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.
16 Severity : enhancement
18 Description: lost problems with lov objid file
19 Details : Fixes some scability and access to not inited memory problems
20 in work with lov objdid file.
25 Description: lfs find does not continue on file error
26 Details : Continue other files processing when a file/dir is absent.
30 Description: Inconsistent usage of lustre_pack_reply()
31 Details : Standardize the usage of lustre_pack_reply() such that it
32 always generate a CERROR on failure.
35 Frequency : occasional
37 Description: MDS hang or stay in waiting lock
38 Details : If client receive lock with CBPENDING flag ldlm need send lock
39 cancel as separate rpc, to avoid situation when cancel request
40 can't processed due all i/o threads stay in wait lock.
44 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
48 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
52 Description: Update to SLES9 kernel-2.6.5-7.287.3.
56 Description: Update to RHEL4 latest kernel.
58 Severity : enhancement
60 Description: Build SLES10 patchless client fails
61 Details : The configure was broken by run ./configure with
62 --with-linux-obj=.... argument for patchless client. When the
63 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
64 can't search header adequately. Use absolute path such as
65 -I($LINUX)/include instead.
69 Description: interrupt oig_wait produce painc on resend.
70 Details : brw_redo_request can be used for resend requests from ptlrpcd and
71 private set, and this produce situation when rq_ptlrpcd_data not
72 copyed to new allocated request and triggered LBUG on assert
73 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
74 ptlrpcd set if request is changed and can be safety changed to use
79 Description: LASSERT_{REQ,REP}SWAB macros are buggy
80 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
81 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
82 these macros and replace them with their difinitions instead.
86 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
87 Details : Update which_patch & target file for SLES10 latest kernel.
90 Description: Remove client patches from SLES 10 kernel.
91 Details : This causes SLES 10 clients to behave as patchless clients
92 even on a Lustre-patched (server) kernel.
94 Severity : enhancement
96 Description: self-adjustable client's lru lists
97 Details : use adaptive algorithm for managing client cached locks lru
98 lists according to current server load, other client's work
99 pattern, memory activities, etc. Both, server and client
100 side namespaces provide number of proc tunables for controlling
103 Severity : enhancement
105 Description: light-weight GSS support
106 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
107 and reduce performance overhead.
109 Severity : enhancement
111 Description: Linux keyring support
112 Details : Support using service of Linux keyring for Lustre GSS internal
113 context refresh/cache mechanism.
117 Description: Fix errors in lfs documentation
118 Details : Fixes man pages
122 Description: don't use GFP_* in generic Lustre code.
123 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
127 Description: obdclass is limited by single OBD_ALLOC(idarray)
128 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
132 Description: warnings with build patchless client with vanila 2.6.19 and up
133 Details : change old ctl_table style and replace ctl_table/ctl_table_header
134 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
138 Frequency : Only for SLES
139 Description: sanity_quota fail test_1
140 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
141 file, which makes TSTID[2] inunique.
145 Frequency : Only for SLES9
146 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
147 Details : Update target/ChangeLog/which_patch .
151 Frequency : Only for test_10 in sanity-quota.sh
152 Description: a bug in quota test code
153 Details : checking a test flag in a wrong place
157 Frequency : Only for test_13 in sanity-quota.sh
158 Description: a bug in quota test script
159 Details : shouldn't check overfull allocated quota
163 Description: jbd statistics
164 Details : Port older jbd statistics patch for sles10
166 Severity : enhancement
168 Description: Add printing inode info into message about error in writepage.
173 Description: reply_lock_interpret crash due to race with it and lock cancel.
174 Details : Do not replay locks that are being cancelled. Do not reference
175 locks by their address during replay, just by their handle.
179 Frequency : When flocks are used.
180 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
181 Details : Flock locks might destroy just granted lock if it could be merged
182 with another existing flock, this is done in completion handler,
183 so teach ldlm_cli_enquque_fini that this is a valid case for
189 Description: Oops in read and write path when failing to allocate lock.
190 Details : Check if lock allocation failed and return error back.
194 Description: lstripe command fails for valid OST index
195 Details : The stripe offset is compared to lov->desc.ld_tgt_count
196 instead of lov->desc.ld_active_tgt_count.
200 Description: sanity.sh failed test 103
201 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
206 Description: df doesn't work properly if diskfs blocksize != 4K
207 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
211 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
212 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
216 Description: obd_health_check_timeout too short
217 Details : set obd_health_check_timeout as 1.5x of obd_timeout
221 Description: llapi_file_create() does not allow some changes
222 Details : add llapi_file_open() that allows specifying the mode and
223 open flags, and also returns an open file handle.
227 Description: merge and cleanup kernel patches.
228 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
232 Description: Add journal checksum support.(Kernel part)
233 Details : The journal checksum feature adds two new flags i.e
234 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
235 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
236 indicates that the commit block contains the checksum for
237 the blocks described by the descriptor blocks. Now commit
238 record can be sent to disk without waiting for descriptor
239 blocks to be written to disk. This behavior is controlled
240 using JBD2_FEATURE_ASYNC_COMMIT flag.
244 Description: OSS needs mutliple precreate threads
245 Details : Add ability to start more than one create thread per OSS.
249 Description: Sanity.sh test_65k failed: setstripe should have succeeded
250 Details : execute lfs setstripe on client
254 Description: mds_obd_create error creating tmp object
255 Details : When the user sets quota on root, llog will be affected and can't
256 create files and write files.
259 Frequency : Always on ia64 patchless client, and possibly others.
261 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
262 Details : This allows the patchless client to be loaded on architectures
267 Description: RedHat Update kernel for RHEL5
268 Details : Add the kernel config file for RHEL5.
272 Description: RedHat Update kernel for RHEL5
273 Details : Modify the kernel config file more closer RHEL5.
277 Description: Build failure against Centos5 (RHEL5)
278 Details : Define PAGE_SIZE when it isn't present.
282 Description: test_7 on sanity-quota.sh can't be used on separate machines
283 Details : add facet to handle it
287 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
288 Details : with NFS, the anon dentry's parent was set to itself in
289 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
290 corresponding dentry other than use rec->ur_name.
292 Severity : enhancement
294 Description: lfs setstripe enhancement
295 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
299 Description: enable data checksumming by default
300 Details : enable checksum by default, allow --disable-checksum
301 configure option and "-o nochecksum" mount option. Checksums
302 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
306 Description: lustre support for RHEL5
307 Details : Add support for RHEL5.
311 Description: Client eviction due to failover config
312 Details : after a connection loss, the lustre client should attempt to
313 reconnect to the last active server first before trying the
314 other potential connections.
318 Description: when mds and osts use different quota unit(32bit and 64bit),
319 quota will be released repeatly.
320 Details : void sending multiple quota reqs to mds, which will keep the
321 status between the reqs.
325 Description: osts not allocated evenly to files
326 Details : change the condition to increase offset_idx
329 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
331 Description: Data corruption for OSTs that are formatted larger than 2TB
333 Details : When generating the bio request for lustre file writes the
334 sector number would overflow a temporary variable before being
335 used for the IO. The data reads correctly from Lustre (which
336 will overflow in a similar manner) but other file data or
337 filesystem metadata may be corrupted in some cases.
341 Description: Tune the kernel for good SCSI performance.
342 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
343 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
348 Description: rewrite ext2-derived code in obdclass/uuid.c
349 Details : rewrite inherited code (uuid parsing code from ext2 utils)
350 from scratch preserving functionality.
354 Description: "lfs find -obd UUID" prints directories
355 Details : "lfs find -obd UUID" will return all directory names instead
356 of just file names. It is incorrect because the directories
357 do not reside on the OSTs.
361 Description: use i_size_read and i_size_write in 2.6 port
362 Details : replace inode->i_size access with i_size_read/write()
365 Frequency : when removing large files
367 Description: scheduling issue during removal of large Lustre files
368 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
369 It causes scheduling issues when removing large files (17TB in the
373 Frequency : only with liblustre clients on XT3
375 Description: evictions taking too long
376 Details : allow llrd to evict clients directly on OSTs
379 Frequency : only on ppc
381 Description: /proc/fs/lustre/devices broken on ppc
382 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
383 We should make sure the type of 'index' is loff_t and then cast
384 explicitly as needed below. Do not assign an explicitly cast
388 Frequency : only for rhel5
390 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
391 Details : Modify the target file & which_kernel.
393 Severity : enhancement
395 Description: omit set fsid for export NFS
396 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
397 as FS_REQUIRES_DEV for avoid problems with generate fsid.
401 Frequency : Always, for kernels after 2.6.16
402 Description: Fix warning idr_remove called for id=.. which is not allocated.
403 Details : Last kernels save old s_dev before kill super and not allow
404 to restore from callback - restore it before call kill_anon_super.
408 Description: buffer overruns could theoretically occur
409 Details : llapi_semantic_traverse() modifies the "path" argument by
410 appending values to the end of the origin string, and a
411 overrun may occur. Adding buffer overrun check in liblustreapi.
415 Description: Fix error on 'ls .' at the top of the Lustre mount.
416 Details : Don't revalidate dentry if it is a root dentry.
420 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
421 Details : Modify vm-tunables-rhel4.patch.
425 Description: Kernel config for 2.6.18-vanilla.
426 Details : Modify targets/2.6-vanilla.target.in.
427 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
428 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
429 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
430 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
434 Description: improve handling recoverable errors
435 Details : if request processig with error which can be recoverable on server
436 request should be resend, otherwise page released from cache and
441 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
442 Details : Modify target file & which_patch.
443 A flaw was found in the IA32 system call emulation provided
444 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
445 value could be stored in the %RAX register, which could trigger an
446 out-of-bounds system call table access. An untrusted local user
447 could exploit this flaw to run code in the kernel
448 (ie a root privilege escalation). (CVE-2007-4573).
452 Description: O_DIRECT bypasses client statistics.
453 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
454 read_ahead_stats were updated. copied the stats section from
455 osc_send_oap_rpc() into async_internal().
459 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
463 Description: change order of libsysio includes
464 Details : '#include sysio.h' should always come before '#include xtio.h'
468 Description: handle "serious error: objid * already exists" more gracefully
469 Details : If LAST_ID value on disk is smaller than the objects existing in
470 the O/0/d* directories, it indicates disk corruption and causes an
471 LBUG(). If the object is 0-length, then we should use the existing
472 object. This will help to avoid a full fsck in most cases.
474 Severity : enhancement
476 Description: adapt the lustre_config script to support the upgrade case
477 Details : Add "-u" option for lustre_config script to support upgrading 1.4
478 server targets to 1.6 in parallel.
482 Description: To avoid grant space > avaible space when the disk is almost
483 full. Without this patch you might see the error "grant XXXX >
484 available" or some LBUG about grant, when the disk is almost
486 Details : In filter_check_grant, for non_grant cache write, we should
487 check the left space by if (*left > ungranted + bytes), instead
488 of (*left > ungranted), because only we are sure the left space
489 is enough for another "bytes", then the ungrant space should be
490 increase. In client, we should update cl_avail_grant only there
491 is OBD_MD_FLGRANT in the reply.
495 Description: Update RHEL 4 kernel to fix local root privilege escalation.
496 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
497 described in CVE-2007-4573. This problem could allow untrusted
498 local users to gain root access.
501 Frequency : when using O_DIRECT and quotas
503 Description: Incorrect file ownership on O_DIRECT output files
504 Details : block usage reported by 'lfs quota' does not take into account
505 files that have been written with O_DIRECT.
510 Description: touch file failed when fs is not full
511 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
512 otherwise we can get ENOSP while fs is not full.
516 Description: parallel lock callbacks
517 Details : Instead of sending blocking and completion callbacks as separated
518 requests, adding them to a set and sending in parallel.
520 --------------------------------------------------------------------------------
522 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
524 * Support for kernels:
525 2.6.5-7.283 (SLES 9),
526 2.6.9-55.EL (RHEL 4),
527 2.6.16.46-0.14 (SLES 10),
528 2.6.18.8 vanilla (kernel.org)
529 * Client support for unpatched kernels:
530 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
531 2.6.16 - 2.6.22 vanilla (kernel.org)
532 * Due to recently discovered recovery problems, we do not recommend
533 using patchless RHEL 4 clients with this or any earlier release.
534 * Recommended e2fsprogs version: 1.39.cfs8
535 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
536 * Starting with this release, the ldiskfs backing filesystem required
537 by Lustre is now in its own package, lustre-ldiskfs. This package
538 should be installed. It is versioned separately from Lustre and
539 may be released separately in future.
543 Description: block reactivating mgc import until all deactivates complete
544 Details : Fix race when failing back MDT/MGS to itself (testing)
547 Frequency : at statup only
549 Description: mds_lov_synchronize race leads to various problems
550 Details : simultaneous MDT->OST connections at startup can cause the
551 sync to abort, leaving the OSC in a bad state.
553 Severity : enhancement
555 Description: add optional extra BUILD_VERSION info
556 Details : add a new environment variable (namely LUSTRE_VERS) which allows
557 to override the lustre version.
560 Frequency : 2.6.18 servers only
562 Description: ll_kern_mount() doesn't release the module reference
563 Details : The ldiskfs module reference count never drops down to 0
564 because ll_kern_mount() doesn't release the module reference.
569 Description: server LBUG when using old ost_num_threads parameter
570 Details : Accept the old ost_num_threads parameter but warn that it
571 is deprecated, and fix an off-by-one error that caused an LBUG.
576 Description: Transient SCSI error results in persistent IO issue
577 Details : iobuf->dr_error is not reinitialized to 0 between two
581 Frequency : sometimes when underlying device returns I/O errors
583 Description: OSTs not going read-only during write failures
584 Details : OSTs are not remounted read-only when the journal commit threads
585 get I/O errors because fsfilt_ext3 calls journal_start/stop()
586 instead of the ext3 wrappers.
590 Description: poor connect scaling with increasing client count
591 Details : Don't run filter_grant_sanity_check for more than 100 exports
592 to improve scaling for large numbers of clients.
595 Frequency : SLES10 only
597 Description: sanity-quota.sh quotacheck failed: rc = -22
598 Details : Quotas cannot be enabled on SLES10.
601 Frequency : liblustre clients only
603 Description: getdirentries does not give error when run on compute nodes
604 Details : getdirentries does not fail when the size specified as an argument
605 is too small to contain at least one entry
607 Severity : enhancement
609 Description: Add LNET router traceability for debug purposes
610 Details : If a checksum failure occurs with a router as part of the
611 IO path, the NID of the last router that forwarded the bulk data
612 is printed so it can be identified.
617 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
618 Details : Due to a race condition, liblustre clients were occasionally
621 Severity : enhancement
623 Description: lfs setstripe use optional parameters instead of postional
626 Severity : enhancement
628 Description: Nanosecond timestamp support for ldiskfs
629 Details : The on-disk ldiskfs filesystem has added support for nanosecond
630 resolution timestamps. There is not yet support for this at
631 the Lustre filesystem level.
634 Frequency : during server recovery
636 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
637 Details : request with rq_no_resend flag not awake l_wait_event if they get a
641 Frequency : nfs export on patchless client
643 Description: connectathon hang when test nfs export over patchless client
644 Details : Disconnected dentry cannot be found with lookup, so we do not need
645 to unhash it or make it invalid
648 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
649 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
650 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
655 Description: open req refcounting wrong on reconnect
656 Details : If reconnect happened between getting open reply from server and
657 call to mdc_set_replay_data in ll_file_open, we will schedule
658 replay for unreferenced request that we are about to free.
659 Subsequent close will crash in variety of ways.
660 Check that request is still eligible for replay in
661 mdc_set_replay_data().
666 Description: disable writes to filesystem when reading health_check file
667 Details : the default for reading the health_check proc file has changed
668 to NOT do a journal transaction and write to disk, because this
669 can cause reads of the /proc file to hang and block HA state
670 checking on a healthy but otherwise heavily loaded system. It
671 is possible to return to the previous behaviour during configure
672 with --enable-health-write.
674 Severity : enhancement
676 Description: 64-bit inode version
677 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
678 to the inode. This will be required for version-based recovery.
683 Description: MDS fails to start if a duplicate client export is detected
684 Details : in some rare cases it was possible for a client to connect to
685 an MDS multiple times. Upon recovery the MDS would detect this
686 and fail during startup. Handle this more gracefully.
688 Severity : enhancement
690 Description: Add -o localflock option to simulate old noflock
692 Details : This will achieve local-only flock/fcntl locks
698 Description: log_commit_thread vs filter_destroy race leads to crash
699 Details : Take import reference before releasing llog record semaphore
704 Description: Wrong request locking in request set processing
705 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
706 imp_delayed_list, in this place should be used imp_lock.
