during non-page-aligned writes to a single file from
2452 both multiple nodes and multiple threads on one node at same time
2453 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2454 inconsistency led to false short-reads, that were cached and later
2455 used by ->prepare_write() to fill in partially written page,
2456 leading to data loss.
2459 Frequency : always, if lconf --abort_recovery used
2461 Description: lconf --abort_recovery fails with 'Operation not supported'
2462 Details : lconf was attempting to abort recovery on the MDT device and not
2465 Severity : enhancement
2467 Description: remove cleanup logs
2468 Details : replace lconf-generated cleanup logs with lustre internal
2469 cleanup routines. Eliminates the need for client-cleanup and
2472 Severity : enhancement
2474 Description: add support for EAs (user and system) on lustre filesystems
2475 Details : it is now possible to store extended attributes in the Lustre
2476 client filesystem, and with the user_xattr mount option it
2477 is possible to allow users to store EAs on their files also
2479 Severity : enhancement
2481 Description: Add possibility (config option) to show minimal available OST free
2483 Details : When compiled with --enable-mindf configure option, statfs(2)
2484 (and so, df) will return least minimal free space available from
2485 all OSTs as amount of free space on FS, instead of summary of
2486 free spaces of all OSTs.
2488 Severity : enhancement
2490 Description: do not expand extent locks acquired on OST-side
2491 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2492 by server: they are not cached anyway.
2495 Frequency : when mmap is used/binaries executed from Lustre
2497 Description: Unmmap pages before throwing them away from read cache.
2498 Details : llap_shrink cache now attempts to unmap pages before discarding
2499 them (if unmapping failed - do not discard). SLES9 kernel has
2500 extra checks that trigger if this unmapping is not done first.
2505 Description: lconf didn't resolve symlinks before checking to see whether a
2506 given mountpoint was already in use
2509 Frequency : when migrating failover services
2510 Bugzilla : 6395, 9514
2511 Description: When migrating a subset of services from a node (e.g. failback
2512 from a failover service node) the remaining services would
2513 time out and evict clients.
2514 Details : lconf --force (implied by --failover) sets the global obd_timeout
2515 to 5 seconds in order to quickly disconnect, but this caused
2516 other RPCs to time out too quickly. Do not change the global
2517 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2519 Severity : enhancement
2520 Frequency : if MDS is started with down OST
2521 Bugzilla : 9439,5706
2522 Description: Allow startup/shutdown of an MDS without depending on the
2523 availability of the OSTs.
2524 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2525 Add appropriate locking and lov-osc refcounts for safe
2526 cleaning. Add osc abort_inflight calls in case the
2527 synchronize never started.
2530 Frequency : occasional (Cray XT3 only)
2532 Description: root not authorized to access files in CRAY_PORTALS environment
2533 Details : The client process capabilities were not honoured on the MDS in
2534 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2535 been accessed by an authorized user then root was able to access
2536 the file on the local client also. The root user capabilities
2537 are now allowed on the MDS, as this environment has secure UID.
2540 Frequency : occasional
2542 Description: ldiskfs "too long searching" message happens too often
2543 Details : A debugging message (otherwise harmless) prints too often on
2544 the OST console. This has been reduced to only happen when
2545 there are fragmentation problems on the filesystem.
2550 Description: Division by zero in statfs when all OSCs are inactive
2551 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2552 lov_statfs divides by value returned by lov_get_stripecnt().
2556 Bugzilla : 9489, 3273
2557 Description: First write from each client to each OST was only 4kB in size,
2558 to initialize client writeback cache, which caused sub-optimal
2559 RPCs and poor layout on disk for the first writen file.
2560 Details : Clients now request an initial cache grant at (re)connect time
2561 and so that they can start streaming writes to the cache right
2562 away and always do full-sized RPCs if there is enough data.
2563 If the OST is rebooted the client also re-establishes its grant
2564 so that client cached writes will be honoured under the grant.
2569 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2570 Details : Now I/O RPCs go to different portal number and (presumably) fast
2571 lock requests (and glimses) and other RPCs get their own service
2572 threads pool that should be able to service those RPCs
2575 Severity : enhancement
2577 Description: Ability to exchange lustre version between client and servers and
2578 issue warnings at client side if client is too old. Also for
2579 liblustre clients there is ability to refuse connection of too old
2581 Details : New 'version' field is added to connect data structure that is
2582 filled with version info. That info is later checked by server and
2586 Frequency : rare, liblustre only.
2587 Bugzilla : 9296, 9581
2588 Description: Two simultaneous writes from liblustre at offset within same page
2589 might proceed at the same time overwriting eachother with stale
2591 Details : I/O lock withing llu_file_prwv was released too early, before data
2592 actually was hitting the wire. Extended lock-holding time until
2593 server acknowledges receiving data.
2596 Frequency : extremely rare. Never observed in practice.
2598 Description: avoid generating lustre_handle cookie of 0.
2599 Details : class_handle_hash() generates handle cookies by incrementing
2600 global counter, and can hit 0 occasionaly (this is unlikely, but
2601 not impossible, because initial value of cookie counter is
2602 selected randonly). Value of 0 is used as a sentinel meaning
2603 "unassigned handle" --- avoid it. Also coalesce two critical
2604 sections in this function into one.
2606 Severity : enhancement
2608 Description: allow liblustre clients to delegate truncate locking to OST
2609 Details : To avoid overhead of locking, liblustre client instructs OST to
2610 take extent lock in ost_punch() on client's behalf. New connection
2611 flag is added to handle backward compatibility.
2613 Severity : enhancement
2614 Bugzilla : 4928, 7341, 9758
2615 Description: allow number of OST service threads to be specified
2616 Details : a module parameter allows the number of OST service threads
2617 to be specified via "options ost ost_num_threads={N}" in the
2618 OSS's /etc/modules.conf or /etc/modprobe.conf.
2622 Bugzilla : 6146, 9635, 9895
2623 Description: servers crash with bad pointer in target_handle_connect()
2624 Details : In rare cases when a client is reconnecting it was possible that
2625 the connection request was the last reference for that export.
2626 We would temporarily drop the export reference and get a new
2627 one, but this may have been the last reference and the export
2628 was just destroyed. Get new reference before dropping old one.
2630 Severity : enhancement
2631 Frequency : if client is started with failover MDS
2633 Description: Allow multiple MDS hostnames in the mount command
2634 Details : Try to read the configuration from all specified MDS
2635 hostnames during a client mount in case the "primary"
2638 Severity : enhancement
2640 Description: Stop sending data to evicted clients as soon as possible.
2641 Details : Check if the client we are about to send or are sending data to
2642 was evicted already. (Check is done every second of waiting,
2643 for which l_wait_event interface was extended to allow checking
2644 of exit condition at specified intervals).
2647 Frequency : rare, normally only when NFS exporting is done from client
2649 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2651 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2654 Frequency : always, if srand() is called before liblustre initialization
2656 Description: Liblustre uses system PRNG disturbing its usage by user application
2657 Details : Introduce internal to lustre fast and high-quality PRNG for
2658 lustre usage and make liblustre and some other places in generic
2659 lustre code to use it.
2661 Severity : enhancement
2662 Bugzilla : 9477, 9557, 9870
2663 Description: Verify that the MDS configuration logs are updated when xml is
2664 Details : Check if the .xml configuration logs are newer than the config
2665 logs stored on the MDS and report an error if this is the case.
2666 Request --write-conf, or allow starting with --old_conf.
2668 Severity : enhancement
2670 Description: Handle symlinks in the path when checking if Lustre is mounted.
2671 Details : Resolve intermediate symlinks when checking if a client has
2672 mounted a filesystem to avoid duplicate client mounts.
2677 Description: lconf can hit an error exception but still return success.
2678 Details : The lconf command catches the Command error exception at the top
2679 level script context and will exit with the associated exit
2680 status, but doesn't ensure that this exit status is non-zero.
2685 Description: failure of ptlrpc thread startup can cause oops
2686 Details : Starting a ptlrpc service thread can fail if there are a large
2687 number of threads or the server memory is very fragmented.
2688 Handle this without oopsing.
2691 Frequency : always, only if liblustre and non-default acceptor port was used
2693 Description: liblustre cannot connect to servers with non-default acceptor port
2694 Details : tcpnal_set_default_params() was not called and was therefore
2695 ignoring the environment varaible TCPNAL_PORT, as well as other
2696 TCPNAL_ environment variables
2701 Description: two objects could be created on the same OST for a single file
2702 Details : If an OST is down, in some cases it was possible to create two
2703 objects on a single OST for a single file. No problems other
2704 than potential performance impact and spurious error messages.
