3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.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-cfs5
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 Severity : enhancement
17 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
19 Severity : enhancement
21 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
25 Frequency : rare, at shutdown
26 Description: access already free / zero obd_namespace.
27 Details : if client_disconnect_export was called without force flag set,
28 and exist connect request in flight, this can produce access to
29 NULL pointer (or already free pointer) when connect_interpret
30 store ocd flags in obd_namespace.
34 Frequency : only at startup
35 Description: not alloc memory with spinlock held.
36 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
39 Severity : enhancement
41 Description: make lustre randomly fail allocating memory
42 Details : Make lustre randomly failed allocating memory for testing purpose.
44 Severity : enhancement
46 Description: lost problems with lov objid file
47 Details : Fixes some scability and access to not inited memory problems
48 in work with lov objdid file.
53 Description: lfs find does not continue on file error
54 Details : Continue other files processing when a file/dir is absent.
58 Description: Inconsistent usage of lustre_pack_reply()
59 Details : Standardize the usage of lustre_pack_reply() such that it
60 always generate a CERROR on failure.
63 Frequency : occasional
65 Description: MDS hang or stay in waiting lock
66 Details : If client receive lock with CBPENDING flag ldlm need send lock
67 cancel as separate rpc, to avoid situation when cancel request
68 can't processed due all i/o threads stay in wait lock.
72 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
76 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
80 Description: Update to SLES9 kernel-2.6.5-7.287.3.
84 Description: Update to RHEL4 latest kernel.
86 Severity : enhancement
88 Description: Build SLES10 patchless client fails
89 Details : The configure was broken by run ./configure with
90 --with-linux-obj=.... argument for patchless client. When the
91 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
92 can't search header adequately. Use absolute path such as
93 -I($LINUX)/include instead.
97 Description: interrupt oig_wait produce painc on resend.
98 Details : brw_redo_request can be used for resend requests from ptlrpcd and
99 private set, and this produce situation when rq_ptlrpcd_data not
100 copyed to new allocated request and triggered LBUG on assert
101 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
102 ptlrpcd set if request is changed and can be safety changed to use
107 Description: LASSERT_{REQ,REP}SWAB macros are buggy
108 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
109 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
110 these macros and replace them with their difinitions instead.
114 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
115 Details : Update which_patch & target file for SLES10 latest kernel.
118 Description: Remove client patches from SLES 10 kernel.
119 Details : This causes SLES 10 clients to behave as patchless clients
120 even on a Lustre-patched (server) kernel.
122 Severity : enhancement
124 Description: self-adjustable client's lru lists
125 Details : use adaptive algorithm for managing client cached locks lru
126 lists according to current server load, other client's work
127 pattern, memory activities, etc. Both, server and client
128 side namespaces provide number of proc tunables for controlling
131 Severity : enhancement
133 Description: light-weight GSS support
134 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
135 and reduce performance overhead.
137 Severity : enhancement
139 Description: Linux keyring support
140 Details : Support using service of Linux keyring for Lustre GSS internal
141 context refresh/cache mechanism.
145 Description: Fix errors in lfs documentation
146 Details : Fixes man pages
150 Description: don't use GFP_* in generic Lustre code.
151 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
155 Description: obdclass is limited by single OBD_ALLOC(idarray)
156 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
160 Description: warnings with build patchless client with vanila 2.6.19 and up
161 Details : change old ctl_table style and replace ctl_table/ctl_table_header
162 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
166 Frequency : Only for SLES
167 Description: sanity_quota fail test_1
168 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
169 file, which makes TSTID[2] inunique.
173 Frequency : Only for SLES9
174 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
175 Details : Update target/ChangeLog/which_patch .
179 Frequency : Only for test_10 in sanity-quota.sh
180 Description: a bug in quota test code
181 Details : checking a test flag in a wrong place
185 Frequency : Only for test_13 in sanity-quota.sh
186 Description: a bug in quota test script
187 Details : shouldn't check overfull allocated quota
191 Description: jbd statistics
192 Details : Port older jbd statistics patch for sles10
194 Severity : enhancement
196 Description: Add printing inode info into message about error in writepage.
201 Description: reply_lock_interpret crash due to race with it and lock cancel.
202 Details : Do not replay locks that are being cancelled. Do not reference
203 locks by their address during replay, just by their handle.