709 Frequency : when reconnection
711 Description: Grant Leak when osc reconnect to OST
712 Details : When osc reconnect ost, OST(filter) should check whether it
713 should grant more space to client by comparing fed_grant and
714 cl_avail_grant, and return the granted space to client instead
715 of "new granted" space, because client will call osc_init_grant
716 to update the client grant space info.
719 Frequency : when client reconnect to OST
721 Description: Grant Leak when osc do resend and replay bulk write
722 Details : When osc reconnect to OST, OST(filter)should clear grant info of
723 bulk write request, because the grant info while be sync between
724 OSC and OST when reconnect, and we should ignore the grant info
725 these of resend/replay write req.
730 Description: Grant space more than avaiable left space sometimes.
731 Details : When then OST is about to be full, if two bulk writing from
732 different clients came to OST. Accord the avaliable space of the
733 OST, the first req should be permitted, and the second one
734 should be denied by ENOSPC. But if the seconde arrived before
735 the first one is commited. The OST might wrongly permit second
736 writing, which will cause grant space > avaiable space.
739 Frequency : when client is evicted
741 Description: Grant might be wrongly erased when osc is evicted by OST
742 Details : when the import is evicted by server, it will fork another
743 thread ptlrpc_invalidate_import_thread to invalidate the
744 import, where the grant will be set to 0. While the original
745 thread will update the grant it got when connecting. So if
746 the former happened latter, the grant will be wrongly errased
747 because of this race.
752 Description: Checking Stale with correct fid
753 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
754 to check whether it is stale, because sometimes, we want the
755 enqueue happened anyway, and op_data.fid2 will not be initialized.
757 Severity : enhancement
759 Description: update patchless client
760 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
763 Frequency : only with 2.4 kernel
765 Description: random memory corruption
766 Details : size of struct ll_inode_info is to big for union inode.u and this
767 can be cause of random memory corruption.
772 Description: Memory leak in recovery
773 Details : Lov_mds_md was not free in an error handler in mds_create_object.
774 It should also check obd_fail before fsfilt_start, otherwise if
775 fsfilt_start return -EROFS,(failover mds during mds recovery).
776 then the req will return with repmsg->transno = 0 and rc = EROFS.
777 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
778 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
779 be freed no matter whether fsfilt_commit success or not.
782 Frequency : only with huge count clients
784 Description: Prevents from taking the superblock lock in llap_from_page for
786 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
787 for llap_from_page call in ll_removepage prevents from taking the
788 superblock lock for a soon died page.
793 Description: Not check open intent error before release open handle
794 Details : in some rare cases, the open intent error is not checked before
795 release open handle, which may cause
796 ASSERTION(open_req->rq_transno != 0), because it tries to release
797 the failed open handle.
802 Description: Set cat log bitmap only after create log success.
803 Details : in some rare cases, the cat log bitmap is set too early. and it
804 should be set only after create log success.
808 Description: Accessing a block bevice can re-enable I/O when Lustre is
809 tearing down a device.
810 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
815 Description: service threads may hog cpus when there are a lot of requests
816 Details : Insert cond_resched to give other threads a chance to use some CPU
821 Description: the cat log was not initialized in recovery
822 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
823 unlink log on mds will not be initialized until mds post
824 recovery. And also in mds post recovery, the unlink log will
825 initialization will be done asynchronausly, so there will be race
826 between add unlink log and unlink log initialization.
830 Description: brw_stats were being printed incorrectly
831 Details : brw_stats were being printed as log2 but all of them were not
832 recorded as log2. Also remove some code duplication arising from
833 filter_tally_{read,write}.
837 Frequency : rare, only in recovery.
838 Description: ASSERTION(req->rq_type != LI_POISON) failed
839 Details : imp_lock should be held while iterating over imp_sending_list for
840 prevent destroy request after get timeout in ptlrpc_queue_wait.
844 Description: replay-single.sh test 52 fails
845 Details : A lock's skiplist need to be cleanup when it being unlinked
846 from its resource list.
850 Description: Short directio read returns full requested size rather than
852 Details : Direct I/O operations should return actual amount of bytes
853 transferred rather than requested size.
855 Severity : enhancement
857 Description: metadata RPC reduction (e.g. for rm performance)
858 Details : decrease the amount of synchronous RPC between clients and servers
859 by canceling conflicing lock before the operation on the client
860 and packing thier handles into the main operation RPC to server.
862 Severity : enhancement
864 Description: Async OSC create to avoid the blocking unnecessarily.
865 Details : If a OST has no remain object, system will block on the creating
866 when need to create a new object on this OST. Now, ways use
867 pre-created objects when available, instead of blocking on an
868 empty osc while others are not empty. If we must block, we block
869 for the shortest possible period of time.
871 Severity : enhancement
873 Description: refine locking for avoid write wrong info into lov_objid file
874 Details : fix possible races with add new target and write/update data in
877 --------------------------------------------------------------------------------
879 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
884 Frequency : on some architectures
886 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
887 Details : Uninitialized flags sometimes cause configuration commands to
891 Frequency : patchless clients only
893 Description: missing __iget() symbol export
894 Details : The __iget() symbol export is missing. To avoid the need for
895 this on patchless clients the deathrow inode reaper is turned
896 off, and we depend on the VM to clean up old inodes. This
897 dependency was during via the fix for bug 12181.
899 --------------------------------------------------------------------------------
901 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
903 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
904 INTEROPERATE with older versions automatically. In many cases a
905 special upgrade step is needed. Please read the
906 user documentation before upgrading any part of a live system.
907 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
908 version will not interoperate with 1.6 betas before beta5 (1.5.95).
909 * WARNING: Lustre configuration and startup changes are required with
910 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
916 Severity : enhancement
918 Description: MountConf
919 Details : Lustre configuration is now managed via mkfs and mount
920 commands instead of lmc and lconf. New obd types (MGS, MGC)
921 are added for dynamic configuration management. See
922 https://mail.clusterfs.com/wikis/lustre/MountConf for
925 Severity : enhancement
927 Description: dynamic OST addition
928 Details : OSTs can now be added to a live filesystem
930 Severity : enhancement
932 Description: startup order invariance
933 Details : MDTs and OSTs can be started in any order. Clients only
934 require the MDT to complete startup.
936 Severity : enhancement
938 Description: parallel, asynchronous orphan cleanup
939 Details : orphan cleanup is now performed in separate threads for each
940 OST, allowing parallel non-blocking operation.
942 Severity : enhancement
944 Description: optimized stripe assignment
945 Details : stripe assignments are now made based on ost space available,
946 ost previous usage, and OSS previous usage, in order to try
947 to optimize storage space and networking resources.
949 Severity : enhancement
951 Description: Permanently set tunables
952 Details : All writable /proc/fs/lustre tunables can now be permanently
953 set on a per-server basis, at mkfs time or on a live system.
955 Severity : enhancement
957 Description: Lustre message v2
958 Details : Add lustre message format v2.
960 Severity : enhancement
962 Description: client OST exclusion list
963 Details : Clients can be started with a list of OSTs that should be
964 declared "inactive" for known non-responsive OSTs.
968 Description: SPEC SFS validation failure on NFS v2 over lustre.
969 Details : Changes the blocksize for regular files to be 2x RPC size,
970 and not depend on stripe size.
972 Severity : enhancement
974 Description: Multiple MD RPCs in flight.
975 Details : Further unserialise some read-only MDS RPCs - learn about intents.
976 To avoid overly-overloading MDS, introduce a limit on number of
977 MDS RPCs in flight for a single client and add /proc controls
978 to adjust this limit.
980 Severity : enhancement
982 Description: client read/write statistics
983 Details : Add client read/write call usage stats for performance
984 analysis of user processes.
985 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
986 file access. extents_stats shows chunk size distribution.
987 extents_stats_per_process show chunk size distribution per
990 Severity : enhancement
992 Description: mds statistics
993 Details : Add detailed mds operations statistics in
994 /proc/fs/lustre/mds/*/stats.
998 Description: Failure of copying files with lustre special EAs.
999 Details : Client side always return success for setxattr call for lustre
1000 special xattr (currently only "trusted.lov").
1005 Description: Refcount LNET uuids
1006 Details : The global LNET uuid list grew linearly with every startup;
1007 refcount repeated list entries instead of always adding to
1010 Severity : enhancement
1012 Description: Dynamic service threads
1013 Details : Within a small range, start extra service threads
1014 automatically when the request queue builds up.
1017 Frequency : mixed-endian client/server environments
1019 Description: mixed-endian crashes
1020 Details : The new msg_v2 system had some failures in mixed-endian
1023 Severity : enhancement
1025 Description: Easy OST removal
1026 Details : OSTs can be permanently deactivated with e.g. 'lctl
1027 conf_param lustre-OST0001.osc.active=0'
1029 Severity : enhancement
1031 Description: MGS proc entries
1032 Details : Added basic proc entries for the MGS showing what filesystems
1035 Severity : enhancement
1037 Description: provide MGS failover
1038 Details : Added config lock reacquisition after MGS server failover.
1040 Severity : enhancement
1042 Description: add Linux 2.4 support
1043 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1047 Description: a large application tries to do I/O to the same resource and dies
1048 in the middle of it.
1049 Details : Check the req->rq_arrival time after the call to
1050 ost_brw_lock_get(), but before we do anything about
1051 processing it & sending the BULK transfer request. This
1052 should help move old stale pending locks off the queue as
1053 quickly as obd_timeout.
1056 Frequency : when an incorrect nid is specified during startup
1058 Description: ptlrpc connect to non-existant node causes kernel crash
1059 Details : LNET can't be re-entered from an event callback, which
1060 happened when we expire a message after the export has been
1061 cleaned up. Instead, hand the zombie cleanup off to another
1064 Severity : enhancement
1066 Description: plain/inodebits lock performance improvement
1067 Details : Grouping plain/inodebits in granted list by their request modes
1068 and bits policy, thus improving the performance of search through
1072 Frequency : only if OST filesystem is corrupted
1074 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1075 Details : for a short time RPCs with bulk IO are in the replay list,
1076 but replay of bulk IOs is unimplemented. If the OST filesystem
1077 is corrupted due to disk cache incoherency and then replay is
1078 started it is possible to trip an assertion. Avoid putting
1079 committed RPCs into the replay list at all to avoid this issue.
1082 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1084 Description: System hang on startup
1085 Details : This bug allowed the liblustre (e.g. catamount) client to
1086 return to the app before handling all startup RPCs. This
1087 could leave the node unresponsive to lustre network traffic
1088 and manifested as a server ptllnd timeout.
1090 Severity : enhancement
1092 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1093 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1094 boolean to control whether to print peer debug info when a
1095 client's RPC times out.
1098 Frequency : only for kernels with patches from Lustre below 1.4.3
1100 Description: Remove old rdonly API
1101 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1104 Frequency : only for devices with external journals
1106 Description: Set external device read-only also
1107 Details : During a commanded failover stop, we set the disk device
1108 read-only while the server shuts down. We now also set any
1109 external journal device read-only at the same time.
1112 Frequency : when upgrading from 1.4 while trying to change parameters
1114 Description: The wrong (new) MDC name was used when setting parameters for
1115 upgraded MDT's. Also allows changing of OSC (and MDC)
1116 parameters if --writeconf is specified at tunefs upgrade time.
1119 Frequency : when setting specific ost indicies
1121 Description: QOS code breaks on skipped indicies
1122 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1123 created with --index need not be sequential.
1125 Severity : enhancement
1127 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1128 Details : The uninit_groups feature works in conjunction with the kernel
1129 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1130 up the pass1 processing of e2fsck. This is a read-only feature
1131 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1132 mount filesystems that have had this feature enabled.
1134 Severity : enhancement
1136 Description: Improve multi-block allocation algorithm to avoid fragmentation
1137 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1138 allocation locality and avoid filesystem fragmentation.
1140 ------------------------------------------------------------------------------
1142 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1144 * Support for kernels:
1145 2.4.21-47.0.1.EL (RHEL 3)
1146 2.6.5-7.283 (SLES 9)
1147 2.6.9-42.0.10.EL (RHEL 4)
1148 2.6.12.6 vanilla (kernel.org)
1149 2.6.16.27-0.9 (SLES 10)
1150 * Recommended e2fsprogs version: 1.39.cfs5
1152 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1156 Frequency : occasional, depends on client load and configuration
1157 Bugzilla : 12181, 12203
1158 Description: data loss for recently-modified files
1160 Details : In some cases it is possible that recently written or created
1161 files may not be written to disk in a timely manner (this should
1162 normally be within 30s unless client IO load is very high).
1163 The problem appears as zero-length files or files that are a
1164 multiple of 1MB in size after a client crash or client eviction
1165 that are missing data at the end of the file.
1167 This problem is more likely to be hit on clients where files are
1168 repeatedly created and unlinked in the same directory, clients
1169 have a large amount of RAM, have many CPUs, the filesystem has
1170 many OSTs, the clients are rebooted frequently, and/or the files
1171 are not accessed by other nodes after being written.
1173 The presence of the problem can be detected by looking at
1174 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1175 smaller than the second (nr_unused) then dirty files will not
1176 be flushed automatically to disk. "sync; sleep 10" should be
1177 run several times on the node before unmounting it to update
1178 Lustre (this is also safe to run on nodes without this problem).
1180 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1181 that can cause this same problem, so customers using that kernel
1182 also need to update the kernel in addition to Lustre. In order
1183 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1185 It is normal that files written just before a client crash (less
1186 than 30s) may not yet have been flushed to disk, even for local
1190 Frequency : frequent on thin XT3 nodes
1192 Description: UUID collision on thin XT3 Linux nodes
1193 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1194 randomly, since we relied on an insufficiently-seeded PRNG.
1199 Description: OSS hangs after "All ost request buffers busy"
1200 Details : A deadlock between quota and journal operations caused OSS
1201 hangs after printing "All ost request buffers busy."
1204 Frequency : always on liblustre builds
1206 Description: Cleanup compiler warnings on liblustre
1209 Frequency : always on liblustre builds on XT3
1211 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1215 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1216 > 32000 subdirectories
1217 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1218 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1219 subdirectory count crosses 32000. This will aid e2fsck to
1220 correctly handle more than 32000 subdirectories.
1223 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1225 Description: System hang on startup
1226 Details : This bug allowed the liblustre (e.g. catamount) client to
1227 return to the app before handling all startup RPCs. This
1228 could leave the node unresponsive to lustre network traffic
1229 and manifested as a server ptllnd timeout.
1231 Severity : enhancement
1233 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1234 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1235 boolean to control whether to print peer debug info when a
1236 client's RPC times out.
1241 Description: make O_SYNC working on 2.6 kernels
1242 Details : 2.6 kernels use different method for mark pages for write,
1243 so need add a code to lustre for O_SYNC work.
1248 Description: Failure to close file and release space on NFS
1249 Details : Put inode details into lock acquired in ll_intent_file_open.
1250 Use mdc_intent_lock in ll_intent_open to properly
1251 detect all kind of errors unhandled by mdc_enqueue.
1256 Description: proc file read during shutdown sometimes raced obd removal,
1258 Details : Add lock to prevent obd access after proc file removal.
1261 Frequency : Only for files larger than 4GB on 32-bit clients.
1263 Description: improperly doing page alignment of locks
1264 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1265 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1270 Description: RPCs being resent when they shouldn't be
1271 Details : Some RPCs that should not be resent are being resent. This
1272 can cause inconsistencies in the RPC state machine. Do not
1273 resend such requests.
1276 Frequency : rare, only with NFS export
1278 Description: Crash on NFS re-export node
1279 Details : under very unusual load conditions an assertion is hit in
1280 ll_intent_file_open()
1283 Frequency : only if OST filesystem is corrupted
1285 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1286 Details : for a short time RPCs with bulk IO are in the replay list,
1287 but replay of bulk IOs is unimplemented. If the OST filesystem
1288 is corrupted due to disk cache incoherency and then replay is
1289 started it is possible to trip an assertion. Avoid putting
1290 committed RPCs into the replay list at all to avoid this issue.
1295 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1296 Details : Large single O_DIRECT read and write calls can fail to allocate
1297 a sufficiently large buffer to process the request. In case of
1298 allocation failure the allocation is retried with a smaller
1299 buffer and broken into smaller requests.
1301 Severity : enhancement
1303 Description: Add -o localflock option to simulate old noflock behaviour.
1304 Details : This will achieve local-only flock/fcntl locks coherentness.
1309 Description: versioning check is incomplete
1310 Details : Checking the version difference of client vs. server, report
1311 error if the gap is too big.
1316 Description: add support PG_writeback bit
1317 Details : add support for PG_writeback bit for Lustre, for more carefull
1318 work with page cache in 2.6 kernel. This also fix some deadlocks
1319 and remove hack for work O_SYNC with 2.6 kernel.