2708 Bugzilla : 5681, 9562
2709 Description: Client may oops in ll_unhash_aliases
2710 Details : Client dcache may become inconsistent in race condition.
2711 In some cases "getcwd" can fail if the current directory is
2717 Description: Inode refcounting problems in NFS export code
2718 Details : link_raw functions used to call d_instantiate without obtaining
2719 extra inode reference first.
2723 Bugzilla : 9942, 9903
2724 Description: Referencing freed requests leading to crash, memleaks with NFS.
2725 Details : We used to require that call to ll_revalidate_it was always
2726 followed by ll_lookup_it. Also with revalidate_special() it is
2727 possible to call ll_revalidate_it() twice for the same dentry
2728 even if first occurence returned success. This fix changes semantic
2729 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2730 reference on a request referred from the intent.
2731 ll_intent_release() then releases such a request.
2734 Frequency : rare, normally benchmark loads only
2736 Description: unlinked inodes were kept in memory on the client
2737 Details : If a client is repeatedly creating and unlinking files it
2738 can accumulate a lot of stale inodes in the inode slab cache.
2739 If there is no other client load running this can cause the
2740 client node to run out of memory. Instead flush old inodes
2741 from client cache that have the same inode number as a new inode.
2744 Frequency : SLES9 2.6.5 kernel and long filenames only
2745 Bugzilla : 9969, 10379
2746 Description: utime reports stale NFS file handle
2747 Details : SLES9 uses out-of-dentry names in some cases, which confused
2748 the lustre dentry revalidation. Change it to always use the
2752 Frequency : rare, unless heavy write-truncate concurrency is continuous
2753 Bugzilla : 4180, 6984, 7171, 9963, 9331
2754 Description: OST becomes very slow and/or deadlocked during object unlink
2755 Details : filter_destroy() was holding onto the parent directory lock
2756 while truncating+unlinking objects. For very large objects this
2757 may block other threads for a long time and slow overall OST
2758 responsiveness. It may also be possible to get a lock ordering
2759 deadlock in this case, or run out of journal credits because of
2760 the combined truncate+unlink. Solution is to do object truncate
2761 first in one transaction without parent lock, and then do the
2762 final unlink in a new transaction with the parent lock. This
2763 reduces the lock hold time dramatically.
2766 Frequency : rare, 2.4 kernels only
2768 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2769 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2770 before unmounting their filesystem in order to avoid deadlock.
2771 The kernel_locked() function in 2.4 kernels only checks whether
2772 the kernel lock is held, not whether it is this process that is
2773 holding it as 2.6 kernels do.
2778 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2779 Details : The client ptlrpc code may be trying to reconnect to a down
2780 server before a previous connection attempt has timed out.
2781 Increase the reconnect interval to be longer than the connection
2782 timeout interval to avoid sending duplicate connections to
2786 Frequency : echo_client brw_test command
2788 Description: fix echo_client to work with OST preallocated code
2789 Details : OST preallocation code (5137) didn't take echo_client IO path
2790 into account: echo_client calls filter methods outside of any
2791 OST thread and, hence, there is no per-thread preallocated
2792 pages and buffers to use. Solution: hijack pga pages for IO. As
2793 a byproduct, this avoids unnecessary data copying.
2797 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2798 Description: Client can oops in mdc_commit_close() after open replay
2799 Details : It was possible for the MDS to return an open request with no
2800 transaction number in mds_finish_transno() if the client was
2801 evicted, but without actually returning an error. Clients
2802 would later try to replay that open and may trip an assertion
2803 Simplify the client close codepath, and always return an error
2804 from the MDS in case the open is not successful.
2807 Frequency : rare, 2.6 OSTs only
2809 Description: OST may deadlock under high load on fragmented files
2810 Details : If there was a heavy load and highly-fragmented OST filesystems
2811 it was possible to have all the OST threads deadlock waiting on
2812 allocation of biovecs, because the biovecs were not released
2813 until the entire RPC IO was completed. Instead, release biovecs
2814 as soon as they are complete to ensure forward IO progress.
2816 Severity : enhancement
2818 Description: Support for specifying external journal device at mount
2819 Details : If an OST or MDS device is formatted with an external journal
2820 device, this device major/minor is stored in the ext3 superblock
2821 and may not be valid for failover. Allow detecting and
2822 specifying the external journal at mount time.
2827 Description: Mounting an MDS with pending unlinked files may cause oops
2828 Details : target_finish_recovery() calls mds_postrecov() which returned
2829 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2830 considers this an error and immediately begins cleaning up the
2831 lov, just after starting the mds_lov process
2833 Severity : enhancement
2835 Description: Implement 'lfs df' to report actual free space on per-OST basis
2836 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2837 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2838 usage of objects; '-h' to report in human readable format.
2840 ------------------------------------------------------------------------------
2842 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2849 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2850 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2851 at startup during orphan unlinking. Instead, skip buddy update
2852 before setup, as it will be regenerated after recovery is complete.
2857 Description: If an OST is inactive, its locks might reference stale inodes.
2858 Details : lov_change_cbdata() must iterate over all namespaces, even if
2859 they are inactive to clear inode references from the lock.
2861 Severity : enhancement
2862 Frequency : occasional, if non-standard max_dirty_mb used
2864 Description: Client will block write RPCs if not enough grant
2865 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2866 then the client will block writes while waiting for another RPC
2867 to complete instead of consuming its dirty limit. With change
2868 we get improved performance when max_dirty_mb is small.
2870 Severity : enhancement
2871 Bugzilla : 3389, 6253
2872 Description: Add support for supplementary groups on the MDS.
2873 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2874 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2875 lookups for user supplementary groups into a cache.
2879 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2880 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2881 for liblustre O_EXCL is enforced by MDS.
2885 Description: readdir never returns NULL in liblustre.
2886 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2887 dentry in struct dirent.
2891 Frequency : liblustre only, depends on application IO pattern
2892 Description: liblustre clients evicted if not contacting servers
2893 Details : Don't put liblustre clients into the ping_evictor list, so
2894 they will not be evicted by the pinger ever.
2896 Severity : enhancement
2898 Description: Add ability to evict clients by NID from MDS.
2899 Details : By echoing "nid:$NID" string into
2900 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2901 $NID would be instantly evicted from this MDS and from all active
2902 OSTs connected to it.
2906 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2907 Details : lookup_it_finish() used to query file size from OSTs that was not
2912 Description: service threads change working directory to that of init
2913 Details : Starting lustre service threads may pin the working directory
2914 of the parent thread, making that filesystem busy. Threads
2915 now change to the working directory of init to avoid this.
2919 Frequency : during shutdown only
2920 Description: shutdown with a failed MDS or OST can cause unmount to hang
2921 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2926 Frequency : 2.6 only
2927 Description: chmod/chown may include an extra supplementary group
2928 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2929 supplementary group and if none is specified this is used.
2932 Bugzilla : 5479 (6816)
2934 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2935 Details : If lookup is in progress on a file that is unlinked we might try
2936 to revalidate the inode and fail in revalidate after lookup is
2937 complete and ll_file_open() enqueues the open again but
2938 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2941 Frequency : always, if lconf --abort_recovery used
2943 Description: lconf --abort_recovery fails with 'Operation not supported'
2944 Details : lconf was attempting to abort recovery on the MDT device and not
2947 ------------------------------------------------------------------------------
2949 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2954 Frequency : rare (only unsupported configurations with a node running as an
2956 Bugzilla : 6514, 5137
2957 Description: Mounting a Lustre file system on a node running as an OST could
2959 Details : OSTs now preallocates memory needed to write out data at
2960 startup, instead of when needed, to avoid having to
2961 allocate memory in possibly low memory situations.
2962 Specifically, if the file system is mounted on on OST,
2963 memory pressure could force it to try to write out data,
2964 which it needed to allocate memory to do. Due to the low
2965 memory, it would be unable to do so and the node would
2966 become unresponsive.
2968 Severity : enhancement
2970 Description: Addition of lconf --service command line option
2971 Details : lconf now accepts a '--service <arg>' option, which is
2972 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2974 Severity : enhancement
2976 Description: Failover mode is now the default for OSTs.
2977 Details : By default, OSTs will now run in failover mode. To return to
2978 the old behaviour, add '--failout' to the lmc line for OSTs.