207 Frequency : When flocks are used.
208 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
209 Details : Flock locks might destroy just granted lock if it could be merged
210 with another existing flock, this is done in completion handler,
211 so teach ldlm_cli_enquque_fini that this is a valid case for
217 Description: Oops in read and write path when failing to allocate lock.
218 Details : Check if lock allocation failed and return error back.
222 Description: lstripe command fails for valid OST index
223 Details : The stripe offset is compared to lov->desc.ld_tgt_count
224 instead of lov->desc.ld_active_tgt_count.
228 Description: sanity.sh failed test 103
229 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
234 Description: df doesn't work properly if diskfs blocksize != 4K
235 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
239 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
240 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
244 Description: obd_health_check_timeout too short
245 Details : set obd_health_check_timeout as 1.5x of obd_timeout
249 Description: llapi_file_create() does not allow some changes
250 Details : add llapi_file_open() that allows specifying the mode and
251 open flags, and also returns an open file handle.
255 Description: merge and cleanup kernel patches.
256 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
260 Description: Add journal checksum support.(Kernel part)
261 Details : The journal checksum feature adds two new flags i.e
262 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
263 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
264 indicates that the commit block contains the checksum for
265 the blocks described by the descriptor blocks. Now commit
266 record can be sent to disk without waiting for descriptor
267 blocks to be written to disk. This behavior is controlled
268 using JBD2_FEATURE_ASYNC_COMMIT flag.
272 Description: OSS needs mutliple precreate threads
273 Details : Add ability to start more than one create thread per OSS.
277 Description: Sanity.sh test_65k failed: setstripe should have succeeded
278 Details : execute lfs setstripe on client
282 Description: mds_obd_create error creating tmp object
283 Details : When the user sets quota on root, llog will be affected and can't
284 create files and write files.
287 Frequency : Always on ia64 patchless client, and possibly others.
289 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
290 Details : This allows the patchless client to be loaded on architectures
295 Description: RedHat Update kernel for RHEL5
296 Details : Add the kernel config file for RHEL5.
300 Description: RedHat Update kernel for RHEL5
301 Details : Modify the kernel config file more closer RHEL5.
305 Description: Build failure against Centos5 (RHEL5)
306 Details : Define PAGE_SIZE when it isn't present.
310 Description: test_7 on sanity-quota.sh can't be used on separate machines
311 Details : add facet to handle it
315 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
316 Details : with NFS, the anon dentry's parent was set to itself in
317 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
318 corresponding dentry other than use rec->ur_name.
320 Severity : enhancement
322 Description: lfs setstripe enhancement
323 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
327 Description: enable data checksumming by default
328 Details : enable checksum by default, allow --disable-checksum
329 configure option and "-o nochecksum" mount option. Checksums
330 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
334 Description: lustre support for RHEL5
335 Details : Add support for RHEL5.
339 Description: Client eviction due to failover config
340 Details : after a connection loss, the lustre client should attempt to
341 reconnect to the last active server first before trying the
342 other potential connections.
346 Description: when mds and osts use different quota unit(32bit and 64bit),
347 quota will be released repeatly.
348 Details : void sending multiple quota reqs to mds, which will keep the
349 status between the reqs.
353 Description: osts not allocated evenly to files
354 Details : change the condition to increase offset_idx
357 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
359 Description: Data corruption for OSTs that are formatted larger than 2TB
361 Details : When generating the bio request for lustre file writes the
362 sector number would overflow a temporary variable before being
363 used for the IO. The data reads correctly from Lustre (which
364 will overflow in a similar manner) but other file data or
365 filesystem metadata may be corrupted in some cases.