1321 Severity : enhancement
1323 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1324 Details : The uninit_groups feature works in conjunction with the kernel
1325 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1326 up the pass1 processing of e2fsck. This is a read-only feature
1327 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1328 mount filesystems that have had this feature enabled.
1330 Severity : enhancement
1332 Description: Improve multi-block allocation algorithm to avoid fragmentation
1333 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1334 allocation locality and avoid filesystem fragmentation.
1336 ------------------------------------------------------------------------------
1338 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1340 * Support for kernels:
1341 2.6.9-42.0.3.EL (RHEL 4)
1342 2.6.5-7.276 (SLES 9)
1343 2.4.21-47.0.1.EL (RHEL 3)
1344 2.6.12.6 vanilla (kernel.org)
1345 2.6.16.21-0.8 (SLES10)
1346 * Recommended e2fsprogs version: 1.39.cfs2-0
1348 * The backwards-compatible /proc/sys/portals symlink has been removed
1349 in this release. Before upgrading, please ensure that you change
1350 any configuration scripts or /etc/sysctl.conf files that access
1351 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1352 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1353 in advance of the upgrade on any system running Lustre 1.4.6 or
1354 newer, since /proc/sys/lnet was added in that version.
1355 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1359 Frequency : only when quota is used
1361 Description: avoid scanning export list for quota master
1362 Details : Change the algorithms to avoid scanning export list in order
1363 to improve the efficiency.
1366 Frequency : MDS failover only, very rarely
1368 Description: "went back in time" messages on mds failover
1369 Details : The greatest transno may be lost when the current operation
1370 finishes with an error (transno==0) and the client's last_rcvd
1371 record is over-written. Save the greatest transno in the
1372 mds_last_transno for this case.
1375 Frequency : always for specific kernels and striping counts
1377 Description: client may get "Matching packet too big" without ACL support
1378 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1379 when trying to access files in certain configurations. The
1380 clients should in fact be denied when mounting because they do
1381 not understand ACLs.
1384 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1386 Description: many clients connecting with IO in progress causes connect timeouts
1387 Details : Avoid synchronous journal commits to avoid delays caused by many
1388 clients connecting/disconnecting when bulk IO is in progress.
1389 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1390 OST_IO_PORTAL to avoid delays behind potentially many pending
1394 Frequency : occasionally with multiple writers to a single file
1396 Description: shared writes to file may result in wrong size reported by stat()
1397 Details : Allow growing of kms when extent lock is cancelled
1400 Frequency : always with random mmap IO to multi-striped file
1402 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1403 Details : Only if the hole is at the end of OST object so that kms is too
1404 small. Fix is to increase kms accordingly in ll_nopage.
1407 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1409 Description: writes to a missing object would leak memory on the OST
1410 Details : If there is an inconsistency between the MDS and OST filesystems,
1411 such that the MDS references an object that doesn't exist, writes
1412 to that object will leak memory due to incorrect cleanup in the
1413 error handling path, eventually running out of memory on the OST.
1418 Description: Creating too long symlink causes lustre errors
1419 Details : Check symlink and name lengths before sending requests to MDS.
1422 Frequency : only if flock is enabled (not on by default)
1424 Description: posix locks not released on fd closure on 2.6.9+
1425 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1426 kernels, this caused such locks not to be released on fd close and
1427 then assertions on fs unmount about still used locks.
1430 Frequency : MDS failover only, very rarely
1432 Description: clients may get ASSERTION(granted_lock != NULL)
1433 Details : When request was taking a long time, and a client was resending
1434 a getattr by name lock request. The were multiple lock requests
1435 with the same client lock handle and
1436 mds_getattr_name->fixup_handle_for_resent_request found one of the
1437 lock handles but later failed with ASSERTION(granted_lock != NULL).
1442 Description: handle->h_buffer_credits > 0, assertion failure
1443 Details : h_buffer_credits is zero after truncate, causing assertion
1444 failure. This patch extends the transaction or creates a new
1448 Frequency : NFS re-export or patchless client
1449 Bugzilla : 11179, 10796
1450 Description: Crash on NFS re-export node (__d_move)
1451 Details : We do not want to hash the dentry if we don't have a lock.
1452 But if this dentry is later used in d_move, we'd hit uninitialised
1453 list head d_hash, so we just do this to init d_hash field but
1454 leave dentry unhashed.
1457 Frequency : NFS re-export or patchless client
1459 Description: NFS exports has problem with symbolic link
1460 Details : lustre client didn't properly install dentry when re-exported
1461 to NFS or running patchless client.
1464 Frequency : NFS re-export or patchless client
1466 Description: Various nfs/patchless fixes.
1467 Details : fixes reuse disconected alias for lookup process - this fixes
1468 warning "find_exported_dentry: npd != pd",
1469 fix permission error with open files at nfs.
1470 fix apply umask when do revalidate.
1473 Frequency : occasional
1475 Description: Crash on NFS re-export node
1476 Details : calling clear_page() on the wrong pointer triggered oops in
1477 generic_mapping_read().
1480 Frequency : rarely, using O_DIRECT IO
1482 Description: unaligned directio crashes client with LASSERT
1483 Details : check for unaligned buffers before trying any requests.
1486 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1488 Description: stale data returned from RAID cache
1489 Details : If only a small amount of IO is done to the RAID device before
1490 reading it again it is possible to get stale data from the RAID
1491 cache instead of reading it from disk.
1494 Frequency : always for sles10 kernel
1496 Description: sles10 support
1497 Details : ll_follow_link: compile fixes and using of nd_set_link
1498 under newer kernels.
1501 Frequency : depends on arch, kernel and compiler version, always on sles10
1504 Description: recursive or deep enough symlinks cause stack overflow
1505 Details : getting rid of large stack-allocated variable in
1509 Frequency : depends on hardware
1511 Description: lustre write performance loss in the SLES10 kernel
1512 Details : the performance loss is caused by using of write barriers in the
1513 ext3 code. The SLES10 kernel turns barrier support on by
1514 default. The fix is to undo that change for ldiskfs.
1516 ------------------------------------------------------------------------------
1518 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1520 * Support for kernels:
1521 2.6.9-42.0.3EL (RHEL 4)
1522 2.6.5-7.276 (SLES 9)
1523 2.4.21-47.0.1.EL (RHEL 3)
1524 2.6.12.6 vanilla (kernel.org)
1528 Frequency : quota enabled and large files being deleted
1530 Description: releasing more than 4GB of quota at once hangs OST
1531 Details : If a user deletes more than 4GB of files on a single OST it
1532 will cause the OST to spin in an infinite loop. Release
1533 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1538 Description: statfs data retrieved from /proc may be stale or zero
1539 Details : When reading per-device statfs data from /proc, in the
1540 {kbytes,files}_{total,free,avail} files, it may appear
1541 as zero or be out of date.
1544 Frequency : systems with MD RAID1 external journal devices
1546 Description: lconf's call to blkid is confused by RAID1 journal devices
1547 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1548 of returning all block devices that match the journal UUID.
1551 Frequency : always, for aggregate stripe size over 4GB
1553 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1554 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1555 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1558 Frequency : always if "lfs find" used on a local file/directory
1560 Description: "lfs find" segfaults if used on a local file/directory
1561 Details : The case where a directory component was not specified wasn't
1562 handled correctly. Handle this properly.
1565 Frequency : always on ppc64
1567 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1568 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1569 perform endianness conversion.
1572 Frequency : rarely (truncate to non-zero file size after write under load)
1573 Bugzilla : 10730, 10687
1574 Description: Files padded with zeros to next 4K multiple
1575 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1576 it is possible that files that are truncated to a non-zero size
1577 immediately after being written are filled with zero bytes beyond
1578 the truncated size. No file data is lost.
1580 Severity : enhancement
1582 Description: Allow recovery/failover for liblustre clients.
1583 Details : liblustre clients were unaware of failover configurations until
1586 Severity : enhancement
1588 Description: user file locks should fail when not mounting with flock option
1589 Details : Set up an error-returning stub in ll_file_operations.lock field
1590 to prevent incorrect behaviour when client is mounted without
1591 flock option. Also, set up properly f_op->flock field for
1595 Frequency : always on ia64
1597 Description: "lfs df" loops on printing out MDS statfs information
1598 Details : The obd_ioctl_data was not initialized and in some systems
1599 this caused a failure during the ioctl that did not return
1600 an error. Initialize the struct and return an error on failure.
1603 Frequency : SLES 9 only
1605 Description: Error of copying files with lustre special EAs as root
1606 Details : Client side always return success for setxattr call for lustre
1607 special xattr (currently only "trusted.lov").
1610 Frequency : rarely on clusters with both ia64+i386 clients
1612 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1613 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1614 clients are doing concurrent writes to the same file it is
1615 possible that the ia64 clients may LASSERT because the OST
1616 extent locks are not PAGE_SIZE aligned. Ensure that grown
1617 locks are always aligned on the request boundary.
1620 Frequency : specific use, occasional
1622 Description: Overwriting in use executable truncates on-disk binary image
1623 Details : If one node attempts to overwrite an executable in use by
1624 another node, we now correctly return ETXTBSY instead of
1625 truncating the file.
1627 Severity : enhancement
1629 Description: Async OSC create to avoid the blocking unnecessarily.
1630 Details : If a OST has no remain object, system will block on the creating
1631 when need to create a new object on this OST. Now, ways use
1632 pre-created objects when available, instead of blocking on an
1633 empty osc while others are not empty. If we must block, we block
1634 for the shortest possible period of time.
1639 Description: chmod on Lustre root is propagated to other clients
1640 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1641 invalid by the follow_mount time.
1644 Frequency : liblustre clients only
1646 Description: Race in 'instant cancel' lock handling could lead to such locks
1647 never to be granted in case of SMP MDS
1648 Details : Do not destroy not yet granted but cbpending locks in
1652 Frequency : replay/resend of open
1654 Description: non null lock assetion failure in mds_intent_policy
1655 Details : Trying to replay/resend lockless open requests resulted in
1656 mds_open() returning 0 with no lock. Now it sets a flag if
1657 a lock is going to be returned.
1659 Severity : enhancement
1661 Description: Checksum enhancements
1662 Details : New checksum enhancements allow for resending RPCs that failed
1665 Severity : enhancement
1667 Description: Tunables on number of dirty pages in cacche
1668 Details : Allow to set limit on number of dirty pages cached.
1673 Description: client crash on unmount - lock still has references
1674 Details : In some error handling cases it was possible to leak a lock
1675 reference on a client while accessing a file. This was not
1676 harmful to the client during operation, but would cause the
1677 client to crash when the filesystem is unmounted.
1680 Frequency : specific case, rare
1682 Description: ETXTBSY on mds though file not in use by client
1683 Details : ETXTBSY is no longer incorrectly returned when attempting to
1684 chmod or chown a directory that the user previously tried to
1685 execute or a currently-executing binary.
1688 Frequency : extremely rare except on liblustre-based clients
1690 Description: Lustre space not freed when files are deleted
1691 Details : Clean up open-unlinked files after client eviction. Previously
1692 the unlink was skipped and the files remained as orphans.
1697 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1698 Details : In some cases it was possible to send a blocking callback to
1699 a client doing a glimpse, even though that client didn't get
1700 a lock granted. When the glimpse lock is cancelled on the OST
1701 the freed lock is left on the waiting list and corrupted the list.
1704 Frequency : all core dumps
1706 Description: Broke core dumps to lustre
1707 Details : Negative dentry may be unhashed if parent does not have UPDATE
1708 lock, but some callers, e.g. do_coredump, expect dentry to be
1709 hashed after successful create, hash it in ll_create_it.
1711 ------------------------------------------------------------------------------
1713 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1715 * Support for kernels:
1716 2.6.9-42.0.2.EL (RHEL 4)
1717 2.6.5-7.276 (SLES 9)
1718 2.4.21-40.EL (RHEL 3)
1719 2.6.12.6 vanilla (kernel.org)
1723 Frequency : always on RHEL 3
1725 Description: Number of open files grows over time
1726 Details : The number of open files grows over time, whether or not
1727 Lustre is started. This was due to a filp leak introduced
1728 by one of our kernel patches.
1730 ------------------------------------------------------------------------------
1732 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1734 * Support for kernels:
1735 2.6.9-42.EL (RHEL 4)
1736 2.6.5-7.276 (SLES 9)
1737 2.4.21-40.EL (RHEL 3)
1738 2.6.12.6 vanilla (kernel.org)
1743 Bugzilla : 5719, 9635, 9792, 9684
1744 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1745 Details : If a server is under heavy load and cannot reply to new
1746 connection requests before the client resends the (re)connect,
1747 the connection handling code can behave badly if two service
1748 threads are concurrently handing separate (re)connections from
1749 the same client. Add better locking to the connection handling
1750 code, and ensure that only a single connection will be processed
1751 for a given client UUID, even if the lock is dropped.
1753 Severity : enhancement
1755 Description: add TCP zero-copy support to kernel
1756 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1757 sends if the hardware supports scatter-gather and checksumming.
1758 This allows socklnd to do client-write and server-read more
1759 efficiently and reduce CPU utilization from skbuf copying.
1762 Frequency : only if NFS exporting from client
1764 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1765 Details : If setting attributes on a file created under NFS that had
1766 never been opened it would be possible to oops the client
1767 if the file had no objects.
1770 Frequency : always for liblustre
1772 Description: liblustre client does MDS+OSTs setattr RPC for each write
1773 Details : When doing a write from a liblustre client, the client
1774 incorrectly issued an RPC to the MDS and each OST the file was
1775 striped over in order to update the timestamps. When writing
1776 with small chunks and many clients this could overwhelm the MDS
1777 with RPCs. In all cases it would slow down the write because
1778 these RPCs are unnecessary.
1780 Severity : enhancement
1782 Description: allow number of MDS service threads to be changed at module load
1783 Details : It is now possible to change the number of MDS service threads
1784 running. Adding "options mds mds_num_threads={N}" to the MDS's
1785 /etc/modprobe.conf will set the number of threads for the next
1786 time Lustre is restarted (assuming the "mds" module is also
1787 reloaded at that time). The default number of threads will
1788 stay the same, 32 for most systems.
1793 Description: OST crash if filesystem is unformatted or corrupt
1794 Details : If an OST is started on a device that has never been formatted
1795 or if the filesystem is corrupt and cannot even mount then the
1796 error handling cleanup routines would dereference a NULL pointer.
1801 Description: NULL pointer deref in llap_from_page.
1802 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1803 incorrect) mapping if the page was truncated/reclaimed while it was
1804 searched for. Check for this condition and skip such pages when
1805 doing readahead. Introduce extra check to llap_from_page() to
1806 verify page->mapping->host is non-NULL (so page is not anonymous).
1809 Frequency : Sometimes when using sys_sendfile
1811 Description: "page not covered by a lock" warnings from ll_readpage
1812 Details : sendfile called ll_readpage without right page locks present.
1813 Now we introduced ll_file_sendfile that does necessary locking
1814 around call to generic_file_sendfile() much like we do in
1818 Frequency : with certain MDS communication failures at client mount time
1820 Description: NULL pointer deref after failed client mount
1821 Details : a client connection request may delayed by the network layer
1822 and not be sent until after the PTLRPC layer has timed out the
1823 request. If the client fails the mount immediately it will try
1824 to clean up before the network times out the request. Add a
1825 reference from the request import to the obd device and delay
1826 the cleanup until the network drops the request.
1829 Frequency : occasionally during client (re)connect
1831 Description: assertion failure during client (re)connect
1832 Details : processing a client connection request may be delayed by the
1833 client or server longer than the client connect timeout. This
1834 causes the client to resend the connection request. If the
1835 original connection request is replied in this interval, the
1836 client may trip an assertion failure in ptlrpc_connect_interpret()
1837 which thought it would be the only running connect process.
1840 Frequency : only with obd_echo servers and clients that are rebooted
1842 Description: kernel BUG accessing uninitialized data structure
1843 Details : When running an obd_echo server it did not start the ping_evictor
1844 thread, and when a client was evicted an uninitialized data
1845 structure was accessed. Start the ping_evictor in the RPC
1846 service startup instead of the OBD startup.
1848 Severity : enhancement
1849 Bugzilla : 10193 (patchless)
1850 Description: Remove dependency on various unexported kernel interfaces.
1851 Details : No longer need reparent_to_init, exit_mm, exit_files,
1852 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1855 Frequency : rare (only users of deprecated and unsupported LDAP config)
1857 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1858 Details : LDAP apparently contains 'lustreName' attributes instead of
1859 'name'. A simple remapping of the name is sufficient.
1862 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1864 Description: waiting_locks_callback trips kernel BUG if client is evicted
1865 Details : Running with the dump_on_timeout debug flag turned on makes
1866 it possible that the waiting_locks_callback() can try to dump
1867 the Lustre kernel debug logs from an interrupt handler. Defer
1868 this log dumping to the expired_lock_main() thread.