2980 Severity : enhancement
2982 Description: Health checks are now provided for MDS and OSTs
2983 Details : Additional detailed health check information on MSD and OSTs
2984 is now provided through the procfs health_check value.
2987 Frequency : occasional, depends on IO load
2989 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2990 after numerous create/delete cycles
2991 Details : The ext3 inode allocation policy would not allocate new inodes
2992 very well on the OSTs because there are no new directories
2993 being created. Instead we look for groups with free space if
2994 the parent directories are nearly full.
2999 Description: Network or server problems during mount may cause partially
3000 mounted clients instead of returning an error.
3001 Details : The config llog parsing code may overwrite the error return
3002 code during mount error handling, returning success instead
3008 Description: MDS can fail to allocate large reply buffers
3009 Details : After long uptimes the MDS can fail to allocate large reply
3010 buffers (e.g. zconf client mount config records) due to memory
3011 fragmentation or consumption by the buffer cache. Preallocate
3012 some large reply buffers so that these replies can be sent even
3013 under memory pressure.
3017 Frequency : rare (liblustre)
3018 Description: fsx running with liblustre complained that using truncate() to
3019 extend the file doesn't work. This patch corrects that issue.
3020 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3021 ATTR_SIZE and lsm use in llu_setattr_raw.
3025 Frequency : rare, only 2.6 kernels
3026 Description: Unusual file access patterns on the MDS may result in inode
3027 data being lost in very rare circumstances.
3028 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3029 code in the 2.6 kernel caused the inode and/or EA data not to
3030 be read from disk, causing single-file corruption.
3034 Frequency : rare, only 2.6 filesystems using extents
3035 Description: Heavy concurrent write and delete load may cause data corruption.
3036 Details : It was possible under high-load situations to have an extent
3037 metadata block in the block device cache from a just-unlinked
3038 file overwrite a newly-allocated data block. We now unmap any
3039 metadata buffers that alias just-allocated data blocks.
3043 Frequency : filesystems with default stripe_count larger than 77
3044 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3045 Details : lconf specifies an inode size of 4096 bytes when the default
3046 stripe_count is larger than 77. This conflicts with the default
3047 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3048 this case to avoid pinning too much memory for large EAs.
3050 ------------------------------------------------------------------------------
3052 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3057 Frequency : rare (extremely heavy IO load with hundreds of clients)
3059 Description: Client is evicted, gets IO error writing to file
3060 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3061 caused clients to be evicted for AST timeouts. The fixes in
3062 bug 5192 mean we no longer need to have such short AST timeouts
3063 so ldlm_timeout has been increased.
3066 Frequency : occasional during --force or --failover shutdown under load
3067 Bugzilla : 5949, 4834
3068 Description: Server oops/LBUG if stopped with --force or --failover under load
3069 Details : a collection of import/export refcount and cleanup ordering
3070 issues fixed for safer force cleanup
3073 Frequency : only filesystems larger than 120 OSTs
3074 Bugzilla : 5990, 6223
3075 Description: lfs getstripe would oops on a very large filesystem
3076 Details : lov_getconfig used kfree on vmalloc'd memory
3079 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3080 Bugzilla : 6242, 6243
3081 Description: reading from files that had been truncated to a non-zero size
3082 but never opened returned no data
3083 Details : ll_file_read() reads zeros from no-object files to EOF
3088 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3089 Details : There's little harm in aborting MDS/OSS recovery and letting it
3090 try again, so I removed the LASSERT and return an error instead.
3092 Severity : enhancement
3094 Description: New debugging infrastructure for tracking down data corruption
3095 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3096 (b) cover more of the client-side code path, and (c) try to narrow
3097 down where problems occurred
3101 Bugzilla : 3819, 4364, 4397, 6313
3102 Description: Racing close and eviction MDS could cause assertion in mds_close
3103 Details : It was possible to get multiple mfd references during close and
3104 client eviction, leading to one thread referencing a freed mfd.
3106 Severity: : enhancement
3107 Bugzilla : 3262, 6359
3108 Description: Attempts to reconnect to servers are now more aggressive.
3109 Details : This builds on the enhanced upcall-less recovery that was added
3110 in 1.4.2. When trying to reconnect to servers, clients will
3111 now try each server in the failover group every 10 seconds. By
3112 default, clients would previously try one server every 25 seconds.
3117 Description: After recovery, certain operations trigger a failed
3118 assertion on a client.
3119 Details : Failing over an mds, using lconf -d --failover, while a
3120 client was doing a readdir() call would cause the client to
3121 LBUG after recovery completed and the readdir() was resent.
3123 Severity : enhancement
3125 Description: Default groups are now added by lconf
3126 Details : You can now run lconf --group <servicename> without having to
3127 manually add groups with lmc.
3130 Frequency : occasional
3132 Description: Nodes with an elan id of 0 trigger a failed assertion
3135 Frequency : always when accessing e.g. tty/console device nodes
3137 Description: tty and some other devices nodes cannot be used on lustre
3138 Details : file's private_data field is used by device data and lustre
3139 values in there got lost. New field was added to struct file to
3140 store fs-specific private data.
3143 Frequency : when exporting Lustre via NFS
3145 Description: NFSD failed occasionally when looking up a path component
3146 Details : NFSD is looking up ".." which was broken in ext3 directories
3147 that had grown large enough to become hashed.
3150 Frequency : Clusters with multiple interfaces not on the same subnet
3152 Description: Nodes will repeatedly try to reconnect to an interface which it
3153 cannot reach and report an error to the log.
3154 Details : Extra peer list entries will be created by lconf with some peers
3155 unreachable. lconf now validates the peer before adding it.
3158 Frequency : Only if a default stripe is set on the filesystem root.
3160 Description: Setting a default stripe on the filesystem root prevented the
3161 filesystem from being remounted.
3162 Details : The client was sending extra request flags in the root getattr
3163 request and did not allocate a reply buffer for the dir EA.
3166 Frequency : occasional, higher if lots of files are accessed by one client
3167 Bugzilla : 6159, 6097
3168 Description: Client trips assertion regarding lsm mismatch/magic
3169 Details : While revalidating inodes the VFS looks up inodes with ifind()
3170 and in rare cases can find an inode that is being freed.
3171 The ll_test_inode() code will free the lsm during ifind()
3172 when it finds an existing inode and then the VFS later attaches
3173 this free lsm to a new inode.
3177 Bugzilla : 6422, 7030
3178 Description: MDS deadlock between mkdir and client eviction
3179 Details : Creating a new file via mkdir or mknod (starting a transaction
3180 and getting the ns lock) can deadlock with client eviction
3181 (gets ns lock and trying to finish a synchronous transaction).
3184 Frequency : occasional
3185 Description: While starting a server, the fsfilt_ext3 module could not be
3187 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3188 kernels. Previously, lconf would still use the ext3 name
3189 when trying to load modules. Now, it will correctly use
3190 ext3 on 2.4 and ldiskfs on 2.6.
3192 Severity : enhancement
3193 Description: The default stripe count has been changed to 1
3194 Details : The interpretation of the default stripe count (0, to lfs
3195 or lmc) has been changed to mean striping across a single
3196 OST, rather than all available. For general usage we have
3197 found a stripe count of 1 or 2 works best.
3199 Severity : enhancement
3200 Description: Add support for compiling against Cray portals.
3201 Details : Conditional compiling for some areas that are different
3205 Frequency : occasional
3206 Bugzilla : 6409, 6834
3207 Description: Creating files with an explicit stripe count may lead to
3208 a failed assertion on the MDS
3209 Details : If some OSTs are full or unavailable, creating files may
3210 trigger a failed assertion on the MDS. Now, Lustre will
3211 try to use other servers or return an error to the
3215 Frequency : occasional
3217 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3218 caused lock timeout during readahead (since 1.4.2).
3219 Details : Processes doing readahead might match a lock that hasn't been
3220 granted yet if there are overlapping and conflicting lock
3221 requests. The readahead process waits on ungranted lock
3222 (original lock is CBPENDING), while OST waits for that process
3223 to cancel CBPENDING read lock and eventually evicts client.
3225 Severity : enhancement
3227 Description: Initial enabling of flock support for clients
3228 Details : Implements fcntl advisory locking and file status functions.
3229 This feature is provided as an optional mount flag (default
3230 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3231 locking are implemented yet, and those that are are not guaranteed
3232 to be completely correct in production environments.
3233 mount -t lustre -o [flock|noflock] ...