369 Description: Tune the kernel for good SCSI performance.
370 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
371 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
376 Description: rewrite ext2-derived code in obdclass/uuid.c
377 Details : rewrite inherited code (uuid parsing code from ext2 utils)
378 from scratch preserving functionality.
382 Description: "lfs find -obd UUID" prints directories
383 Details : "lfs find -obd UUID" will return all directory names instead
384 of just file names. It is incorrect because the directories
385 do not reside on the OSTs.
389 Description: use i_size_read and i_size_write in 2.6 port
390 Details : replace inode->i_size access with i_size_read/write()
393 Frequency : when removing large files
395 Description: scheduling issue during removal of large Lustre files
396 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
397 It causes scheduling issues when removing large files (17TB in the
401 Frequency : only with liblustre clients on XT3
403 Description: evictions taking too long
404 Details : allow llrd to evict clients directly on OSTs
407 Frequency : only on ppc
409 Description: /proc/fs/lustre/devices broken on ppc
410 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
411 We should make sure the type of 'index' is loff_t and then cast
412 explicitly as needed below. Do not assign an explicitly cast
416 Frequency : only for rhel5
418 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
419 Details : Modify the target file & which_kernel.
421 Severity : enhancement
423 Description: omit set fsid for export NFS
424 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
425 as FS_REQUIRES_DEV for avoid problems with generate fsid.
429 Frequency : Always, for kernels after 2.6.16
430 Description: Fix warning idr_remove called for id=.. which is not allocated.
431 Details : Last kernels save old s_dev before kill super and not allow
432 to restore from callback - restore it before call kill_anon_super.
436 Description: buffer overruns could theoretically occur
437 Details : llapi_semantic_traverse() modifies the "path" argument by
438 appending values to the end of the origin string, and a
439 overrun may occur. Adding buffer overrun check in liblustreapi.
443 Description: Fix error on 'ls .' at the top of the Lustre mount.
444 Details : Don't revalidate dentry if it is a root dentry.
448 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
449 Details : Modify vm-tunables-rhel4.patch.
453 Description: Kernel config for 2.6.18-vanilla.
454 Details : Modify targets/2.6-vanilla.target.in.
455 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
456 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
457 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
458 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
462 Description: improve handling recoverable errors
463 Details : if request processig with error which can be recoverable on server
464 request should be resend, otherwise page released from cache and
469 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
470 Details : Modify target file & which_patch.
471 A flaw was found in the IA32 system call emulation provided
472 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
473 value could be stored in the %RAX register, which could trigger an
474 out-of-bounds system call table access. An untrusted local user
475 could exploit this flaw to run code in the kernel
476 (ie a root privilege escalation). (CVE-2007-4573).
480 Description: O_DIRECT bypasses client statistics.
481 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
482 read_ahead_stats were updated. copied the stats section from
483 osc_send_oap_rpc() into async_internal().
487 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
491 Description: change order of libsysio includes
492 Details : '#include sysio.h' should always come before '#include xtio.h'
496 Description: handle "serious error: objid * already exists" more gracefully
497 Details : If LAST_ID value on disk is smaller than the objects existing in
498 the O/0/d* directories, it indicates disk corruption and causes an
499 LBUG(). If the object is 0-length, then we should use the existing
500 object. This will help to avoid a full fsck in most cases.
502 Severity : enhancement
504 Description: adapt the lustre_config script to support the upgrade case
505 Details : Add "-u" option for lustre_config script to support upgrading 1.4
506 server targets to 1.6 in parallel.
510 Description: To avoid grant space > avaible space when the disk is almost
511 full. Without this patch you might see the error "grant XXXX >
512 available" or some LBUG about grant, when the disk is almost
514 Details : In filter_check_grant, for non_grant cache write, we should
515 check the left space by if (*left > ungranted + bytes), instead
516 of (*left > ungranted), because only we are sure the left space
517 is enough for another "bytes", then the ungrant space should be
518 increase. In client, we should update cl_avail_grant only there
519 is OBD_MD_FLGRANT in the reply.