1870 Severity : enhancement
1872 Description: Support NFS exporting on 2.6 kernels.
1873 Details : Implement non-rawops metadata methods for NFS server to use without
1874 changing NFS server code.
1877 Frequency : very rare (synthetic metadata workload only)
1879 Description: two racing renames might cause an MDS thread to deadlock
1880 Details : Running the "racer" program may cause one MDS thread to rename
1881 a file from being the source of a rename to being the target of
1882 a rename at exactly the same time that another thread is doing
1883 so, and the second thread has already enqueued these locks after
1884 doing a lookup of the target and is trying to relock them in
1885 order. Ensure that we don't try to re-lock the same resource.
1888 Frequency : only very large systems with liblustre clients
1890 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1891 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1892 all OSTs in parallel. This allows the eviction of stale liblustre
1893 clients to proceed much faster than if they were done in series,
1894 and also offers similar improvements for other set_info RPCs.
1899 Description: excessive CPU usage during initial read phase on client
1900 Details : During the initial read phase on a client, it would agressively
1901 retry readahead on the file, consuming too much CPU and impacting
1902 performance (since 1.4.5.8). Improve the readahead algorithm
1903 to avoid this, and also improve some other common cases (read
1904 of small files in particular, where "small" is files smaller than
1905 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1910 Description: MDS crash when receiving packet with unknown intent.
1911 Details : Do not LBUG in unknown intent case, just return -EFAULT
1913 Severity : enhancement
1914 Bugzilla : 9293, 9385
1915 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1916 Details : Do not serialize getattr (non-intent version) and statfs.
1919 Frequency : occasional, when OST network is overloaded/intermittent
1921 Description: client evicted by OST after bulk IO timeout
1922 Details : If a client sends a bulk IO request (read or write) the OST
1923 may evict the client if it is unresposive to its data GET/PUT
1924 request. This is incorrect if the network is overloaded (takes
1925 too long to transfer the RPC data) or dropped the OST GET/PUT
1926 request. There is no need to evict the client at all, since
1927 the pinger and/or lock callbacks will handle this, and the
1928 client can restart the bulk request.
1931 Frequency : Always when mmapping file with no objects
1933 Description: client crashes when mmapping file with no objects
1934 Details : Check that we actually have objects in a file before doing any
1935 operations on objects in ll_vm_open, ll_vm_close and
1941 Description: Request leak when working with deleted CWD
1942 Details : Introduce advanced request refcount tracking for requests
1943 referenced from lustre intent.
1945 Severity : Enhancement
1947 Description: Cache open file handles on client.
1948 Details : MDS now will return special lock along with openhandle, if
1949 requested and client is allowed to hold openhandle, even if unused,
1950 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1951 closing files for every read/write openration.
1953 Severity : Enhancement
1955 Description: Cache open negative dentries on client when possible.
1956 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1957 negative dentries on lock revocation.
1962 Description: Remounting a client read-only wasn't possible with a zconf mount
1963 Details : It wasn't possible to remount a client read-only with llmount.
1965 Severity : enhancement
1966 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1967 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1968 MPICH is included to add Lustre-specific ADIO interfaces.
1969 This is based closely on the UFS ADIO layer and only differs
1970 in file creation, in order to allow the OST striping to be set.
1971 This is user-contributed code and not supported by CFS.
1976 Description: extended inode attributes (immutable, append-only) work improperly
1977 when 2.4 and 2.6 kernels are used on client/server or vice versa
1978 Details : Introduce kernel-independent values for these flags.
1980 Severity : enhancement
1983 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1984 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1985 granularity. Now, Fractional MB granularity is be supported,
1986 this is very useful for low memory system.
1988 Severity : enhancement
1990 Description: Getattr by fid
1991 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1992 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1993 have locks and avoids getattr rpc requests for it on every stat.
1996 Frequency : Always, for filesystems larger than 2TB
1998 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1999 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2000 filesystems larger than 2TB to be mounted with mballoc enabled.
2003 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2005 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2006 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2007 filesystems larger than 2TB is mounted then the kernel will
2008 silently corrupt the start of the filesystem. CONFIG_LBD is
2009 enabled for all CFS-supported kernels, but the possibility of
2010 this happening with a modified kernel config exists.
2012 Severity : enhancement
2014 Description: add client O_DIRECT support for 2.6 kernels
2015 Details : It is now possible to do O_DIRECT reads and writes to files
2016 in the Lustre client mountpoint on 2.6 kernel clients.
2018 Severity : enhancement
2020 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2021 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2022 parallel, not waiting for response from every OST before sending
2023 a rpc to the next OST.
2028 Description: setattr vs write race when updating file timestamps
2029 Details : Client processes that update a file timestamp into the past
2030 right after writing to the file (e.g. tar) it is possible that
2031 the updated file modification time can be reset to the current
2032 time due to a race between processing the setattr and write RPC.
2034 Severity : enhancement
2036 Description: Bring 'lfs find' closer in line with regular Linux find.
2037 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2038 -print0 options and obtains all the needed info through the lustre
2041 Severity : enhancement
2043 Description: support up to 1024 configured devices on one node
2044 Details : change obd_dev array from statically allocated to dynamically
2045 allocated structs as they are first used to reduce memory usage
2050 Description: Flush dirty partially truncated pages during truncate
2051 Details : Immediatelly flush partially truncated pages in filter_setattr,
2052 this way we completely avoid having any pages in page cache on OST
2053 and can retire ugly workarounds during writes to flush such pages.
2058 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2059 Details : protect inode from truncation within vfs_unlink() context
2060 just take a reference before calling vfs_unlink() and release it
2061 when parent's i_sem is free.
2066 Description: last_id value checked outside lock on OST caused LASSERT failure
2067 Details : If there were multiple MDS->OST object precreate requests in
2068 flight, it was possible that the OST's last object id was checked
2069 outside a lock and incorrectly tripped an assertion. Move checks
2070 inside locks, and discard old precreate requests.
2073 Frequency : always, if extents are used on OSTs
2075 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2076 Details : OSTs using the extents format would not zero the high 16 bits of
2077 the index physical block number. This is not a problem for any
2078 OST filesystems smaller than 16TB, and no kernels support ext3
2079 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2080 new/modified files) and can be fixed for existing filesystems
2081 with e2fsprogs-1.39-cfs1.
2086 Description: import connection selection may be incorrect if timer wraps
2087 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2088 import connections to be ignored if the 32-bit jiffies counter
2089 wraps. Use a 64-bit jiffies counter.
2092 Frequency : very large clusters immediately after boot
2094 Description: LNET request buffers exhausted under heavy short-term load
2095 Details : If a large number of client requests are generated on a service
2096 that has previously never seen so many requests it is possible
2097 that the request buffer growth cannot keep up with the spike in
2098 demand. Instead of dropping incoming requests, they are held in
2099 the LND until the RPC service can accept more requests.
2102 Frequency : Sometimes during replay
2104 Description: Assertion failure in ll_local_open after replay.
2105 Details : If replay happened on an open request reply before we were able
2106 to set replay handler, reply will become not swabbed tripping the
2107 assertion in ll_local_open. Now we set the handler right after
2108 recognising of open request
2111 Frequency : very rare
2113 Description: kernel reports "badness in vsnprintf"
2114 Details : Reading from the "recovery_status" /proc file in small chunks
2115 may cause a negative length in lprocfs_obd_rd_recovery_status()
2116 call to vsnprintf() (which is otherwise harmless). Exit early
2117 if there is no more space in the output buffer.
2119 Severity : enhancement
2121 Description: clear OBD RPC statistics by writing to them
2122 Details : It is now possible to clear the OBD RPC statistics by writing
2123 to the "stats" file.
2128 Description: Client mtime is not the same on different clients after utimes
2129 Details : In some cases, the client was using the utimes() syscall on
2130 a file cached on another node. The clients now validate the
2131 ctime from the MDS + OSTs to determine which one is right.
2136 Description: Inability to activate failout mode
2137 Details : lconf script incorrectly assumed that in pythong string's numeric
2138 value is used in comparisons.
2141 Frequency : always with multiple stripes per file
2143 Description: Inefficient object allocation for mutli-stripe files
2144 Details : When selecting which OSTs to stripe files over, for files with
2145 a stripe count that divides evenly into the number of OSTs,
2146 the MDS is always picking the same starting OST for each file.
2147 Return the OST selection heuristic to the original design.
2152 Description: mount failures may take full timeout to return an error
2153 Details : Under some heavy load conditions it is possible that a
2154 failed mount can wait for the full obd_timeout interval,
2155 possibly several minutes, before reporting an error.
2156 Instead return an error as soon as the status is known.
2158 Frequency : quota enabled and large files being deleted
2160 Description: releasing more than 4GB of quota at once hangs OST
2161 Details : If a user deletes more than 4GB of files on a single OST it
2162 will cause the OST to spin in an infinite loop. Release
2163 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2168 Description: statfs data retrieved from /proc may be stale or zero
2169 Details : When reading per-device statfs data from /proc, in the
2170 {kbytes,files}_{total,free,avail} files, it may appear
2171 as zero or be out of date.
2174 Frequency : systems with MD RAID1 external journal devices
2176 Description: lconf's call to blkid is confused by RAID1 journal devices
2177 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2178 of returning all block devices that match the journal UUID.
2181 Frequency : always, for aggregate stripe size over 4GB
2183 Description: assertion fails when trying to use 4GB stripe size
2184 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2185 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2188 Frequency : always on ppc64
2190 Description: the first write on an ext3 filesystem with mballoc got stuck
2191 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2192 perform endianness conversion.
2194 ------------------------------------------------------------------------------
2196 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2198 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2199 INTEROPERATE with older versions automatically. Please read the
2200 user documentation before upgrading any part of a live system.
2201 * WARNING: Lustre networking configuration changes are required with
2202 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2205 * Support for kernels:
2206 2.6.9-22.0.2.EL (RHEL 4)
2207 2.6.5-7.244 (SLES 9)
2208 2.6.12.6 vanilla (kernel.org)
2211 Severity : enhancement
2212 Bugzilla : 7981/8208
2213 Description: Introduced Lustre Networking (LNET)
2214 Details : LNET is new networking infrastructure for Lustre, it includes
2215 a reorganized network configuration mode (see the user
2216 documentation for full details) as well as support for routing
2217 between different network fabrics. Lustre Networking Devices
2218 (LNDs) for the supported network fabrics have also been
2219 created for this new infrastructure.
2221 Severity : enhancement
2222 Description: Introduced Access control lists
2223 Details : clients can set ACLs on files and directories in order to have
2224 more fine-grained permissions than the standard Unix UGO+RWX.
2225 The MDS must be started with the "-o acl" mount option.
2227 Severity : enhancement
2228 Description: Introduced filesystem quotas
2229 Details : Administrators may now establish per-user quotas on the
2232 Severity : enhancement
2234 Description: Configuration change for the XT3
2235 The PTLLND is now used to run Lustre over Portals on the XT3
2236 The configure option(s) --with-cray-portals are no longer used.
2237 Rather --with-portals=<path-to-portals-includes> is used to
2238 enable building on the XT3. In addition to enable XT3 specific
2239 features the option --enable-cray-xt3 must be used.
2244 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2245 Details : A race between allocating a new llcd and re-getting the llcd_lock
2246 allowed another thread to grab newly-allocated llcd.
2248 Severity : enhancement
2250 Description: 2.6 OST async journal commit and locking fix to improve performance
2251 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2252 2.6 kernels are now async as they already were in 2.4 kernels so
2253 that they can commit concurrently with the network bulk transfer.
2254 For block-allocated files the filter allocation semaphore is held
2255 to avoid filesystem fragmentation during allocation. BKL lock
2256 removed for 2.6 xattr operations where it is no longer needed.
2261 Description: lconf incorrectly determined whether two IP networks could talk
2262 Details : In some more complicated routing and multiple-network
2263 configurations, lconf will avoid trying to make a network
2264 connection to a disjoint part of the IP space. It was doing the
2265 math incorrectly for one set of cases.
2270 Description: Fix for potential infinite loop processing records in an llog.
2271 Details : If an llog record is corrupted/zeroed, it is possible to loop
2272 forever in llog_process(). Validate the llog record length
2273 and skip the remainder of the block on error.
2276 Frequency : occasional (liblustre only)
2278 Description: liblustre could not open files whose last component is a symlink
2279 Details : sysio_path_walk() would incorrectly pass the open intent to
2280 intermediate path components.
2283 Frequency : rare (liblustre only with non-standard tuning)
2284 Bugzilla : 7201 (7350)
2285 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2286 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2287 after it has been released, only do this for referenced locks
2289 Severity : enhancement
2291 Description: specifying an (invalid) directory default stripe_size of -1
2292 would reset the directory default striping
2293 Details : stripe_size -1 was used internally to signal directory stripe
2294 removal, now use "all default" to signal dir stripe removal
2295 as a directory striping of "all default" is not useful
2298 Frequency : common for large clusters running liblustre clients
2300 Description: doing an ls when liblustre clients are running is slow
2301 Details : sending a glimpse AST to a liblustre client waits for every AST
2302 to time out, as liblustre clients will not respond. Since they
2303 cannot cache data we refresh the OST lock LVB from disk instead.
2305 Severity : enhancement
2307 Description: doing an ls at the same time as file IO can be slow
2308 Details : enqueue and other "small" requests can be blocked behind many
2309 large IO requests. Create a new OST IO portal for non-IO
2310 requests so they can be processed faster.
2313 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2315 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2316 created file with mode 000
2317 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2318 allow an ftruncate() as a non-root user to a file with mode 000.
2319 Lustre now respects this flag to disable mode checks when
2320 truncating a file owned by the user
2323 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2326 Description: Revoking locks from clients that went dead or catatonic might take
2328 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2329 cancellation of such locks instant on servers without waiting for
2330 any reply from clients. Clients drops these locks when cancel
2331 notification from server is received without replying.
2334 Frequency : liblustre-only, when liblustre client dies or becomes busy
2336 Description: Doing ls on Linux clients can take a long time with active
2338 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2339 granting such locks to it in the first place if possible. Locks
2340 are taken by proxy on the OST during the read or write and
2341 dropped immediately afterward. Add connect flags handling, do
2342 not grant locks to liblustre clients for glimpse ASTs.
2344 Severity : enhancement
2346 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2347 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2348 sequential reads and behaves badly with applications doing random
2349 reads. Improve it by reading ahead at least read region, and
2350 avoiding excessive large RPC for small reads.
2352 Severity : enhancement
2354 Description: Creating more than 1000 files for a single job may cause a load
2355 imbalance on the OSTs if there are also a large number of OSTs.
2356 Details : qos_prep_create() uses an OST index reseed value that is an
2357 even multiple of the number of available OSTs so that if the
2358 reseed happens in the middle of the object allocation it will
2359 still utilize the OSTs as uniformly as possible.
2364 Description: OST or MDS may oops in ping_evictor_main()
2365 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2366 but doesn't restart at beginning of obd_exports_timed list
2369 Severity : enhancement
2371 Description: improve by-nid export eviction on the MDS and OST
2372 Details : allow multiple exports with the same NID to be evicted at one
2373 time without re-searching the exports list.
2376 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2378 Description: MDS may oops in groups_free()
2379 Details : in rare race conditions a newly allocated group_info struct is
2380 freed again, and this can be NULL. The 2.4 compatibility code
2381 for groups_free() checked for a NULL pointer, but 2.6 did not.
2384 Frequency : common for liblustre clients doing little filesystem IO
2385 Bugzilla : 9352, 7313
2386 Description: server may evict liblustre clients accessing contended locks
2387 Details : if a client is granted a lock or receives a completion AST
2388 with a blocking AST already set it would not reply to the AST
2389 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2392 Frequency : lfs setstripe, only systems with more than 160 OSTs
2394 Description: unable to set striping with a starting offset beyond OST 160
2395 Details : llapi_create_file() incorrectly limited the starting stripe
2396 index to the maximum single-file stripe count.
2399 Frequency : LDAP users only
2401 Description: lconf did not handle in-kernel recovery with LDAP properly
2402 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2404 Severity : enhancement
2406 Description: bind OST threads to NUMA nodes to improve performance
2407 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2408 node and do their allocations there to localize memory access
2410 Severity : enhancement
2412 Description: llmount can determine client NID directly from Myrinet (GM)
2413 Details : the client NID code from gmnalnid was moved directly into
2414 llmount, removing the need to use this or specifying the
2415 client NID explicitly when mounting GM clients with zeroconf
2418 Frequency : if client is started with down MDS
2420 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2421 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2422 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2423 ptlrpc_import_delayed_req() because we didn't check if the
2424 request was interrupted, and we also didn't break out of the
2425 event loop if there was no timeout
2430 Description: data loss during non-page-aligned writes to a single file from
2431 both multiple nodes and multiple threads on one node at same time
2432 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2433 inconsistency led to false short-reads, that were cached and later
2434 used by ->prepare_write() to fill in partially written page,
2435 leading to data loss.