3236 Frequency : occasional
3238 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3239 assertion in the ext3 JBD (journaling) layer.
3240 Details : The b_committed_data struct is protected by the big kernel lock
3241 in 2.4 kernels, serializing journal_commit_transaction() and
3242 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3243 access to this struct. In 2.6 kernels there is finer grained
3244 locking to improve SMP performance of the JBD layer.
3248 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3253 Description: merge and cleanup kernel patches.
3256 -----------------------------------------------------------------------------
3258 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3260 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3261 versions of Lustre. Please update all servers and clients to
3262 version 1.4.2 or later at the same time. You must also run
3263 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3264 to update the configuration logs.
3266 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3267 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3268 - swab llog records of type '0' so we get proper header size/idx (5861)
3269 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3270 - fix rename of one directory over another leaking an inode (5953)
3271 - avoid SetPageDirty on 2.6 (5981)
3272 - don't re-add just-being-destroyed locks to the waiting list (5653)
3273 - when creating new directories, inherit the parent's custom
3274 striping settings if present parent (3048)
3275 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3276 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3277 - don't hold client locks on temporary worklist from l_lru (5666)
3278 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3279 - automatically evict dead clients (5921)
3280 - Update file size properly in create+truncate+fstat case (6196)
3281 - Do not unhash mountpoint dentries, do not allow removal of
3283 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3284 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3285 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3286 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3287 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3288 - Correctly handle reads of files with no objects (6243)
3289 - lctl recover will also mark a device active if deactivate used (5933)
3291 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3292 - allow --write-conf on an MDS with different nettype than client (5619)
3293 - don't write config llogs to MDS for mounts not from that MDS (5617)
3294 - lconf should create multiple TCP connections from a client (5201)
3295 - init scripts are now turned off by default; run chkconfig --on
3296 lustre and chkconfig --on lustrefs to use them
3297 - upcalls are no longer needed for clients to recover to failover
3299 - add --abort-recovery option to lconf to abort recovery on device
3301 - add support for an arbitrary number of OSTs (3026)
3302 - Quota support protocol changes.
3303 - forward compatibility changes to wire structs (6007)
3304 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3305 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3306 - improved "lustre status" script
3307 - initialize blocksize for non-regular files (6062)
3308 - added --disable-server and --disable-client configure options (5782)
3309 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3310 - Vanilla 2.4.29 support
3311 - increase maximum number of obd devices to 520 (6242)
3312 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3313 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3314 SLES 9 2.6.5 kernels
3315 - limit stripes per file to 160 (the maximum EA size) (6093)
3317 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3320 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3321 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3322 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3323 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3324 - small ext3 extents cleanups and fixes (5733)
3325 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3326 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3327 - avoid needless client->OST connect, fix handle mismatch (5317)
3328 - fix DLM error path that led to out-of-sync client, long delays (5779)
3329 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3330 - fix several locking issues related to i_size (5492,5624,5654,5672)
3331 - don't move pending lock onto export if it is already evicted (5683)
3332 - fix kernel oops when creating .foo in unlinked directory (5548)
3333 - fix deadlock in obdfilter statistics vs. object create (5811)
3334 - use time_{before,after} to avoid timer jiffies wrap (5882)
3335 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3336 - Do not leak request if server was not able to process it (5154)
3337 - If mds_open unable to find parent dir, make that negative lookup(5154)
3338 - don't create new directories with extent-mapping (5909, 5936)
3340 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3341 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3342 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3343 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3344 - support for RHEL4 kernel on i686 (5773)
3345 - provide error messages when incompatible logs are encountered (5898)
3347 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3348 * version 1.4.0.10 (1.4.1 release candidate 1)
3350 - don't keep a lock reference when lock is not granted (4238)
3351 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3352 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3353 - drop import inflight refcount on signal_completed_replay error (5255)
3354 - unlock page after async write error during send (3677)
3355 - handle missing objects in filter_preprw_read properly (5265)
3356 - no transno return for symlink open, don't save no-trasno open (3440)
3357 - don't try to complete elan receive that already failed (4012)
3358 - free RPC server reply state on error (5406)
3359 - clean up thread from ptlrpc_start_thread() on error (5160)
3360 - readahead could read extra page into cache that wasn't ejected (5388)
3361 - prevent races in class_attach/setup/cleanup/detach (5260)
3362 - don't dereference de->d_inode after l_dput of de (5458)
3363 - use "int" for stripe value returned from lock_to_stripe (5544)
3364 - mballoc allocation and error-checking fixes in 2.6 (5504)
3365 - block device patches to fix I/O request sizes in 2.6 (5482)
3366 - look up hostnames for IB nals (5602)
3367 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3368 - don't start multiple acceptors for the same port (5277)
3369 - fix incorrect LASSERT in mds_getattr_name (5635)
3370 - export a proc file for general "ping" checking (5628)
3371 - fix "lfs check" to not block when the MDS is down (5628)
3373 - service request history (4965)
3374 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3375 - create an "evict_client" /proc entry on OSTs, like the MDS has
3376 - fix mount usage message, return errors per mount(8) (5168)
3377 - change grep [] to grep "[]" in tests so they work in more UMLs
3378 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3379 - remove ancient LOV_MAGIC_V0 EA support (5047)
3380 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3381 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3382 - allow readahead allocations to fail when low on memory (5383)
3383 - mmap locking landed again, after considerable improvement (2828)
3384 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3386 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3389 - send OST transaction number in read/write reply to free req (4966)
3390 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3391 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3392 - the watchdog thread now runs as interruptible (5246)
3393 - flock/lockf fixes (but it's still disabled, pending 5135)
3394 - don't use EXT3 constants in llite code (5094)
3395 - memory shortage at startup could cause assertion (5176)
3397 - reorganization of lov code
3398 - single portals codebase
3400 - add extents/mballoc support (5025)
3401 - direct I/O reads in the obdfilter (4048)
3402 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3404 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3407 - send OST transaction number in read/write reply to free req (4966)
3408 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3409 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3410 - the watchdog thread now runs as interruptible (5246)
3411 - handle missing objects in filter_preprw_read properly (5265)
3412 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3413 - drop import inflight refcount on signal_completed_replay error (5255)
3414 - unlock page after async write error during send (3677)
3415 - return original error code on reconstructed replies (3761)
3416 - no transno return for symlink open, don't save no-trasno open (3440)
3418 - add pid to ldlm debugging output (4922)
3419 - bump the watchdog timeouts -- we can't handle 30sec yet
3420 - extra debugging for orphan dentry/inode bug (5259)
3422 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3425 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3426 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3427 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3428 - fix NULL dereference in /proc/sys/portals/routes (4827)
3429 - allow failed mdc_close() operations to be interrupted (4561)
3430 - stop precreate on OST before MDS would time out on it (4778)
3431 - don't send partial-page writes before EOF from client (4410)
3432 - discard client grant for sub-page writes on large-page clients (4520)
3433 - don't free dentries not owned by NFS code, check generation (4806)
3434 - fix lsm leak if mds_create_objects() fails (4801)
3435 - limit debug_daemon file size, always print CERROR messages (4789)
3436 - use transno after validating reply (3892)
3437 - process timed out requests if import state changes (3754)
3438 - update mtime on OST during writes, return in glimpse (4829)
3439 - add mkfsoptions to LDAP (4679)
3440 - use ->max_readahead method instead of zapping global ra (5039)
3441 - don't interrupt __l_wait_event() during strace
3443 - add software watchdogs to catch hung threads quickly (4941)
3444 - make lustrefs init script start after nfs is mounted
3445 - fix CWARN/ERROR duplication (4930)
3446 - return async write errors to application if possible (2248)
3447 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3448 - print NAL number in %x format (4645)
3449 - update barely-supported suse-2.4.21-171 series (4842)
3450 - support for sles 9 %post scripts
3451 - support for building 2.6 kernel-source packages
3452 - support for sles km_* packages
3454 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3457 - ignore -ENOENT errors in osc_destroy (3639)
3458 - notify osc create thread that OSC is being cleaned up (4600)
3459 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3460 - reconstruct ost_handle() like mds_handle() (4657)
3461 - create a new thread to do import eviction to avoid deadlock (3969)
3462 - let lconf resolve symlinked-to devices (4629)
3463 - don't unlink "objects" from directory with default EA (4554)
3464 - hold socknal file ref over connect in case target is down (4394)
3465 - allow more than 32000 subdirectories in a single directory (3244)
3466 - fix blocks count for O_DIRECT writes (3751)
3467 - OST returns ENOSPC from object create when no space left (4539)
3468 - don't send truncate RPC if file size isn't changing (4410)
3469 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3470 - bind to privileged port in socknal and tcpnal (3689)
3472 - rate limit CERROR/CWARN console message to avoid overload (4519)
3473 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3474 - basic mmap support (3918)
3475 - kernel patch series update from b1_4 (4711)
3477 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3480 - avoid crash during MDS cleanup with OST shut down (2775)
3481 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3482 - don't use bad inodes on the MDS (3744)
3483 - dynamic object preallocation to improve recovery speed (4236)
3484 - don't hold spinlock over lock dumping or change debug flags (4401)
3485 - don't zero obd_dev when it is force cleaned (3651)
3486 - print grants to console if they go negative (4431)
3487 - "lctl deactivate" will stop automatic recovery attempts (3406)
3488 - look for existing locks in ldlm_handle_enqueue() (3764)
3489 - don't resolve lock handle twice in recovery avoiding race (4401)
3490 - revalidate should check working dir is a directory (4134)
3492 - don't always mark "slow" obdfilter messages as errors (4418)
3494 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3497 - don't close LustreDB during write_conf until it is done (3860)
3498 - fix typo in lconf for_each_profile (3821)
3499 - allow dumping logs from multiple threads at one time (3820)
3500 - don't allow multiple threads in OSC recovery (3812)
3501 - fix debug_size parameters (3864)
3502 - fix mds_postrecov to initialize import for llog ctxt (3121)
3503 - replace config semaphore with spinlock (3306)
3504 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3505 - don't allow enqueue to complete on a destroyed export (3822)
3506 - down write_lock before checking llog header bitmap (3825)
3507 - recover from lock replay timeout (3764)
3508 - up llog sem before sending rpc (3652)
3509 - reduce ns lock hold times when setting kms (3267)
3510 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3511 - fix NULL deref and obd_dev leak on setup error (3312)
3512 - replace some LBUG about llog ops with error handling (3841)
3513 - don't match INVALID dentries from d_lookup and spin (3784)
3514 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3515 - fix invalid assertion in ptlrpc_set_wait (3880)
3517 - add libwrap support for the TCP acceptor (3996)
3518 - add /proc/sys/portals/routes for non-root route listing (3994)
3519 - allow setting MDS UUID in .xml (2580)
3520 - print the stack of a process that LBUGs (4228)
3522 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3525 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3526 - make sure to unset replay flag from failed open requests (3440)
3527 - if default stripe count is 0, use OST count for inode size (3636)
3528 - update parent mtime/ctime on client for create/unlink (2611)
3529 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3530 - free recovery state on server during a forced cleanup (3571)
3531 - unregister_reply for resent reqs (3063)
3532 - loop back devices mounting and status check on 2.6 (3563)
3533 - fix resource-creation race that can provoke i_size == 0 (3513)
3534 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3535 - more debugging for page-accounting assertion (3746)
3536 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3537 - avoid deadlock after precreation failure (3758)
3538 - fix race and lock order deadlock in orphan handling (3450, 3750)
3539 - add validity checks when grabbing inodes from l_ast_data (3599)
3541 - add /proc/.../recovery_status to obdfilter (3428)
3542 - lightweight CDEBUG infrastructure, debug daemon (3668)
3543 - change default OSC RPC parameters to be better on small clusters
3544 - turn off OST read cache for files smaller than 32MB
3545 - install man pages and include them in rpms (3100)
3546 - add new init script for (un)mounting lustre filesystems (2593)
3547 - run chkconfig in %post for init scripts (3701)
3548 - drop scimac NAL (unmaintained)
3550 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3553 - clean kiobufs before and after use (3485)
3554 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3555 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3556 - change init script to fail more clearly if not run as root (1528)
3557 - allow clients to reconnect during replay (1742)
3558 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3559 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3560 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3561 - fix cond_resched (3554)
3562 - only evict unfinished clients after recovery (3515)
3563 - allow bulk resend, prevent data loss (3570)
3564 - dynamic ptlrpc request buffer allocation (2102)
3565 - don't allow unlinking open directory if it isn't empty (2904)
3566 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3567 - remove extraneous obd dereference causing LASSERT failure (3334)
3568 - don't use get_cycles() when creating temp. files on the mds (3156)
3569 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3570 - handle EEXIST for set-stripe, set proper directory name (3336)
3572 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3573 - fix ksocknal_fmb_callback() error messages (2918)
3575 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3578 - don't copy lvb into (possibly NULL) reply on error (2983)
3579 - don't deref dentry after dput, don't free lvb on error (2922)
3580 - use the kms to determine writeback rpc length (2947)
3581 - increment oti_logcookies when osc is inactive (2948)
3582 - update client's i_blocks count via lvb messages (2543)
3583 - handle intent open/close of special files properly (1557)
3584 - mount MDS with errors=remount-ro, like obdfilter (2009)
3585 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3586 - don't use cancelling-locks' kms values (2947)
3587 - use highest lock extent for kms, not last one (2925)
3588 - don't dereference ERR_PTR() dentry in error handling path (3107)
3589 - fix thread race in portals_debug_dumplog() (3122)
3590 - create lprocfs device entries at setup instead of at attach (1519)
3591 - common AST error handler, don't evict client on completion race (3145)
3592 - zero nameidata in detach_mnt in 2.6 (3118)
3593 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3594 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3595 - initialize RPC timeout timer earlier for 2.6 (3219)
3596 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3597 - print nal/nid for unknown nid (3258)
3598 - additional checks for oscc recovery before doing precreate (3284)
3599 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3600 - don't crash in mdc_close for bad permissions on open (3285)
3601 - zero i_rdev for non-device files (3147)
3602 - clear page->private before handing to FS, better assertion (3119)
3603 - tune the read pipeline (3236)
3604 - fix incorrect decref of invalidated dentry (2350)
3605 - provide read-ahead stats and refine rpc in flight stats (3328)
3606 - don't hold journal transaction open across create RPC (3313)
3607 - update atime on MDS at close time (3265)
3608 - close LDAP connection when recovering to avoid server load (3315)
3609 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3610 - don't leak open file on MDS after open resend (3325)
3611 - serialize filter_precreate and filter_destroy_precreated (3329)
3612 - loop device shouldn't call sync_dev() for nul device (3092)
3613 - clear page cache after eviction (2766)
3614 - resynchronize MDS->OST in background (2824)
3615 - refuse to mount the same filesystem twice on same mountpoint (3394)
3616 - allow llmount to create routes for mounting behind routers (3320)
3617 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3618 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3619 - fix rare problem with rename on htree directories (3417)
3621 - allow default OST striping configuration per directory (1414)
3622 - fix compilation for qswnal for 2.6 kernels (3125)
3623 - increase maximum number of MDS request buffers for large systems
3624 - change liblustreapi to be useful for external progs like lfsck (3098)
3625 - increase local configuration timeout for slow disks (3353)
3626 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3628 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3631 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3632 - don't overwrite extent policy data in reply if lock was blocked (2901)
3633 - drop filter export grants atomically with removal from device (2663)
3634 - del obd_self_export from work_list in class_disconnect_exports (2908)
3635 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3636 - swab reply message in mdc_close, other PPC fixes (2464)
3637 - fix destroying of named logs (2325)
3638 - overwrite old logs when running lconf --write_conf (2264)
3639 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3640 - fix race in target_handle_connect (2898)
3641 - mds_reint_create() should take same inode create lock (2926)
3642 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3643 - don't close files for self_export to avoid uninitialized obd (2936)