2438 Frequency : always, if lconf --abort_recovery used
2440 Description: lconf --abort_recovery fails with 'Operation not supported'
2441 Details : lconf was attempting to abort recovery on the MDT device and not
2444 Severity : enhancement
2446 Description: remove cleanup logs
2447 Details : replace lconf-generated cleanup logs with lustre internal
2448 cleanup routines. Eliminates the need for client-cleanup and
2451 Severity : enhancement
2453 Description: add support for EAs (user and system) on lustre filesystems
2454 Details : it is now possible to store extended attributes in the Lustre
2455 client filesystem, and with the user_xattr mount option it
2456 is possible to allow users to store EAs on their files also
2458 Severity : enhancement
2460 Description: Add possibility (config option) to show minimal available OST free
2462 Details : When compiled with --enable-mindf configure option, statfs(2)
2463 (and so, df) will return least minimal free space available from
2464 all OSTs as amount of free space on FS, instead of summary of
2465 free spaces of all OSTs.
2467 Severity : enhancement
2469 Description: do not expand extent locks acquired on OST-side
2470 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2471 by server: they are not cached anyway.
2474 Frequency : when mmap is used/binaries executed from Lustre
2476 Description: Unmmap pages before throwing them away from read cache.
2477 Details : llap_shrink cache now attempts to unmap pages before discarding
2478 them (if unmapping failed - do not discard). SLES9 kernel has
2479 extra checks that trigger if this unmapping is not done first.
2484 Description: lconf didn't resolve symlinks before checking to see whether a
2485 given mountpoint was already in use
2488 Frequency : when migrating failover services
2489 Bugzilla : 6395, 9514
2490 Description: When migrating a subset of services from a node (e.g. failback
2491 from a failover service node) the remaining services would
2492 time out and evict clients.
2493 Details : lconf --force (implied by --failover) sets the global obd_timeout
2494 to 5 seconds in order to quickly disconnect, but this caused
2495 other RPCs to time out too quickly. Do not change the global
2496 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2498 Severity : enhancement
2499 Frequency : if MDS is started with down OST
2500 Bugzilla : 9439,5706
2501 Description: Allow startup/shutdown of an MDS without depending on the
2502 availability of the OSTs.
2503 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2504 Add appropriate locking and lov-osc refcounts for safe
2505 cleaning. Add osc abort_inflight calls in case the
2506 synchronize never started.
2509 Frequency : occasional (Cray XT3 only)
2511 Description: root not authorized to access files in CRAY_PORTALS environment
2512 Details : The client process capabilities were not honoured on the MDS in
2513 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2514 been accessed by an authorized user then root was able to access
2515 the file on the local client also. The root user capabilities
2516 are now allowed on the MDS, as this environment has secure UID.
2519 Frequency : occasional
2521 Description: ldiskfs "too long searching" message happens too often
2522 Details : A debugging message (otherwise harmless) prints too often on
2523 the OST console. This has been reduced to only happen when
2524 there are fragmentation problems on the filesystem.
2529 Description: Division by zero in statfs when all OSCs are inactive
2530 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2531 lov_statfs divides by value returned by lov_get_stripecnt().
2535 Bugzilla : 9489, 3273
2536 Description: First write from each client to each OST was only 4kB in size,
2537 to initialize client writeback cache, which caused sub-optimal
2538 RPCs and poor layout on disk for the first writen file.
2539 Details : Clients now request an initial cache grant at (re)connect time
2540 and so that they can start streaming writes to the cache right
2541 away and always do full-sized RPCs if there is enough data.
2542 If the OST is rebooted the client also re-establishes its grant
2543 so that client cached writes will be honoured under the grant.
2548 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2549 Details : Now I/O RPCs go to different portal number and (presumably) fast
2550 lock requests (and glimses) and other RPCs get their own service
2551 threads pool that should be able to service those RPCs
2554 Severity : enhancement
2556 Description: Ability to exchange lustre version between client and servers and
2557 issue warnings at client side if client is too old. Also for
2558 liblustre clients there is ability to refuse connection of too old
2560 Details : New 'version' field is added to connect data structure that is
2561 filled with version info. That info is later checked by server and
2565 Frequency : rare, liblustre only.
2566 Bugzilla : 9296, 9581
2567 Description: Two simultaneous writes from liblustre at offset within same page
2568 might proceed at the same time overwriting eachother with stale
2570 Details : I/O lock withing llu_file_prwv was released too early, before data
2571 actually was hitting the wire. Extended lock-holding time until
2572 server acknowledges receiving data.
2575 Frequency : extremely rare. Never observed in practice.
2577 Description: avoid generating lustre_handle cookie of 0.
2578 Details : class_handle_hash() generates handle cookies by incrementing
2579 global counter, and can hit 0 occasionaly (this is unlikely, but
2580 not impossible, because initial value of cookie counter is
2581 selected randonly). Value of 0 is used as a sentinel meaning
2582 "unassigned handle" --- avoid it. Also coalesce two critical
2583 sections in this function into one.
2585 Severity : enhancement
2587 Description: allow liblustre clients to delegate truncate locking to OST
2588 Details : To avoid overhead of locking, liblustre client instructs OST to
2589 take extent lock in ost_punch() on client's behalf. New connection
2590 flag is added to handle backward compatibility.
2592 Severity : enhancement
2593 Bugzilla : 4928, 7341, 9758
2594 Description: allow number of OST service threads to be specified
2595 Details : a module parameter allows the number of OST service threads
2596 to be specified via "options ost ost_num_threads={N}" in the
2597 OSS's /etc/modules.conf or /etc/modprobe.conf.
2601 Bugzilla : 6146, 9635, 9895
2602 Description: servers crash with bad pointer in target_handle_connect()
2603 Details : In rare cases when a client is reconnecting it was possible that
2604 the connection request was the last reference for that export.
2605 We would temporarily drop the export reference and get a new
2606 one, but this may have been the last reference and the export
2607 was just destroyed. Get new reference before dropping old one.
2609 Severity : enhancement
2610 Frequency : if client is started with failover MDS
2612 Description: Allow multiple MDS hostnames in the mount command
2613 Details : Try to read the configuration from all specified MDS
2614 hostnames during a client mount in case the "primary"
2617 Severity : enhancement
2619 Description: Stop sending data to evicted clients as soon as possible.
2620 Details : Check if the client we are about to send or are sending data to
2621 was evicted already. (Check is done every second of waiting,
2622 for which l_wait_event interface was extended to allow checking
2623 of exit condition at specified intervals).
2626 Frequency : rare, normally only when NFS exporting is done from client
2628 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2630 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2633 Frequency : always, if srand() is called before liblustre initialization
2635 Description: Liblustre uses system PRNG disturbing its usage by user application
2636 Details : Introduce internal to lustre fast and high-quality PRNG for
2637 lustre usage and make liblustre and some other places in generic
2638 lustre code to use it.
2640 Severity : enhancement
2641 Bugzilla : 9477, 9557, 9870
2642 Description: Verify that the MDS configuration logs are updated when xml is
2643 Details : Check if the .xml configuration logs are newer than the config
2644 logs stored on the MDS and report an error if this is the case.
2645 Request --write-conf, or allow starting with --old_conf.
2647 Severity : enhancement
2649 Description: Handle symlinks in the path when checking if Lustre is mounted.
2650 Details : Resolve intermediate symlinks when checking if a client has
2651 mounted a filesystem to avoid duplicate client mounts.
2656 Description: lconf can hit an error exception but still return success.
2657 Details : The lconf command catches the Command error exception at the top
2658 level script context and will exit with the associated exit
2659 status, but doesn't ensure that this exit status is non-zero.
2664 Description: failure of ptlrpc thread startup can cause oops
2665 Details : Starting a ptlrpc service thread can fail if there are a large
2666 number of threads or the server memory is very fragmented.
2667 Handle this without oopsing.
2670 Frequency : always, only if liblustre and non-default acceptor port was used
2672 Description: liblustre cannot connect to servers with non-default acceptor port
2673 Details : tcpnal_set_default_params() was not called and was therefore
2674 ignoring the environment varaible TCPNAL_PORT, as well as other
2675 TCPNAL_ environment variables
2680 Description: two objects could be created on the same OST for a single file
2681 Details : If an OST is down, in some cases it was possible to create two
2682 objects on a single OST for a single file. No problems other
2683 than potential performance impact and spurious error messages.
2687 Bugzilla : 5681, 9562
2688 Description: Client may oops in ll_unhash_aliases
2689 Details : Client dcache may become inconsistent in race condition.
2690 In some cases "getcwd" can fail if the current directory is
2696 Description: Inode refcounting problems in NFS export code
2697 Details : link_raw functions used to call d_instantiate without obtaining
2698 extra inode reference first.
2702 Bugzilla : 9942, 9903
2703 Description: Referencing freed requests leading to crash, memleaks with NFS.
2704 Details : We used to require that call to ll_revalidate_it was always
2705 followed by ll_lookup_it. Also with revalidate_special() it is
2706 possible to call ll_revalidate_it() twice for the same dentry
2707 even if first occurence returned success. This fix changes semantic
2708 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2709 reference on a request referred from the intent.
2710 ll_intent_release() then releases such a request.
2713 Frequency : rare, normally benchmark loads only
2715 Description: unlinked inodes were kept in memory on the client
2716 Details : If a client is repeatedly creating and unlinking files it
2717 can accumulate a lot of stale inodes in the inode slab cache.
2718 If there is no other client load running this can cause the
2719 client node to run out of memory. Instead flush old inodes
2720 from client cache that have the same inode number as a new inode.
2723 Frequency : SLES9 2.6.5 kernel and long filenames only
2724 Bugzilla : 9969, 10379
2725 Description: utime reports stale NFS file handle
2726 Details : SLES9 uses out-of-dentry names in some cases, which confused
2727 the lustre dentry revalidation. Change it to always use the
2731 Frequency : rare, unless heavy write-truncate concurrency is continuous
2732 Bugzilla : 4180, 6984, 7171, 9963, 9331
2733 Description: OST becomes very slow and/or deadlocked during object unlink
2734 Details : filter_destroy() was holding onto the parent directory lock
2735 while truncating+unlinking objects. For very large objects this
2736 may block other threads for a long time and slow overall OST
2737 responsiveness. It may also be possible to get a lock ordering
2738 deadlock in this case, or run out of journal credits because of
2739 the combined truncate+unlink. Solution is to do object truncate
2740 first in one transaction without parent lock, and then do the
2741 final unlink in a new transaction with the parent lock. This
2742 reduces the lock hold time dramatically.
2745 Frequency : rare, 2.4 kernels only
2747 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2748 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2749 before unmounting their filesystem in order to avoid deadlock.
2750 The kernel_locked() function in 2.4 kernels only checks whether
2751 the kernel lock is held, not whether it is this process that is
2752 holding it as 2.6 kernels do.
2757 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2758 Details : The client ptlrpc code may be trying to reconnect to a down
2759 server before a previous connection attempt has timed out.
2760 Increase the reconnect interval to be longer than the connection
2761 timeout interval to avoid sending duplicate connections to
2765 Frequency : echo_client brw_test command
2767 Description: fix echo_client to work with OST preallocated code
2768 Details : OST preallocation code (5137) didn't take echo_client IO path
2769 into account: echo_client calls filter methods outside of any
2770 OST thread and, hence, there is no per-thread preallocated
2771 pages and buffers to use. Solution: hijack pga pages for IO. As
2772 a byproduct, this avoids unnecessary data copying.
2776 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2777 Description: Client can oops in mdc_commit_close() after open replay
2778 Details : It was possible for the MDS to return an open request with no
2779 transaction number in mds_finish_transno() if the client was
2780 evicted, but without actually returning an error. Clients
2781 would later try to replay that open and may trip an assertion
2782 Simplify the client close codepath, and always return an error
2783 from the MDS in case the open is not successful.
2786 Frequency : rare, 2.6 OSTs only
2788 Description: OST may deadlock under high load on fragmented files
2789 Details : If there was a heavy load and highly-fragmented OST filesystems
2790 it was possible to have all the OST threads deadlock waiting on
2791 allocation of biovecs, because the biovecs were not released
2792 until the entire RPC IO was completed. Instead, release biovecs
2793 as soon as they are complete to ensure forward IO progress.
2795 Severity : enhancement
2797 Description: Support for specifying external journal device at mount
2798 Details : If an OST or MDS device is formatted with an external journal
2799 device, this device major/minor is stored in the ext3 superblock
2800 and may not be valid for failover. Allow detecting and
2801 specifying the external journal at mount time.
2806 Description: Mounting an MDS with pending unlinked files may cause oops
2807 Details : target_finish_recovery() calls mds_postrecov() which returned
2808 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2809 considers this an error and immediately begins cleaning up the
2810 lov, just after starting the mds_lov process
2812 Severity : enhancement
2814 Description: Implement 'lfs df' to report actual free space on per-OST basis
2815 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2816 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2817 usage of objects; '-h' to report in human readable format.
2819 ------------------------------------------------------------------------------
2821 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2828 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2829 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2830 at startup during orphan unlinking. Instead, skip buddy update
2831 before setup, as it will be regenerated after recovery is complete.
2836 Description: If an OST is inactive, its locks might reference stale inodes.
2837 Details : lov_change_cbdata() must iterate over all namespaces, even if
2838 they are inactive to clear inode references from the lock.
2840 Severity : enhancement
2841 Frequency : occasional, if non-standard max_dirty_mb used
2843 Description: Client will block write RPCs if not enough grant
2844 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2845 then the client will block writes while waiting for another RPC
2846 to complete instead of consuming its dirty limit. With change
2847 we get improved performance when max_dirty_mb is small.
2849 Severity : enhancement
2850 Bugzilla : 3389, 6253
2851 Description: Add support for supplementary groups on the MDS.
2852 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2853 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2854 lookups for user supplementary groups into a cache.
2858 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2859 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2860 for liblustre O_EXCL is enforced by MDS.
2864 Description: readdir never returns NULL in liblustre.
2865 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2866 dentry in struct dirent.
2870 Frequency : liblustre only, depends on application IO pattern
2871 Description: liblustre clients evicted if not contacting servers
2872 Details : Don't put liblustre clients into the ping_evictor list, so
2873 they will not be evicted by the pinger ever.
2875 Severity : enhancement
2877 Description: Add ability to evict clients by NID from MDS.
2878 Details : By echoing "nid:$NID" string into
2879 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2880 $NID would be instantly evicted from this MDS and from all active
2881 OSTs connected to it.
2885 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2886 Details : lookup_it_finish() used to query file size from OSTs that was not
2891 Description: service threads change working directory to that of init
2892 Details : Starting lustre service threads may pin the working directory
2893 of the parent thread, making that filesystem busy. Threads
2894 now change to the working directory of init to avoid this.
2898 Frequency : during shutdown only
2899 Description: shutdown with a failed MDS or OST can cause unmount to hang
2900 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2905 Frequency : 2.6 only
2906 Description: chmod/chown may include an extra supplementary group
2907 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2908 supplementary group and if none is specified this is used.
2911 Bugzilla : 5479 (6816)
2913 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2914 Details : If lookup is in progress on a file that is unlinked we might try
2915 to revalidate the inode and fail in revalidate after lookup is
2916 complete and ll_file_open() enqueues the open again but
2917 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2920 Frequency : always, if lconf --abort_recovery used
2922 Description: lconf --abort_recovery fails with 'Operation not supported'
2923 Details : lconf was attempting to abort recovery on the MDT device and not
2926 ------------------------------------------------------------------------------
2928 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2933 Frequency : rare (only unsupported configurations with a node running as an
2935 Bugzilla : 6514, 5137
2936 Description: Mounting a Lustre file system on a node running as an OST could
2938 Details : OSTs now preallocates memory needed to write out data at
2939 startup, instead of when needed, to avoid having to
2940 allocate memory in possibly low memory situations.
2941 Specifically, if the file system is mounted on on OST,
2942 memory pressure could force it to try to write out data,
2943 which it needed to allocate memory to do. Due to the low
2944 memory, it would be unable to do so and the node would
2945 become unresponsive.
2947 Severity : enhancement
2949 Description: Addition of lconf --service command line option
2950 Details : lconf now accepts a '--service <arg>' option, which is
2951 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2953 Severity : enhancement
2955 Description: Failover mode is now the default for OSTs.
2956 Details : By default, OSTs will now run in failover mode. To return to
2957 the old behaviour, add '--failout' to the lmc line for OSTs.
2959 Severity : enhancement
2961 Description: Health checks are now provided for MDS and OSTs
2962 Details : Additional detailed health check information on MSD and OSTs
2963 is now provided through the procfs health_check value.
2966 Frequency : occasional, depends on IO load
2968 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2969 after numerous create/delete cycles
2970 Details : The ext3 inode allocation policy would not allocate new inodes
2971 very well on the OSTs because there are no new directories
2972 being created. Instead we look for groups with free space if
2973 the parent directories are nearly full.
2978 Description: Network or server problems during mount may cause partially
2979 mounted clients instead of returning an error.
2980 Details : The config llog parsing code may overwrite the error return
2981 code during mount error handling, returning success instead
2987 Description: MDS can fail to allocate large reply buffers
2988 Details : After long uptimes the MDS can fail to allocate large reply
2989 buffers (e.g. zconf client mount config records) due to memory
2990 fragmentation or consumption by the buffer cache. Preallocate
2991 some large reply buffers so that these replies can be sent even
2992 under memory pressure.
2996 Frequency : rare (liblustre)
2997 Description: fsx running with liblustre complained that using truncate() to
2998 extend the file doesn't work. This patch corrects that issue.
2999 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3000 ATTR_SIZE and lsm use in llu_setattr_raw.
3004 Frequency : rare, only 2.6 kernels
3005 Description: Unusual file access patterns on the MDS may result in inode
3006 data being lost in very rare circumstances.
3007 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3008 code in the 2.6 kernel caused the inode and/or EA data not to
3009 be read from disk, causing single-file corruption.
3013 Frequency : rare, only 2.6 filesystems using extents
3014 Description: Heavy concurrent write and delete load may cause data corruption.
3015 Details : It was possible under high-load situations to have an extent
3016 metadata block in the block device cache from a just-unlinked
3017 file overwrite a newly-allocated data block. We now unmap any
3018 metadata buffers that alias just-allocated data blocks.
3022 Frequency : filesystems with default stripe_count larger than 77
3023 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3024 Details : lconf specifies an inode size of 4096 bytes when the default
3025 stripe_count is larger than 77. This conflicts with the default
3026 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3027 this case to avoid pinning too much memory for large EAs.
3029 ------------------------------------------------------------------------------
3031 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3036 Frequency : rare (extremely heavy IO load with hundreds of clients)
3038 Description: Client is evicted, gets IO error writing to file
3039 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3040 caused clients to be evicted for AST timeouts. The fixes in
3041 bug 5192 mean we no longer need to have such short AST timeouts
3042 so ldlm_timeout has been increased.
3045 Frequency : occasional during --force or --failover shutdown under load
3046 Bugzilla : 5949, 4834
3047 Description: Server oops/LBUG if stopped with --force or --failover under load
3048 Details : a collection of import/export refcount and cleanup ordering
3049 issues fixed for safer force cleanup
3052 Frequency : only filesystems larger than 120 OSTs
3053 Bugzilla : 5990, 6223
3054 Description: lfs getstripe would oops on a very large filesystem
3055 Details : lov_getconfig used kfree on vmalloc'd memory
3058 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3059 Bugzilla : 6242, 6243
3060 Description: reading from files that had been truncated to a non-zero size
3061 but never opened returned no data
3062 Details : ll_file_read() reads zeros from no-object files to EOF
3067 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3068 Details : There's little harm in aborting MDS/OSS recovery and letting it
3069 try again, so I removed the LASSERT and return an error instead.
3071 Severity : enhancement
3073 Description: New debugging infrastructure for tracking down data corruption
3074 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3075 (b) cover more of the client-side code path, and (c) try to narrow
3076 down where problems occurred
3080 Bugzilla : 3819, 4364, 4397, 6313
3081 Description: Racing close and eviction MDS could cause assertion in mds_close
3082 Details : It was possible to get multiple mfd references during close and
3083 client eviction, leading to one thread referencing a freed mfd.
3085 Severity: : enhancement
3086 Bugzilla : 3262, 6359
3087 Description: Attempts to reconnect to servers are now more aggressive.
3088 Details : This builds on the enhanced upcall-less recovery that was added
3089 in 1.4.2. When trying to reconnect to servers, clients will
3090 now try each server in the failover group every 10 seconds. By
3091 default, clients would previously try one server every 25 seconds.
3096 Description: After recovery, certain operations trigger a failed
3097 assertion on a client.
3098 Details : Failing over an mds, using lconf -d --failover, while a
3099 client was doing a readdir() call would cause the client to
3100 LBUG after recovery completed and the readdir() was resent.
3102 Severity : enhancement
3104 Description: Default groups are now added by lconf
3105 Details : You can now run lconf --group <servicename> without having to
3106 manually add groups with lmc.
3109 Frequency : occasional
3111 Description: Nodes with an elan id of 0 trigger a failed assertion
3114 Frequency : always when accessing e.g. tty/console device nodes
3116 Description: tty and some other devices nodes cannot be used on lustre
3117 Details : file's private_data field is used by device data and lustre
3118 values in there got lost. New field was added to struct file to
3119 store fs-specific private data.
3122 Frequency : when exporting Lustre via NFS
3124 Description: NFSD failed occasionally when looking up a path component
3125 Details : NFSD is looking up ".." which was broken in ext3 directories
3126 that had grown large enough to become hashed.
3129 Frequency : Clusters with multiple interfaces not on the same subnet
3131 Description: Nodes will repeatedly try to reconnect to an interface which it
3132 cannot reach and report an error to the log.
3133 Details : Extra peer list entries will be created by lconf with some peers
3134 unreachable. lconf now validates the peer before adding it.
3137 Frequency : Only if a default stripe is set on the filesystem root.
3139 Description: Setting a default stripe on the filesystem root prevented the
3140 filesystem from being remounted.
3141 Details : The client was sending extra request flags in the root getattr
3142 request and did not allocate a reply buffer for the dir EA.
3145 Frequency : occasional, higher if lots of files are accessed by one client
3146 Bugzilla : 6159, 6097
3147 Description: Client trips assertion regarding lsm mismatch/magic
3148 Details : While revalidating inodes the VFS looks up inodes with ifind()
3149 and in rare cases can find an inode that is being freed.
3150 The ll_test_inode() code will free the lsm during ifind()
3151 when it finds an existing inode and then the VFS later attaches
3152 this free lsm to a new inode.
3156 Bugzilla : 6422, 7030
3157 Description: MDS deadlock between mkdir and client eviction
3158 Details : Creating a new file via mkdir or mknod (starting a transaction
3159 and getting the ns lock) can deadlock with client eviction
3160 (gets ns lock and trying to finish a synchronous transaction).
3163 Frequency : occasional
3164 Description: While starting a server, the fsfilt_ext3 module could not be
3166 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3167 kernels. Previously, lconf would still use the ext3 name
3168 when trying to load modules. Now, it will correctly use
3169 ext3 on 2.4 and ldiskfs on 2.6.
3171 Severity : enhancement
3172 Description: The default stripe count has been changed to 1
3173 Details : The interpretation of the default stripe count (0, to lfs
3174 or lmc) has been changed to mean striping across a single
3175 OST, rather than all available. For general usage we have
3176 found a stripe count of 1 or 2 works best.
3178 Severity : enhancement
3179 Description: Add support for compiling against Cray portals.
3180 Details : Conditional compiling for some areas that are different
3184 Frequency : occasional
3185 Bugzilla : 6409, 6834
3186 Description: Creating files with an explicit stripe count may lead to
3187 a failed assertion on the MDS
3188 Details : If some OSTs are full or unavailable, creating files may
3189 trigger a failed assertion on the MDS. Now, Lustre will
3190 try to use other servers or return an error to the
3194 Frequency : occasional
3196 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3197 caused lock timeout during readahead (since 1.4.2).
3198 Details : Processes doing readahead might match a lock that hasn't been
3199 granted yet if there are overlapping and conflicting lock
3200 requests. The readahead process waits on ungranted lock
3201 (original lock is CBPENDING), while OST waits for that process
3202 to cancel CBPENDING read lock and eventually evicts client.
3204 Severity : enhancement
3206 Description: Initial enabling of flock support for clients
3207 Details : Implements fcntl advisory locking and file status functions.
3208 This feature is provided as an optional mount flag (default
3209 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3210 locking are implemented yet, and those that are are not guaranteed
3211 to be completely correct in production environments.
3212 mount -t lustre -o [flock|noflock] ...
3215 Frequency : occasional
3217 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3218 assertion in the ext3 JBD (journaling) layer.
3219 Details : The b_committed_data struct is protected by the big kernel lock
3220 in 2.4 kernels, serializing journal_commit_transaction() and
3221 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3222 access to this struct. In 2.6 kernels there is finer grained
3223 locking to improve SMP performance of the JBD layer.
3227 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3232 Description: merge and cleanup kernel patches.
3235 -----------------------------------------------------------------------------
3237 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3239 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3240 versions of Lustre. Please update all servers and clients to
3241 version 1.4.2 or later at the same time. You must also run
3242 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3243 to update the configuration logs.
3245 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3246 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3247 - swab llog records of type '0' so we get proper header size/idx (5861)
3248 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3249 - fix rename of one directory over another leaking an inode (5953)
3250 - avoid SetPageDirty on 2.6 (5981)
3251 - don't re-add just-being-destroyed locks to the waiting list (5653)
3252 - when creating new directories, inherit the parent's custom
3253 striping settings if present parent (3048)
3254 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3255 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3256 - don't hold client locks on temporary worklist from l_lru (5666)
3257 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3258 - automatically evict dead clients (5921)
3259 - Update file size properly in create+truncate+fstat case (6196)
3260 - Do not unhash mountpoint dentries, do not allow removal of
3262 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3263 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3264 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3265 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3266 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3267 - Correctly handle reads of files with no objects (6243)
3268 - lctl recover will also mark a device active if deactivate used (5933)
3270 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3271 - allow --write-conf on an MDS with different nettype than client (5619)
3272 - don't write config llogs to MDS for mounts not from that MDS (5617)
3273 - lconf should create multiple TCP connections from a client (5201)
3274 - init scripts are now turned off by default; run chkconfig --on
3275 lustre and chkconfig --on lustrefs to use them
3276 - upcalls are no longer needed for clients to recover to failover
3278 - add --abort-recovery option to lconf to abort recovery on device
3280 - add support for an arbitrary number of OSTs (3026)
3281 - Quota support protocol changes.
3282 - forward compatibility changes to wire structs (6007)
3283 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3284 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3285 - improved "lustre status" script
3286 - initialize blocksize for non-regular files (6062)
3287 - added --disable-server and --disable-client configure options (5782)
3288 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3289 - Vanilla 2.4.29 support
3290 - increase maximum number of obd devices to 520 (6242)
3291 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3292 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3293 SLES 9 2.6.5 kernels
3294 - limit stripes per file to 160 (the maximum EA size) (6093)
3296 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3299 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3300 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3301 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3302 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3303 - small ext3 extents cleanups and fixes (5733)
3304 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3305 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3306 - avoid needless client->OST connect, fix handle mismatch (5317)
3307 - fix DLM error path that led to out-of-sync client, long delays (5779)
3308 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3309 - fix several locking issues related to i_size (5492,5624,5654,5672)
3310 - don't move pending lock onto export if it is already evicted (5683)
3311 - fix kernel oops when creating .foo in unlinked directory (5548)
3312 - fix deadlock in obdfilter statistics vs. object create (5811)
3313 - use time_{before,after} to avoid timer jiffies wrap (5882)
3314 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3315 - Do not leak request if server was not able to process it (5154)
3316 - If mds_open unable to find parent dir, make that negative lookup(5154)
3317 - don't create new directories with extent-mapping (5909, 5936)
3319 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3320 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3321 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3322 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3323 - support for RHEL4 kernel on i686 (5773)
3324 - provide error messages when incompatible logs are encountered (5898)
3326 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3327 * version 1.4.0.10 (1.4.1 release candidate 1)
3329 - don't keep a lock reference when lock is not granted (4238)
3330 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3331 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3332 - drop import inflight refcount on signal_completed_replay error (5255)
3333 - unlock page after async write error during send (3677)
3334 - handle missing objects in filter_preprw_read properly (5265)
3335 - no transno return for symlink open, don't save no-trasno open (3440)
3336 - don't try to complete elan receive that already failed (4012)
3337 - free RPC server reply state on error (5406)
3338 - clean up thread from ptlrpc_start_thread() on error (5160)
3339 - readahead could read extra page into cache that wasn't ejected (5388)
3340 - prevent races in class_attach/setup/cleanup/detach (5260)
3341 - don't dereference de->d_inode after l_dput of de (5458)
3342 - use "int" for stripe value returned from lock_to_stripe (5544)
3343 - mballoc allocation and error-checking fixes in 2.6 (5504)
3344 - block device patches to fix I/O request sizes in 2.6 (5482)
3345 - look up hostnames for IB nals (5602)
3346 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3347 - don't start multiple acceptors for the same port (5277)
3348 - fix incorrect LASSERT in mds_getattr_name (5635)
3349 - export a proc file for general "ping" checking (5628)
3350 - fix "lfs check" to not block when the MDS is down (5628)
3352 - service request history (4965)
3353 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3354 - create an "evict_client" /proc entry on OSTs, like the MDS has
3355 - fix mount usage message, return errors per mount(8) (5168)
3356 - change grep [] to grep "[]" in tests so they work in more UMLs
3357 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3358 - remove ancient LOV_MAGIC_V0 EA support (5047)
3359 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3360 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3361 - allow readahead allocations to fail when low on memory (5383)
3362 - mmap locking landed again, after considerable improvement (2828)
3363 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3365 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3368 - send OST transaction number in read/write reply to free req (4966)
3369 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3370 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3371 - the watchdog thread now runs as interruptible (5246)
3372 - flock/lockf fixes (but it's still disabled, pending 5135)
3373 - don't use EXT3 constants in llite code (5094)
3374 - memory shortage at startup could cause assertion (5176)
3376 - reorganization of lov code
3377 - single portals codebase
3379 - add extents/mballoc support (5025)
3380 - direct I/O reads in the obdfilter (4048)
3381 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3383 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3386 - send OST transaction number in read/write reply to free req (4966)
3387 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3388 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3389 - the watchdog thread now runs as interruptible (5246)
3390 - handle missing objects in filter_preprw_read properly (5265)
3391 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3392 - drop import inflight refcount on signal_completed_replay error (5255)
3393 - unlock page after async write error during send (3677)
3394 - return original error code on reconstructed replies (3761)
3395 - no transno return for symlink open, don't save no-trasno open (3440)
3397 - add pid to ldlm debugging output (4922)
3398 - bump the watchdog timeouts -- we can't handle 30sec yet
3399 - extra debugging for orphan dentry/inode bug (5259)
3401 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3404 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3405 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3406 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3407 - fix NULL dereference in /proc/sys/portals/routes (4827)
3408 - allow failed mdc_close() operations to be interrupted (4561)
3409 - stop precreate on OST before MDS would time out on it (4778)
3410 - don't send partial-page writes before EOF from client (4410)
3411 - discard client grant for sub-page writes on large-page clients (4520)
3412 - don't free dentries not owned by NFS code, check generation (4806)
3413 - fix lsm leak if mds_create_objects() fails (4801)
3414 - limit debug_daemon file size, always print CERROR messages (4789)
3415 - use transno after validating reply (3892)
3416 - process timed out requests if import state changes (3754)
3417 - update mtime on OST during writes, return in glimpse (4829)
3418 - add mkfsoptions to LDAP (4679)
3419 - use ->max_readahead method instead of zapping global ra (5039)
3420 - don't interrupt __l_wait_event() during strace
3422 - add software watchdogs to catch hung threads quickly (4941)
3423 - make lustrefs init script start after nfs is mounted
3424 - fix CWARN/ERROR duplication (4930)
3425 - return async write errors to application if possible (2248)
3426 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3427 - print NAL number in %x format (4645)
3428 - update barely-supported suse-2.4.21-171 series (4842)