3644 - allow MDS with the same name as client node (2939)
3645 - hold dentry reference for closed log files for unlink (2325)
3646 - reserve space for all logs during transactions (2059)
3647 - don't evict page beyond end of stripe extent (2925)
3648 - don't oops on a deleted current working directory (2399)
3649 - handle hard links to targets without a parent properly (2517)
3650 - don't dereference NULL lock when racing during eviction (2867)
3651 - don't grow lock extents when lots of conflicting locks (2919)
3653 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3656 - account for cache space usage on clients to avoid data loss (974)
3657 - lfsck support in lustre kernel code (2349)
3658 - reduce journal credits needed for BRW writes (2370)
3659 - orphan handling to avoid losing space on client/server crashes
3660 - ptlrpcd can be blocked, stopping ALL progress (2477)
3661 - use lock value blocks to assist in proper KMS, faster stat (1021)
3662 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3663 - recovery for initial connections (2355)
3664 - fixes for mds_cleanup_orphans (1934)
3665 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3666 - block all file creations until orphan recovery completes (1901)
3667 - client remove rq_connection from request struct (2423)
3668 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3669 - recovery timer race (2670)
3670 - mdc_close recovey bug (2532)
3671 - ptlrpc cleanup bug (2710)
3672 - mds timeout on local locks (2588)
3673 - namespace lock held during RPCs (2431)
3674 - handle interrupted sync write properly (2503)
3675 - don't try to handle a message that hasn't been replied to (2699)
3676 - client assert failure during cleanup after abort recovery (2701)
3677 - leak mdc device after failed mount (2712)
3678 - ptlrpc_check_set allows timedout requests to complete (2714)
3679 - wait for inflight reqs when ptlrpcd finishes (2710)
3680 - make sure unregistered services are removed from the srv_list
3681 - reset bulk XID's when resending them (caught by 1138 test)
3682 - unregister_bulk after timeout
3683 - fix lconf error (2694)
3684 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3685 - readahead locks pages, leaves pending causing memory pressure (2673)
3686 - increase OST request buffers to 4096 on large machines (2729)
3687 - fix up permission of existing directories in simple_mkdir (2661)
3688 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3689 - don't assign transno to errored transactions (2742)
3690 - don't delete objects on OST if given a bogus objid from MDS (2751)
3691 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3692 - if rq_no_resend, then timeout request after recovery (2432)
3693 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3694 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3695 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3696 - MDS recovery shouldn't delete objects that it has given out (2730)
3697 - if enqueue arrives after completion, don't clobber LVB (2819)
3698 - don't unlock pages twice when trigger_group_io returns error (2814)
3699 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3700 - don't write pages to disk if there was an error (1450)
3701 - don't ping imports that have recovery disabled (2676)
3702 - take buffered bytes into account when balancing socknal conn (2817)
3703 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3704 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3705 - remove little-endian swabbing of llog records (1987)
3706 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3707 - retry reposting request buffers if they fail (1191)
3708 - grow extent at grant time to avoid granting a revoked lock (2809)
3709 - lock revoke doesn't evict page if covered by a second lock (2765)
3710 - disable VM readahead to avoid reading outside lock extents (2805)
3712 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3713 - updated kernel patches for hp-2.4.20 kernel (2681)
3715 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3718 - fix truncated write corruption (2366)
3719 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3721 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3722 - ensure len(uuid) < 37 in lmc (1171)
3723 - fix ia64 OOPS in llog_test (2255)
3724 - zero end of page at obdfilter for partial page writes (2648)
3725 - don't leave stale dentries around after renames (bug 2428)
3726 - fix timeouts when evicting a client with a single lock held (2642)
3727 - set deadline for the initial HELLO message to drain (2634)
3728 - print out dotted-quad IP addresses in the socknal (2302)
3730 - additional debugging for MDS client eviction problem (2443)
3731 - fix mkfsoptions support for osts (2603, 2604)
3733 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3736 - add series for the vanilla 2.6.0 kernel
3737 - add series for the vanilla 2.4.24 kernel
3738 - add series for a cray x86/64 UL kernel drop
3739 - fix xattr patches for the vanilla 2.4.19 series
3741 - generate true UUIDs in lmc (1171)
3742 - have portals stack dumping break in UML (2466)
3743 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3744 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3745 - wake callback waiting threads on client eviction (2460)
3746 - Add --ptldebug and --subsystem to lmc (1719)
3747 - update assertion to allow safe interrupt allocation
3748 - set rq_no_resend for cancel requests (2432)
3749 - recalculate ptlrpcd timeout after resend (2494)
3750 - call vfs_rmdir when removing pending directories (2368)
3751 - fix renaming a file to itself (2429)
3752 - lmc creates a default one-stripe lov (2454)
3753 - expand procfs space to handle large clusters (2326)
3754 - increase UML stack to avoid overflow
3755 - update lconf's list of debug and subsystem masks
3756 - fix lfs find --obd (2510)
3757 - /proc tunable for disabling filter read caching (2591)
3758 - stop rpm packages from altering slapd.conf (2301)
3759 - disable nagle in the socknal under 0conf (2578)
3760 - choose mds inode size based on stripe count (2572)
3761 - fix kernel-source rpm problems (2516)
3763 - add --disable-doc to avoid pdf generation (2421)
3764 - update documentation, tests, type-os, comments
3765 - avoid format warnings on ia64
3766 - remove the TOE NAL
3767 - tiny code cleanups by removing unused fields
3769 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3772 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3773 - remove the most blatant lies from BUILDING file (2371)
3774 - change default debug level to reasonable production setting
3775 - reduce client side cache size to reduce cache flush time
3776 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3777 - make TCP zerocopy and pinger support enabled by default (2476)
3778 - sync writes completed after process exits caused crashes (2319)
3779 - maintain correct mount count on the MDS (2356)
3780 - backout 1557, because 2316 wasn't really fixed
3781 - better file I/O statistics gathering in /proc
3782 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3783 - another kernel patch to fix zero-copy TCP function export
3784 - don't take duplicate lock when processing re-sent getattr (2420)
3785 - lctl uses obd_self_export instead of creating new conn (2353)
3786 - MDS/OST recovery case which requires object creation asserted (2425)
3787 - move lfs from /usr/sbin to /usr/bin in packages
3788 - fix race between mds_client_add and mds_client_free (2417)
3789 - use kmalloc instead of slabs in portals (2430)
3790 - don't create duplicate records when a failover MDS is present (2442)
3791 - remove unnecessary mount age check (2332)
3792 - don't remove directory inodes from locks prematurely (2451)
3793 - don't break if MDS service name is the same as hostname (2103)
3794 - fix races in client write RPC generation when cache full (2482)
3796 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3799 - remove now-unused request->rq_obd (278)
3800 - if an allocation fails, print out how much memory we've used (1933)
3801 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3802 - add the "configurable stack size" patch to most series files (1256)
3803 - ability to write large log records, for 100+ OST configs (2306)
3804 - fix NULL deref when filter_prep fails (2314)
3805 - fix operator precedence error in filter_sync
3806 - dynamic allocation of socknal TX descriptors (2315)
3807 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3808 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3809 - ensure that CWARN messages in Portals always get to the syslog
3810 - __init/__exit are not for prototype decls (ldlm_init/exit)
3811 - x86-64 compile warning fixes
3812 - fix gateway LMC keyword conflict (2318)
3813 - fix MDS lock inversions in getattr/reint paths (1844)
3814 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3815 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3816 - DLM race condition prevented some lock evictions (2328)
3817 - ENOMEM detection and retry on socknal sends (2230)
3818 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3819 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3820 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3821 - better error messages when a client is rejected during recovery (1505)
3822 - avoid cancelling locks which were never granted, after failure (2330)
3823 - fix i_sem/journal inversion in mds_client_add (2333)
3824 - fix truncate/getattr lock cycle deadlock (2334)
3825 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3826 - fix two rare exit paths which could leak an l_lock() ref (2321)
3827 - fencepost error in MDS/OST orphan recovery (2226)
3828 - make log record alignment 8 bytes (1988)
3829 - lstripe now fails when requested offset > ost_count (2237)
3830 - ensure that all kernel series have a complete list.h (1607)
3831 - fix crashes in special-file operations (2316)
3832 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3834 - allow configurable automake binary, for testing new versions
3835 - small update to the lfs documentation
3837 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3839 * fix negative export reference count in fsfilt_sync (2312)
3841 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3842 * release candidate 0.9.1
3844 - orphans are moved into the PENDING directory for possible recovery
3845 - replayed opens now open by fid for orphan/rename safety (1042)