3429 - support for sles 9 %post scripts
3430 - support for building 2.6 kernel-source packages
3431 - support for sles km_* packages
3433 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3436 - ignore -ENOENT errors in osc_destroy (3639)
3437 - notify osc create thread that OSC is being cleaned up (4600)
3438 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3439 - reconstruct ost_handle() like mds_handle() (4657)
3440 - create a new thread to do import eviction to avoid deadlock (3969)
3441 - let lconf resolve symlinked-to devices (4629)
3442 - don't unlink "objects" from directory with default EA (4554)
3443 - hold socknal file ref over connect in case target is down (4394)
3444 - allow more than 32000 subdirectories in a single directory (3244)
3445 - fix blocks count for O_DIRECT writes (3751)
3446 - OST returns ENOSPC from object create when no space left (4539)
3447 - don't send truncate RPC if file size isn't changing (4410)
3448 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3449 - bind to privileged port in socknal and tcpnal (3689)
3451 - rate limit CERROR/CWARN console message to avoid overload (4519)
3452 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3453 - basic mmap support (3918)
3454 - kernel patch series update from b1_4 (4711)
3456 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3459 - avoid crash during MDS cleanup with OST shut down (2775)
3460 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3461 - don't use bad inodes on the MDS (3744)
3462 - dynamic object preallocation to improve recovery speed (4236)
3463 - don't hold spinlock over lock dumping or change debug flags (4401)
3464 - don't zero obd_dev when it is force cleaned (3651)
3465 - print grants to console if they go negative (4431)
3466 - "lctl deactivate" will stop automatic recovery attempts (3406)
3467 - look for existing locks in ldlm_handle_enqueue() (3764)
3468 - don't resolve lock handle twice in recovery avoiding race (4401)
3469 - revalidate should check working dir is a directory (4134)
3471 - don't always mark "slow" obdfilter messages as errors (4418)
3473 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3476 - don't close LustreDB during write_conf until it is done (3860)
3477 - fix typo in lconf for_each_profile (3821)
3478 - allow dumping logs from multiple threads at one time (3820)
3479 - don't allow multiple threads in OSC recovery (3812)
3480 - fix debug_size parameters (3864)
3481 - fix mds_postrecov to initialize import for llog ctxt (3121)
3482 - replace config semaphore with spinlock (3306)
3483 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3484 - don't allow enqueue to complete on a destroyed export (3822)
3485 - down write_lock before checking llog header bitmap (3825)
3486 - recover from lock replay timeout (3764)
3487 - up llog sem before sending rpc (3652)
3488 - reduce ns lock hold times when setting kms (3267)
3489 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3490 - fix NULL deref and obd_dev leak on setup error (3312)
3491 - replace some LBUG about llog ops with error handling (3841)
3492 - don't match INVALID dentries from d_lookup and spin (3784)
3493 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3494 - fix invalid assertion in ptlrpc_set_wait (3880)
3496 - add libwrap support for the TCP acceptor (3996)
3497 - add /proc/sys/portals/routes for non-root route listing (3994)
3498 - allow setting MDS UUID in .xml (2580)
3499 - print the stack of a process that LBUGs (4228)
3501 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3504 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3505 - make sure to unset replay flag from failed open requests (3440)
3506 - if default stripe count is 0, use OST count for inode size (3636)
3507 - update parent mtime/ctime on client for create/unlink (2611)
3508 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3509 - free recovery state on server during a forced cleanup (3571)
3510 - unregister_reply for resent reqs (3063)
3511 - loop back devices mounting and status check on 2.6 (3563)
3512 - fix resource-creation race that can provoke i_size == 0 (3513)
3513 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3514 - more debugging for page-accounting assertion (3746)
3515 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3516 - avoid deadlock after precreation failure (3758)
3517 - fix race and lock order deadlock in orphan handling (3450, 3750)
3518 - add validity checks when grabbing inodes from l_ast_data (3599)
3520 - add /proc/.../recovery_status to obdfilter (3428)
3521 - lightweight CDEBUG infrastructure, debug daemon (3668)
3522 - change default OSC RPC parameters to be better on small clusters
3523 - turn off OST read cache for files smaller than 32MB
3524 - install man pages and include them in rpms (3100)
3525 - add new init script for (un)mounting lustre filesystems (2593)
3526 - run chkconfig in %post for init scripts (3701)
3527 - drop scimac NAL (unmaintained)
3529 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3532 - clean kiobufs before and after use (3485)
3533 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3534 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3535 - change init script to fail more clearly if not run as root (1528)
3536 - allow clients to reconnect during replay (1742)
3537 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3538 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3539 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3540 - fix cond_resched (3554)
3541 - only evict unfinished clients after recovery (3515)
3542 - allow bulk resend, prevent data loss (3570)
3543 - dynamic ptlrpc request buffer allocation (2102)
3544 - don't allow unlinking open directory if it isn't empty (2904)
3545 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3546 - remove extraneous obd dereference causing LASSERT failure (3334)
3547 - don't use get_cycles() when creating temp. files on the mds (3156)
3548 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3549 - handle EEXIST for set-stripe, set proper directory name (3336)
3551 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3552 - fix ksocknal_fmb_callback() error messages (2918)
3554 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3557 - don't copy lvb into (possibly NULL) reply on error (2983)
3558 - don't deref dentry after dput, don't free lvb on error (2922)
3559 - use the kms to determine writeback rpc length (2947)
3560 - increment oti_logcookies when osc is inactive (2948)
3561 - update client's i_blocks count via lvb messages (2543)
3562 - handle intent open/close of special files properly (1557)
3563 - mount MDS with errors=remount-ro, like obdfilter (2009)
3564 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3565 - don't use cancelling-locks' kms values (2947)
3566 - use highest lock extent for kms, not last one (2925)
3567 - don't dereference ERR_PTR() dentry in error handling path (3107)
3568 - fix thread race in portals_debug_dumplog() (3122)
3569 - create lprocfs device entries at setup instead of at attach (1519)
3570 - common AST error handler, don't evict client on completion race (3145)
3571 - zero nameidata in detach_mnt in 2.6 (3118)
3572 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3573 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3574 - initialize RPC timeout timer earlier for 2.6 (3219)
3575 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3576 - print nal/nid for unknown nid (3258)
3577 - additional checks for oscc recovery before doing precreate (3284)
3578 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3579 - don't crash in mdc_close for bad permissions on open (3285)
3580 - zero i_rdev for non-device files (3147)
3581 - clear page->private before handing to FS, better assertion (3119)
3582 - tune the read pipeline (3236)
3583 - fix incorrect decref of invalidated dentry (2350)
3584 - provide read-ahead stats and refine rpc in flight stats (3328)
3585 - don't hold journal transaction open across create RPC (3313)
3586 - update atime on MDS at close time (3265)
3587 - close LDAP connection when recovering to avoid server load (3315)
3588 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3589 - don't leak open file on MDS after open resend (3325)
3590 - serialize filter_precreate and filter_destroy_precreated (3329)
3591 - loop device shouldn't call sync_dev() for nul device (3092)
3592 - clear page cache after eviction (2766)
3593 - resynchronize MDS->OST in background (2824)
3594 - refuse to mount the same filesystem twice on same mountpoint (3394)
3595 - allow llmount to create routes for mounting behind routers (3320)
3596 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3597 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3598 - fix rare problem with rename on htree directories (3417)
3600 - allow default OST striping configuration per directory (1414)
3601 - fix compilation for qswnal for 2.6 kernels (3125)
3602 - increase maximum number of MDS request buffers for large systems
3603 - change liblustreapi to be useful for external progs like lfsck (3098)
3604 - increase local configuration timeout for slow disks (3353)
3605 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3607 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3610 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3611 - don't overwrite extent policy data in reply if lock was blocked (2901)
3612 - drop filter export grants atomically with removal from device (2663)
3613 - del obd_self_export from work_list in class_disconnect_exports (2908)
3614 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3615 - swab reply message in mdc_close, other PPC fixes (2464)
3616 - fix destroying of named logs (2325)
3617 - overwrite old logs when running lconf --write_conf (2264)
3618 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3619 - fix race in target_handle_connect (2898)
3620 - mds_reint_create() should take same inode create lock (2926)
3621 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3622 - don't close files for self_export to avoid uninitialized obd (2936)
3623 - allow MDS with the same name as client node (2939)
3624 - hold dentry reference for closed log files for unlink (2325)
3625 - reserve space for all logs during transactions (2059)
3626 - don't evict page beyond end of stripe extent (2925)
3627 - don't oops on a deleted current working directory (2399)
3628 - handle hard links to targets without a parent properly (2517)
3629 - don't dereference NULL lock when racing during eviction (2867)
3630 - don't grow lock extents when lots of conflicting locks (2919)
3632 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3635 - account for cache space usage on clients to avoid data loss (974)
3636 - lfsck support in lustre kernel code (2349)
3637 - reduce journal credits needed for BRW writes (2370)
3638 - orphan handling to avoid losing space on client/server crashes
3639 - ptlrpcd can be blocked, stopping ALL progress (2477)
3640 - use lock value blocks to assist in proper KMS, faster stat (1021)
3641 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3642 - recovery for initial connections (2355)
3643 - fixes for mds_cleanup_orphans (1934)
3644 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3645 - block all file creations until orphan recovery completes (1901)
3646 - client remove rq_connection from request struct (2423)
3647 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3648 - recovery timer race (2670)
3649 - mdc_close recovey bug (2532)
3650 - ptlrpc cleanup bug (2710)
3651 - mds timeout on local locks (2588)
3652 - namespace lock held during RPCs (2431)
3653 - handle interrupted sync write properly (2503)
3654 - don't try to handle a message that hasn't been replied to (2699)
3655 - client assert failure during cleanup after abort recovery (2701)
3656 - leak mdc device after failed mount (2712)
3657 - ptlrpc_check_set allows timedout requests to complete (2714)
3658 - wait for inflight reqs when ptlrpcd finishes (2710)
3659 - make sure unregistered services are removed from the srv_list
3660 - reset bulk XID's when resending them (caught by 1138 test)
3661 - unregister_bulk after timeout
3662 - fix lconf error (2694)
3663 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3664 - readahead locks pages, leaves pending causing memory pressure (2673)
3665 - increase OST request buffers to 4096 on large machines (2729)
3666 - fix up permission of existing directories in simple_mkdir (2661)
3667 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3668 - don't assign transno to errored transactions (2742)
3669 - don't delete objects on OST if given a bogus objid from MDS (2751)
3670 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3671 - if rq_no_resend, then timeout request after recovery (2432)
3672 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3673 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3674 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3675 - MDS recovery shouldn't delete objects that it has given out (2730)
3676 - if enqueue arrives after completion, don't clobber LVB (2819)
3677 - don't unlock pages twice when trigger_group_io returns error (2814)
3678 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3679 - don't write pages to disk if there was an error (1450)
3680 - don't ping imports that have recovery disabled (2676)
3681 - take buffered bytes into account when balancing socknal conn (2817)
3682 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3683 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3684 - remove little-endian swabbing of llog records (1987)
3685 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3686 - retry reposting request buffers if they fail (1191)
3687 - grow extent at grant time to avoid granting a revoked lock (2809)
3688 - lock revoke doesn't evict page if covered by a second lock (2765)
3689 - disable VM readahead to avoid reading outside lock extents (2805)
3691 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3692 - updated kernel patches for hp-2.4.20 kernel (2681)
3694 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3697 - fix truncated write corruption (2366)
3698 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3700 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3701 - ensure len(uuid) < 37 in lmc (1171)
3702 - fix ia64 OOPS in llog_test (2255)
3703 - zero end of page at obdfilter for partial page writes (2648)
3704 - don't leave stale dentries around after renames (bug 2428)
3705 - fix timeouts when evicting a client with a single lock held (2642)
3706 - set deadline for the initial HELLO message to drain (2634)
3707 - print out dotted-quad IP addresses in the socknal (2302)
3709 - additional debugging for MDS client eviction problem (2443)
3710 - fix mkfsoptions support for osts (2603, 2604)
3712 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3715 - add series for the vanilla 2.6.0 kernel
3716 - add series for the vanilla 2.4.24 kernel
3717 - add series for a cray x86/64 UL kernel drop
3718 - fix xattr patches for the vanilla 2.4.19 series
3720 - generate true UUIDs in lmc (1171)
3721 - have portals stack dumping break in UML (2466)
3722 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3723 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3724 - wake callback waiting threads on client eviction (2460)
3725 - Add --ptldebug and --subsystem to lmc (1719)
3726 - update assertion to allow safe interrupt allocation
3727 - set rq_no_resend for cancel requests (2432)
3728 - recalculate ptlrpcd timeout after resend (2494)
3729 - call vfs_rmdir when removing pending directories (2368)
3730 - fix renaming a file to itself (2429)
3731 - lmc creates a default one-stripe lov (2454)
3732 - expand procfs space to handle large clusters (2326)
3733 - increase UML stack to avoid overflow
3734 - update lconf's list of debug and subsystem masks
3735 - fix lfs find --obd (2510)
3736 - /proc tunable for disabling filter read caching (2591)
3737 - stop rpm packages from altering slapd.conf (2301)
3738 - disable nagle in the socknal under 0conf (2578)
3739 - choose mds inode size based on stripe count (2572)
3740 - fix kernel-source rpm problems (2516)
3742 - add --disable-doc to avoid pdf generation (2421)
3743 - update documentation, tests, type-os, comments
3744 - avoid format warnings on ia64
3745 - remove the TOE NAL
3746 - tiny code cleanups by removing unused fields
3748 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3751 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3752 - remove the most blatant lies from BUILDING file (2371)
3753 - change default debug level to reasonable production setting
3754 - reduce client side cache size to reduce cache flush time
3755 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3756 - make TCP zerocopy and pinger support enabled by default (2476)
3757 - sync writes completed after process exits caused crashes (2319)
3758 - maintain correct mount count on the MDS (2356)
3759 - backout 1557, because 2316 wasn't really fixed
3760 - better file I/O statistics gathering in /proc
3761 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3762 - another kernel patch to fix zero-copy TCP function export
3763 - don't take duplicate lock when processing re-sent getattr (2420)
3764 - lctl uses obd_self_export instead of creating new conn (2353)
3765 - MDS/OST recovery case which requires object creation asserted (2425)
3766 - move lfs from /usr/sbin to /usr/bin in packages
3767 - fix race between mds_client_add and mds_client_free (2417)
3768 - use kmalloc instead of slabs in portals (2430)
3769 - don't create duplicate records when a failover MDS is present (2442)
3770 - remove unnecessary mount age check (2332)
3771 - don't remove directory inodes from locks prematurely (2451)
3772 - don't break if MDS service name is the same as hostname (2103)
3773 - fix races in client write RPC generation when cache full (2482)
3775 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3778 - remove now-unused request->rq_obd (278)
3779 - if an allocation fails, print out how much memory we've used (1933)
3780 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3781 - add the "configurable stack size" patch to most series files (1256)
3782 - ability to write large log records, for 100+ OST configs (2306)
3783 - fix NULL deref when filter_prep fails (2314)
3784 - fix operator precedence error in filter_sync
3785 - dynamic allocation of socknal TX descriptors (2315)
3786 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3787 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3788 - ensure that CWARN messages in Portals always get to the syslog
3789 - __init/__exit are not for prototype decls (ldlm_init/exit)
3790 - x86-64 compile warning fixes
3791 - fix gateway LMC keyword conflict (2318)
3792 - fix MDS lock inversions in getattr/reint paths (1844)
3793 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3794 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3795 - DLM race condition prevented some lock evictions (2328)
3796 - ENOMEM detection and retry on socknal sends (2230)
3797 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3798 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3799 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3800 - better error messages when a client is rejected during recovery (1505)
3801 - avoid cancelling locks which were never granted, after failure (2330)
3802 - fix i_sem/journal inversion in mds_client_add (2333)
3803 - fix truncate/getattr lock cycle deadlock (2334)
3804 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3805 - fix two rare exit paths which could leak an l_lock() ref (2321)
3806 - fencepost error in MDS/OST orphan recovery (2226)
3807 - make log record alignment 8 bytes (1988)
3808 - lstripe now fails when requested offset > ost_count (2237)
3809 - ensure that all kernel series have a complete list.h (1607)
3810 - fix crashes in special-file operations (2316)
3811 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3813 - allow configurable automake binary, for testing new versions
3814 - small update to the lfs documentation
3816 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3818 * fix negative export reference count in fsfilt_sync (2312)
3820 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3821 * release candidate 0.9.1
3823 - orphans are moved into the PENDING directory for possible recovery
3824 - replayed opens now open by fid for orphan/rename safety (1042)
3825 - last close of an orphan inode generates a transno (683)
3826 - chdir() and mount() now pin the directory entry (1020)