3846 - last close of an orphan inode generates a transno (683)
3847 - chdir() and mount() now pin the directory entry (1020)
3848 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3849 - discard very old requests without processing them (1502)
3850 - remove some common, well-understood CERRORs (1505)
3851 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3852 - clear "grant" flags in OST replies until OST grant code lands (1644)
3853 - fix read performance by not clobbering i_blksize on client (1598)
3854 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3855 - make LRU size a /proc tunable, clears locks when reduced (707)
3856 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3857 - server locks take references on exports now (1558)
3858 - build fixes for 2.4.20-rh trees (1663)
3859 - return an error from lov_create if all OSCs are inactive (1751)
3860 - fix import levels when a reconnect happens without a timeout (1597)
3861 - exit early from mds_open if we get a lookup error (1749)
3862 - partial page read at EOF wouldn't wait for disk before sending (1642)
3863 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3864 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3865 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3866 - write/truncate lock inversion (1639)
3867 - Don't auto-load obdclass, portals modules during cleanup (1495)
3868 - fix timestamps from jumping to "now" (1763)
3869 - extra journal assertions (1648)
3870 - add an extra multiunlink test (1771)
3871 - fix read_record/write_record API (1776)
3872 - fix leak of offset_extent, possible incorrect i_size later (1772)
3873 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3874 - Debugging for the kqswnal_get_idle_tx problems (1820)
3875 - Allow recovery to be attempted multiple times (1536)
3876 - Write out MDS last_rcvd file after it is first created (1600)
3877 - Fix tx_descriptor leak in failed transmit situations (1827)
3878 - ext3 journaling fixes for assertion failure after IO error (1871)
3879 - class_export_put() on freed export after completion AST error (1896)
3880 - Fix revalidate looping in VFS (1322)
3881 - Don't access a freed export during MDS_REINT timeout (1521)
3882 - Add open-unlink recovery support on the MDS (1673,1764)
3883 - Return an error if no MDS data was read from last_rcvd (1946)
3884 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3885 - Don't setup a disk device that doesn't match exported UUID (317)
3886 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3887 - avoid committing NULL handle in force close
3888 - local.sh is now a one-stripe LOV configuration
3889 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3890 - fix ext3 htree duplicate directory entry corruption (1516)
3891 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3892 - update to new LOV EA format (2097)
3893 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3895 2003-06-15 Phil Schwan <phil@clusterfs.com>
3898 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3899 - per-import/export recovery handling (958, 931, 959)
3900 - multiple last-rcvd slots, for serving multiple FSes (949)
3901 - connections are again shared between multiple imp/exports (963, 964)
3902 - "umount -f" would hang if any requests needed to be sent (393, 978)
3903 - avoid pinning large req buffer by copying for queued messages (989)
3904 - add "uuid" to "lctl device" command to help upcalls (991)
3905 - "open" RPCs with transnos would confuse recovery counters (1037)
3906 - do proper endian conversion of all wire messages (288, 340, 891)
3907 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3908 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3909 - fix ptlrpc_connection leak in target_handle_connect (1174)
3910 - fix import refcounting bug in OST and MDS cleanup (1134)
3911 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3912 - fix very unlikely obd_types race condition (501)
3913 - remove osc_open hack for echo_client (1187)
3914 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3915 - a failure in read_inode2 leads to deadlock (1139)
3916 - cancel ack-locks as soon as transaction is committed (1072)
3917 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3918 - make sure to commitrw after any preprw to avoid deadlock (1162)
3919 - failing to execute a file in a lustre FS would lock inode (1203)
3920 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3921 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3922 - obd_uuid2tgt would walk off the end of the list (1255)
3923 - on IA64 the osc would give portals incorrect bulk size (1258)
3924 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3925 - fix lock inversion caused by new llite matching code (1282)
3926 - limit the number of dirty pages on a client to 10MB (1286)
3927 - timed out locks were not being corrected cancelled (1289)
3928 - fix O_DIRECT above 4GB on IA-32 (1292)
3929 * major user-visible changes
3930 - fail out/fail over policy now controlled by the upcall (993)
3932 - add OBD_PING to check server availability and failure (954)
3933 - lustre messages are now sent in sending host order (288, 340, 891)
3934 - add eadatalen to MDS getattr reply (340)
3935 - OST read replies may contain second buffer, with per-page status (593)
3937 2003-03-11 Phil Schwan <phil@clusterfs.com>
3940 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3941 - failed open()s could cause deadlock; fixed (867, 869)
3942 - stop cancelling OST locks when files are closed (481)
3943 - overlapping XID spaces caused network corruption (851, 853)
3944 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3945 - setattr_raw added, to do single-RPC, server-side setattrs
3946 - lmc/lconf syntax change for OST UUIDs
3947 - fix crashy race condition between ptlrpc_free_req and osc_close
3948 - don't use request in mdc_enqueue if we hit a timeout (889)
3949 - don't set the inode i_size for regular files from the MDS (896)
3950 - handle out of order completion AST (842)
3951 - don't LBUG if a lock request times out after receiving AST (913)
3952 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3953 - if a bad lock AST arrives, send an error instead of dropping entirely
3954 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3955 - fix leak in bulk IO when only partially completed (899, 900, 926)
3956 - fix O_DIRECT for ia64 (55)
3957 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3958 - C-z after timeout could hang a process forever; fixed (977)
3960 - client-side I/O cache (678, 924, 929, 941, 970)
3962 - READPAGE and SETATTRs which don't take server-side locks get
3965 2003-02-11 Phil Schwan <phil@clusterfs.com>
3968 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3969 - Fix fsfilt_extN_readpage() to read a full page of directory
3970 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3971 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3972 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3973 - Most intent operations were removed, in favour of a new RPC mode
3974 that does a single RPC to the server and bypasses most of the VFS
3975 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3976 - Aggressively cancel local locks on DLM servers
3977 - mds_reint_unlink sends EA to the client if it's the last nlink.
3978 client uses that EA to unlink OST objects.
3979 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3980 - recursive symlinks were fixed (439)
3981 - fixed NULL deref in DEBUG_REQ
3982 - filter_update_lastobjid no longer calls sync, which annoyed extN
3983 - fixed multi-client small-writes to a single file problem (445)
3984 - fixed mtime updates during file writes (607)
3985 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3986 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3987 - fixed Linux OST size reporting problem (444, 656)
3988 - OST now updates object mtime with writes or setattr (607, 619)
3989 - client verifies file size before zeroing page past EOF (445)
3990 - OST now writes last allocated objid to disk with allocation (108)
3991 - LOV on echo now works (409)
3993 - mds_reint_unlink sends a new buffer, with the EA included. this
3994 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3995 set if a regular file was being unlinked, and it was the last link
3996 - use PtlGet from the target for bulk writes (315)
3997 - OST now updates object mtime with writes or setattr (607, 619)
3998 - LDLM now has a grant-time callback to revalidate locked items, if
4000 - Many MDS operations were reorganized to combat race conditions
4002 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4003 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4005 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4008 - Fully reactivate OST imports after reconnection (512, others)
4009 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4010 - More graceful error handling for truncating on dead OST (515)
4011 - Don't error out unless we're actually accessing dead stripes (474)
4012 - Fix garbage sizes when stripes are missing (410)
4013 - LRU counters were broken, causing constant lock purge (433, 432)
4014 - garbage on read from stripes with failed OSTs (441)
4015 - mark OSCs as active before reconnecting during recovery (438)
4016 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4017 - lfind did not preserve OST order in output (443)
4018 - symlinks cause hung clients, incorrect data (439)
4019 - stop dereferencing request after dropping refcount (457)
4020 - don't LASSERT(spin_is_locked) on non-SMP (455)
4021 - fixes for many rename() bugs
4022 - fstat didn't correctly synchronize attributes (399)
4023 - server must handle lock cancellation during blocking AST prep (487)
4024 - bulk descriptors were free()d too soon (511)
4025 - fix paths in lconf, which would load incorrect modules (451, 507)
4026 - fix confusing lconf 'host not found' error message (386)
4027 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4028 - fix race condition in mdc_blocking_ast() for inode access (526)
4029 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4030 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4031 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4032 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4033 - fix lstripe in several ways (kernel side) (527)
4034 - fix request leak in ldlm_cli_enqueue (262)
4035 - incorrect OSC was marked inactive after OST failure
4036 - call mds_fs_cleanup before unmounting filesystem (524)
4037 - fix races between taking ns_lock and ldlm_lock_change_resource
4038 - fix races updating LOV export open file list
4039 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4040 - fix recovery NULL deref in ldlm_cli_cancel_unused
4041 - fix some DLM races by using new hash table for lock handles (419)
4042 - permit the client to specify desired inodes, at replay
4043 - duplicate requests when we queue them for replay reintegration
4044 - fix last_rcvd offset calculation
4045 - sync after each recovered transaction, so we always make progress
4046 - never, not always, ERESTART requests without transnos
4047 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4049 - skip replay if the MDS says that the client is already connected
4050 - don't check for a recovery-enabled export to match lctl's UUID
4051 - don't INC_USE_COUNT for phantom exports
4052 - don't crash when cleaning up phantom exports (567)
4053 - don't double-finish or set replay data for errored mdc_open requests
4054 - abort requests when they time out, so we don't get old replies
4055 - send/receive replies for AST messages again
4056 - if the client says that it doesn't have the lock, cancel it on the
4058 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4059 mark it as destroyed, it will all work out when decref is called
4060 - fix module use counts (22, 581)
4062 - ASTs now expect a reply (server cancels lock on error reply)
4064 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4067 - fix many simultaneous client startup (392)
4068 - fix dentry->d_it clobbering
4069 - credentials weren't being shipped for readdir/getattr operations
4070 - remove invalid assertions triggered during some concurrent MD
4072 - proper Lustre versions added (336, 389)
4073 - fix memory leak for create error case (398)
4074 - fix LOV locking bug that would get cli/srv out of sync
4075 - fix echo client over LOV (409)
4076 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4077 - fix double-O_EXCL intent crash (424)
4078 - avoid sending multiple lock CANCELs (352)
4080 - MDS can do multi-client recovery (modulo bugs in new code)
4082 - many updates, edits, cleanups
4084 2002-11-18 Phil Schwan <phil@clusterfs.com>
4087 - fix null d_it dereference (346)
4088 - fix full OST/dbench hang (333)
4089 - fix permission problem with file removal (286)
4090 - fix removal of OSCs from LOV when they fail
4091 - fix NULL deref during bulk timeout (214)
4092 - fix problems related to multiple filesystems on one MDS (241)
4093 - fixed serious subtle metadata locking bugs
4094 - free locks on clients when inodes are removed due to memory
4096 - fix inode pointer in lock data (285)
4097 - partial support for multiple MDS on a single host (241)
4098 - data locks weren't cancelled at clear_inode time (290, 311)
4099 - intent locks could lead to unbounded lock growth (205)
4100 - added a maximum lock count, an LRU list, and a flusher
4101 - fix multiple rename (365)
4102 - properly abstracted the echo client
4103 - OSC locked 1 byte too many; fixed
4104 - rewrote brw callback code:
4105 - fixed recovery bugs related to LOVs (306)
4106 - fixed too-many-pages-in-one-write crash (191)
4107 - fixed (again) crash in sync_io_timeout (214)
4108 - probably fixed callback-related race (385)
4110 - Add capability to MDS protocol
4111 - LDLM cancellations and callbacks on different portals
4113 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4116 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4117 - defer open object destruction to close time (601981, 138)
4118 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4119 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4120 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4121 - don't call obd_setattr in ll_file_release for destroyed objects
4122 * protocol change to lustre_msg: move |version| and add |flags|
4123 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4124 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4125 * added replay of create, unlink, link and rename operations during
4126 MDS failover; recovery should be much more robust now
4127 * remove failed OSCs from LOVs (only lov_create uses this so far)
4128 * the lustre-HOWTO was brought (more) up to date (582544)
4130 2002-10-23 Phil Schwan <phil@clusterfs.com>
4133 - in-use dentries weren't being reused properly (617851)
4134 - prevent multiple LDLM setup (599178)
4135 - fix LOV size calculations for truncate (617853)
4136 - fix client handling of MDS intent errors (POSIX)
4137 - fix permission bug in lovstripe.c test (624321)
4138 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4139 - truncate past end of file could corrupt data
4140 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4141 - a race in recovery could return ETIMEDOUT to apps (623947)
4142 - building outside the source directory was fixed
4143 * the lustre-HOWTO was brought (more) up to date (582544)
4144 * major progress was made on recovery functionality
4146 2002-10-10 Phil Schwan <phil@clusterfs.com>
4149 - recovery deadlock fix
4150 - rm -rf causes LBUG fix (617817)
4151 - file open by multiple tasks fix (618962)
4152 - directory permissions bugs (602707 and 620007)
4153 - journal_stop fixed with locking (611313)
4154 - O_APPEND failures resolved (618273, perhaps 614459)
4155 - lconf PATH fix (619770)
4156 - IA64 build fix (621450)
4157 - RPC buffer sizes scale with amount of memory
4159 2002-10-01 Phil Schwan <phil@clusterfs.com>
4162 - locks would be cancelled without throwing away data pages,
4163 resulting in inconsistent data (605627)
4164 - inode attributes were not always being refreshed (605627, 612449)
4165 - lconf now continues to cleanup after lctl reports an error
4166 - MDS now enforces user permissions (602707)
4167 - lprocfs cleanup fixed, but not yet enabled (614157)
4168 - fixed infinite server hang, should a client not respond to an AST
4169 - avoid going into recovery if user calls readlink() with a buffer
4170 that's too small (613941)
4171 - AST RPCs no longer require replies (614867) -- this may be changed
4172 - don't crash server if client sends an IOV that's too big (611336)
4173 - fixed lock conversion deadlock (611892)
4174 - fixed the following of symlinks (614622)
4175 * recovery: the server can remove locks from a client that dies, other
4176 clients can make progress
4177 * more extN patch fixes
4178 * compile-time configurable ptlrpc buffer allocations
4180 - collaborative read cache document
4181 - Lustre Lite Performance CDR document-in-progress
4183 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4186 - fix typo in patch-2.4.18
4188 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4191 - clear ptlrpc request each time in handle_incoming_request()
4192 - unlink of files now destroys the object on the OST
4194 2002-09-19 Peter Braam <braam@clusterfs.com>
4196 * add hard link support
4197 * change obdfile creation method
4198 * kernel patch changed
4200 2002-09-19 Peter Braam <braam@clusterfs.com>
4203 - stack overflow bug in extN fixed
4205 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4207 * documentation updates
4208 - add man pages for config tools
4209 - update tests/README to describe testing with new config tools
4210 - finish metadata API descriptions
4211 * bug fixes and cleanups
4212 - statfs workaround for 16TB limit
4213 - LOV stripe allocation improved, can stripe on subset of OSTs
4214 - LOV file size/IO offset was wrong for files > 4GB in size
4215 - object EA data was being dropped, caused files to be unreadable
4216 - memory overflow with non-LOV OST caused memory corruption
4217 - fixed regression tests to work with new config tools, obdfilter
4218 - fixed bug when directory size became larger than 1 block
4219 - fixed bug (for single client case) when PWD was deleted
4220 - invalidate local directory pages when doing intent-based ops
4221 - avoid LDLM oops when lock callback contained bad data
4223 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4225 * documentation updates
4226 * bug fixes and cleanups
4227 - configuration tools
4230 - 64-bit compile warnings
4231 - 64-bit internal statfs data
4233 * test_brw on persistent OST devices
4235 * lprocfs (disabled)
4237 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4239 * documentation updates
4240 * bug fixes and cleanups
4241 * configuration tools
4243 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4246 * many small fixes to 0_5_4
4247 * io/network handling
4248 * thinkos in MDS operations
4250 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4253 * crucial basic fixes to 0.5.3
4254 * IOR, Iozone work over Elan
4257 2002-08-07 Phil Schwan <phil@clusterfs.com>
4258 * version 0_5_3, our first alpha
4259 * we use the new Portals iovs
4260 * documentation updates
4261 * bug fixes and cleanups
4262 * small changes in the DLM wire protocol
4264 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4265 * version 0_5_1 with some initial stability,
4266 * locking on MD and file I/O.
4267 * documentation updates
4268 * several bug fixes since 0.5.0
4269 * small changes in wire protocol
4271 2002-07-18 Phil Schwan <phil@clusterfs.com>
4273 * delivered as Lustre Light Alpha
4274 * fixed a crash after handling invalid MDS requests
4275 * fixed directory pages for architectures with non-4k pages sizes
4277 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4278 * release version v0_4_4
4279 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4281 * Fixes a number of interruption problems with OST operations.
4282 * Update documentation for portals header changes
4283 * Move all wire protocol structs/defines to lustre_idl.h
4284 * Fixes symlink length bug.
4285 * Add tcpdump to repository.
4287 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4288 * release version v0_4_3
4289 * Fixes statfs for inodes on extN.
4290 * Fixes bug in runtests which would delete /etc/hosts.
4291 * Use 64-bit object IDs wherever possible (not into VFS though)
4292 Remove ost_get_info, which is unused by lustre, and out of date.
4294 2002-07-03 Peter Braam <braam@clusterfs.com>
4295 * release version v0_4_2 Fixes a lookup error (type not passed)
4296 * move forward to head of Portals
4297 * move forward to latest Lustre kernel
4299 2002-06-25 Peter Braam <braam@clusterfs.com>
4300 * release version v0_4_1. Hopefully stable on single node use.