3827 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3828 - discard very old requests without processing them (1502)
3829 - remove some common, well-understood CERRORs (1505)
3830 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3831 - clear "grant" flags in OST replies until OST grant code lands (1644)
3832 - fix read performance by not clobbering i_blksize on client (1598)
3833 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3834 - make LRU size a /proc tunable, clears locks when reduced (707)
3835 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3836 - server locks take references on exports now (1558)
3837 - build fixes for 2.4.20-rh trees (1663)
3838 - return an error from lov_create if all OSCs are inactive (1751)
3839 - fix import levels when a reconnect happens without a timeout (1597)
3840 - exit early from mds_open if we get a lookup error (1749)
3841 - partial page read at EOF wouldn't wait for disk before sending (1642)
3842 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3843 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3844 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3845 - write/truncate lock inversion (1639)
3846 - Don't auto-load obdclass, portals modules during cleanup (1495)
3847 - fix timestamps from jumping to "now" (1763)
3848 - extra journal assertions (1648)
3849 - add an extra multiunlink test (1771)
3850 - fix read_record/write_record API (1776)
3851 - fix leak of offset_extent, possible incorrect i_size later (1772)
3852 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3853 - Debugging for the kqswnal_get_idle_tx problems (1820)
3854 - Allow recovery to be attempted multiple times (1536)
3855 - Write out MDS last_rcvd file after it is first created (1600)
3856 - Fix tx_descriptor leak in failed transmit situations (1827)
3857 - ext3 journaling fixes for assertion failure after IO error (1871)
3858 - class_export_put() on freed export after completion AST error (1896)
3859 - Fix revalidate looping in VFS (1322)
3860 - Don't access a freed export during MDS_REINT timeout (1521)
3861 - Add open-unlink recovery support on the MDS (1673,1764)
3862 - Return an error if no MDS data was read from last_rcvd (1946)
3863 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3864 - Don't setup a disk device that doesn't match exported UUID (317)
3865 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3866 - avoid committing NULL handle in force close
3867 - local.sh is now a one-stripe LOV configuration
3868 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3869 - fix ext3 htree duplicate directory entry corruption (1516)
3870 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3871 - update to new LOV EA format (2097)
3872 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3874 2003-06-15 Phil Schwan <phil@clusterfs.com>
3877 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3878 - per-import/export recovery handling (958, 931, 959)
3879 - multiple last-rcvd slots, for serving multiple FSes (949)
3880 - connections are again shared between multiple imp/exports (963, 964)
3881 - "umount -f" would hang if any requests needed to be sent (393, 978)
3882 - avoid pinning large req buffer by copying for queued messages (989)
3883 - add "uuid" to "lctl device" command to help upcalls (991)
3884 - "open" RPCs with transnos would confuse recovery counters (1037)
3885 - do proper endian conversion of all wire messages (288, 340, 891)
3886 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3887 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3888 - fix ptlrpc_connection leak in target_handle_connect (1174)
3889 - fix import refcounting bug in OST and MDS cleanup (1134)
3890 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3891 - fix very unlikely obd_types race condition (501)
3892 - remove osc_open hack for echo_client (1187)
3893 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3894 - a failure in read_inode2 leads to deadlock (1139)
3895 - cancel ack-locks as soon as transaction is committed (1072)
3896 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3897 - make sure to commitrw after any preprw to avoid deadlock (1162)
3898 - failing to execute a file in a lustre FS would lock inode (1203)
3899 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3900 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3901 - obd_uuid2tgt would walk off the end of the list (1255)
3902 - on IA64 the osc would give portals incorrect bulk size (1258)
3903 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3904 - fix lock inversion caused by new llite matching code (1282)
3905 - limit the number of dirty pages on a client to 10MB (1286)
3906 - timed out locks were not being corrected cancelled (1289)
3907 - fix O_DIRECT above 4GB on IA-32 (1292)
3908 * major user-visible changes
3909 - fail out/fail over policy now controlled by the upcall (993)
3911 - add OBD_PING to check server availability and failure (954)
3912 - lustre messages are now sent in sending host order (288, 340, 891)
3913 - add eadatalen to MDS getattr reply (340)
3914 - OST read replies may contain second buffer, with per-page status (593)
3916 2003-03-11 Phil Schwan <phil@clusterfs.com>
3919 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3920 - failed open()s could cause deadlock; fixed (867, 869)
3921 - stop cancelling OST locks when files are closed (481)
3922 - overlapping XID spaces caused network corruption (851, 853)
3923 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3924 - setattr_raw added, to do single-RPC, server-side setattrs
3925 - lmc/lconf syntax change for OST UUIDs
3926 - fix crashy race condition between ptlrpc_free_req and osc_close
3927 - don't use request in mdc_enqueue if we hit a timeout (889)
3928 - don't set the inode i_size for regular files from the MDS (896)
3929 - handle out of order completion AST (842)
3930 - don't LBUG if a lock request times out after receiving AST (913)
3931 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3932 - if a bad lock AST arrives, send an error instead of dropping entirely
3933 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3934 - fix leak in bulk IO when only partially completed (899, 900, 926)
3935 - fix O_DIRECT for ia64 (55)
3936 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3937 - C-z after timeout could hang a process forever; fixed (977)
3939 - client-side I/O cache (678, 924, 929, 941, 970)
3941 - READPAGE and SETATTRs which don't take server-side locks get
3944 2003-02-11 Phil Schwan <phil@clusterfs.com>
3947 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3948 - Fix fsfilt_extN_readpage() to read a full page of directory
3949 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3950 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3951 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3952 - Most intent operations were removed, in favour of a new RPC mode
3953 that does a single RPC to the server and bypasses most of the VFS
3954 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3955 - Aggressively cancel local locks on DLM servers
3956 - mds_reint_unlink sends EA to the client if it's the last nlink.
3957 client uses that EA to unlink OST objects.
3958 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3959 - recursive symlinks were fixed (439)
3960 - fixed NULL deref in DEBUG_REQ
3961 - filter_update_lastobjid no longer calls sync, which annoyed extN
3962 - fixed multi-client small-writes to a single file problem (445)
3963 - fixed mtime updates during file writes (607)
3964 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3965 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3966 - fixed Linux OST size reporting problem (444, 656)
3967 - OST now updates object mtime with writes or setattr (607, 619)
3968 - client verifies file size before zeroing page past EOF (445)
3969 - OST now writes last allocated objid to disk with allocation (108)
3970 - LOV on echo now works (409)
3972 - mds_reint_unlink sends a new buffer, with the EA included. this
3973 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3974 set if a regular file was being unlinked, and it was the last link
3975 - use PtlGet from the target for bulk writes (315)
3976 - OST now updates object mtime with writes or setattr (607, 619)
3977 - LDLM now has a grant-time callback to revalidate locked items, if
3979 - Many MDS operations were reorganized to combat race conditions
3981 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3982 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3984 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3987 - Fully reactivate OST imports after reconnection (512, others)
3988 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3989 - More graceful error handling for truncating on dead OST (515)
3990 - Don't error out unless we're actually accessing dead stripes (474)
3991 - Fix garbage sizes when stripes are missing (410)
3992 - LRU counters were broken, causing constant lock purge (433, 432)
3993 - garbage on read from stripes with failed OSTs (441)
3994 - mark OSCs as active before reconnecting during recovery (438)
3995 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3996 - lfind did not preserve OST order in output (443)
3997 - symlinks cause hung clients, incorrect data (439)
3998 - stop dereferencing request after dropping refcount (457)
3999 - don't LASSERT(spin_is_locked) on non-SMP (455)
4000 - fixes for many rename() bugs
4001 - fstat didn't correctly synchronize attributes (399)
4002 - server must handle lock cancellation during blocking AST prep (487)
4003 - bulk descriptors were free()d too soon (511)
4004 - fix paths in lconf, which would load incorrect modules (451, 507)
4005 - fix confusing lconf 'host not found' error message (386)
4006 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4007 - fix race condition in mdc_blocking_ast() for inode access (526)
4008 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4009 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4010 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4011 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4012 - fix lstripe in several ways (kernel side) (527)
4013 - fix request leak in ldlm_cli_enqueue (262)
4014 - incorrect OSC was marked inactive after OST failure
4015 - call mds_fs_cleanup before unmounting filesystem (524)
4016 - fix races between taking ns_lock and ldlm_lock_change_resource
4017 - fix races updating LOV export open file list
4018 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4019 - fix recovery NULL deref in ldlm_cli_cancel_unused
4020 - fix some DLM races by using new hash table for lock handles (419)
4021 - permit the client to specify desired inodes, at replay
4022 - duplicate requests when we queue them for replay reintegration
4023 - fix last_rcvd offset calculation
4024 - sync after each recovered transaction, so we always make progress
4025 - never, not always, ERESTART requests without transnos
4026 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4028 - skip replay if the MDS says that the client is already connected
4029 - don't check for a recovery-enabled export to match lctl's UUID
4030 - don't INC_USE_COUNT for phantom exports
4031 - don't crash when cleaning up phantom exports (567)
4032 - don't double-finish or set replay data for errored mdc_open requests
4033 - abort requests when they time out, so we don't get old replies
4034 - send/receive replies for AST messages again
4035 - if the client says that it doesn't have the lock, cancel it on the
4037 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4038 mark it as destroyed, it will all work out when decref is called
4039 - fix module use counts (22, 581)
4041 - ASTs now expect a reply (server cancels lock on error reply)
4043 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4046 - fix many simultaneous client startup (392)
4047 - fix dentry->d_it clobbering
4048 - credentials weren't being shipped for readdir/getattr operations
4049 - remove invalid assertions triggered during some concurrent MD
4051 - proper Lustre versions added (336, 389)
4052 - fix memory leak for create error case (398)
4053 - fix LOV locking bug that would get cli/srv out of sync
4054 - fix echo client over LOV (409)
4055 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4056 - fix double-O_EXCL intent crash (424)
4057 - avoid sending multiple lock CANCELs (352)
4059 - MDS can do multi-client recovery (modulo bugs in new code)
4061 - many updates, edits, cleanups
4063 2002-11-18 Phil Schwan <phil@clusterfs.com>
4066 - fix null d_it dereference (346)
4067 - fix full OST/dbench hang (333)
4068 - fix permission problem with file removal (286)
4069 - fix removal of OSCs from LOV when they fail
4070 - fix NULL deref during bulk timeout (214)
4071 - fix problems related to multiple filesystems on one MDS (241)
4072 - fixed serious subtle metadata locking bugs
4073 - free locks on clients when inodes are removed due to memory
4075 - fix inode pointer in lock data (285)
4076 - partial support for multiple MDS on a single host (241)
4077 - data locks weren't cancelled at clear_inode time (290, 311)
4078 - intent locks could lead to unbounded lock growth (205)
4079 - added a maximum lock count, an LRU list, and a flusher
4080 - fix multiple rename (365)
4081 - properly abstracted the echo client
4082 - OSC locked 1 byte too many; fixed
4083 - rewrote brw callback code:
4084 - fixed recovery bugs related to LOVs (306)
4085 - fixed too-many-pages-in-one-write crash (191)
4086 - fixed (again) crash in sync_io_timeout (214)
4087 - probably fixed callback-related race (385)
4089 - Add capability to MDS protocol
4090 - LDLM cancellations and callbacks on different portals
4092 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4095 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4096 - defer open object destruction to close time (601981, 138)
4097 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4098 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4099 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4100 - don't call obd_setattr in ll_file_release for destroyed objects
4101 * protocol change to lustre_msg: move |version| and add |flags|
4102 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4103 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4104 * added replay of create, unlink, link and rename operations during
4105 MDS failover; recovery should be much more robust now
4106 * remove failed OSCs from LOVs (only lov_create uses this so far)
4107 * the lustre-HOWTO was brought (more) up to date (582544)
4109 2002-10-23 Phil Schwan <phil@clusterfs.com>
4112 - in-use dentries weren't being reused properly (617851)
4113 - prevent multiple LDLM setup (599178)
4114 - fix LOV size calculations for truncate (617853)
4115 - fix client handling of MDS intent errors (POSIX)
4116 - fix permission bug in lovstripe.c test (624321)
4117 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4118 - truncate past end of file could corrupt data
4119 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4120 - a race in recovery could return ETIMEDOUT to apps (623947)
4121 - building outside the source directory was fixed
4122 * the lustre-HOWTO was brought (more) up to date (582544)
4123 * major progress was made on recovery functionality
4125 2002-10-10 Phil Schwan <phil@clusterfs.com>
4128 - recovery deadlock fix
4129 - rm -rf causes LBUG fix (617817)
4130 - file open by multiple tasks fix (618962)
4131 - directory permissions bugs (602707 and 620007)
4132 - journal_stop fixed with locking (611313)
4133 - O_APPEND failures resolved (618273, perhaps 614459)
4134 - lconf PATH fix (619770)
4135 - IA64 build fix (621450)
4136 - RPC buffer sizes scale with amount of memory
4138 2002-10-01 Phil Schwan <phil@clusterfs.com>
4141 - locks would be cancelled without throwing away data pages,
4142 resulting in inconsistent data (605627)
4143 - inode attributes were not always being refreshed (605627, 612449)
4144 - lconf now continues to cleanup after lctl reports an error
4145 - MDS now enforces user permissions (602707)
4146 - lprocfs cleanup fixed, but not yet enabled (614157)
4147 - fixed infinite server hang, should a client not respond to an AST
4148 - avoid going into recovery if user calls readlink() with a buffer
4149 that's too small (613941)
4150 - AST RPCs no longer require replies (614867) -- this may be changed
4151 - don't crash server if client sends an IOV that's too big (611336)
4152 - fixed lock conversion deadlock (611892)
4153 - fixed the following of symlinks (614622)
4154 * recovery: the server can remove locks from a client that dies, other
4155 clients can make progress
4156 * more extN patch fixes
4157 * compile-time configurable ptlrpc buffer allocations
4159 - collaborative read cache document
4160 - Lustre Lite Performance CDR document-in-progress
4162 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4165 - fix typo in patch-2.4.18
4167 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4170 - clear ptlrpc request each time in handle_incoming_request()
4171 - unlink of files now destroys the object on the OST
4173 2002-09-19 Peter Braam <braam@clusterfs.com>
4175 * add hard link support
4176 * change obdfile creation method
4177 * kernel patch changed
4179 2002-09-19 Peter Braam <braam@clusterfs.com>
4182 - stack overflow bug in extN fixed
4184 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4186 * documentation updates
4187 - add man pages for config tools
4188 - update tests/README to describe testing with new config tools
4189 - finish metadata API descriptions
4190 * bug fixes and cleanups
4191 - statfs workaround for 16TB limit
4192 - LOV stripe allocation improved, can stripe on subset of OSTs
4193 - LOV file size/IO offset was wrong for files > 4GB in size
4194 - object EA data was being dropped, caused files to be unreadable
4195 - memory overflow with non-LOV OST caused memory corruption
4196 - fixed regression tests to work with new config tools, obdfilter
4197 - fixed bug when directory size became larger than 1 block
4198 - fixed bug (for single client case) when PWD was deleted
4199 - invalidate local directory pages when doing intent-based ops
4200 - avoid LDLM oops when lock callback contained bad data
4202 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4204 * documentation updates
4205 * bug fixes and cleanups
4206 - configuration tools
4209 - 64-bit compile warnings
4210 - 64-bit internal statfs data
4212 * test_brw on persistent OST devices
4214 * lprocfs (disabled)
4216 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4218 * documentation updates
4219 * bug fixes and cleanups
4220 * configuration tools
4222 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4225 * many small fixes to 0_5_4
4226 * io/network handling
4227 * thinkos in MDS operations
4229 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4232 * crucial basic fixes to 0.5.3
4233 * IOR, Iozone work over Elan
4236 2002-08-07 Phil Schwan <phil@clusterfs.com>
4237 * version 0_5_3, our first alpha
4238 * we use the new Portals iovs
4239 * documentation updates
4240 * bug fixes and cleanups
4241 * small changes in the DLM wire protocol
4243 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4244 * version 0_5_1 with some initial stability,
4245 * locking on MD and file I/O.
4246 * documentation updates
4247 * several bug fixes since 0.5.0
4248 * small changes in wire protocol
4250 2002-07-18 Phil Schwan <phil@clusterfs.com>
4252 * delivered as Lustre Light Alpha
4253 * fixed a crash after handling invalid MDS requests
4254 * fixed directory pages for architectures with non-4k pages sizes
4256 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4257 * release version v0_4_4
4258 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4260 * Fixes a number of interruption problems with OST operations.
4261 * Update documentation for portals header changes
4262 * Move all wire protocol structs/defines to lustre_idl.h
4263 * Fixes symlink length bug.
4264 * Add tcpdump to repository.
4266 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4267 * release version v0_4_3
4268 * Fixes statfs for inodes on extN.
4269 * Fixes bug in runtests which would delete /etc/hosts.
4270 * Use 64-bit object IDs wherever possible (not into VFS though)
4271 Remove ost_get_info, which is unused by lustre, and out of date.
4273 2002-07-03 Peter Braam <braam@clusterfs.com>
4274 * release version v0_4_2 Fixes a lookup error (type not passed)
4275 * move forward to head of Portals
4276 * move forward to latest Lustre kernel
4278 2002-06-25 Peter Braam <braam@clusterfs.com>
4279 * release version v0_4_1. Hopefully stable on single node use.