523 Description: Update RHEL 4 kernel to fix local root privilege escalation.
524 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
525 described in CVE-2007-4573. This problem could allow untrusted
526 local users to gain root access.
529 Frequency : when using O_DIRECT and quotas
531 Description: Incorrect file ownership on O_DIRECT output files
532 Details : block usage reported by 'lfs quota' does not take into account
533 files that have been written with O_DIRECT.
538 Description: touch file failed when fs is not full
539 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
540 otherwise we can get ENOSP while fs is not full.
544 Description: parallel lock callbacks
545 Details : Instead of sending blocking and completion callbacks as separated
546 requests, adding them to a set and sending in parallel.
549 Frequency : only for Cray XT3
550 Bugzilla : 12829/13455
551 Description: Changing primary group doesn't change the group lustre assigns to
553 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
554 overridden with the primary group provided by the group upcall,
555 whereas the supplied fsgid can be trusted if it is in the list of
556 supplementary groups returned by the group upcall.
558 --------------------------------------------------------------------------------
560 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
562 * Support for kernels:
563 2.6.5-7.283 (SLES 9),
564 2.6.9-55.EL (RHEL 4),
565 2.6.16.46-0.14 (SLES 10),
566 2.6.18.8 vanilla (kernel.org)
567 * Client support for unpatched kernels:
568 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
569 2.6.16 - 2.6.22 vanilla (kernel.org)
570 * Due to recently discovered recovery problems, we do not recommend
571 using patchless RHEL 4 clients with this or any earlier release.
572 * Recommended e2fsprogs version: 1.39.cfs8
573 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
574 * Starting with this release, the ldiskfs backing filesystem required
575 by Lustre is now in its own package, lustre-ldiskfs. This package
576 should be installed. It is versioned separately from Lustre and
577 may be released separately in future.
581 Description: block reactivating mgc import until all deactivates complete
582 Details : Fix race when failing back MDT/MGS to itself (testing)
585 Frequency : at statup only
587 Description: mds_lov_synchronize race leads to various problems
588 Details : simultaneous MDT->OST connections at startup can cause the
589 sync to abort, leaving the OSC in a bad state.
591 Severity : enhancement
593 Description: add optional extra BUILD_VERSION info
594 Details : add a new environment variable (namely LUSTRE_VERS) which allows
595 to override the lustre version.
598 Frequency : 2.6.18 servers only
600 Description: ll_kern_mount() doesn't release the module reference
601 Details : The ldiskfs module reference count never drops down to 0
602 because ll_kern_mount() doesn't release the module reference.
607 Description: server LBUG when using old ost_num_threads parameter
608 Details : Accept the old ost_num_threads parameter but warn that it
609 is deprecated, and fix an off-by-one error that caused an LBUG.
614 Description: Transient SCSI error results in persistent IO issue
615 Details : iobuf->dr_error is not reinitialized to 0 between two
619 Frequency : sometimes when underlying device returns I/O errors
621 Description: OSTs not going read-only during write failures
622 Details : OSTs are not remounted read-only when the journal commit threads
623 get I/O errors because fsfilt_ext3 calls journal_start/stop()
624 instead of the ext3 wrappers.
628 Description: poor connect scaling with increasing client count
629 Details : Don't run filter_grant_sanity_check for more than 100 exports
630 to improve scaling for large numbers of clients.
633 Frequency : SLES10 only
635 Description: sanity-quota.sh quotacheck failed: rc = -22
636 Details : Quotas cannot be enabled on SLES10.
639 Frequency : liblustre clients only
641 Description: getdirentries does not give error when run on compute nodes
642 Details : getdirentries does not fail when the size specified as an argument
643 is too small to contain at least one entry
645 Severity : enhancement
647 Description: Add LNET router traceability for debug purposes
648 Details : If a checksum failure occurs with a router as part of the
649 IO path, the NID of the last router that forwarded the bulk data
650 is printed so it can be identified.
655 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
656 Details : Due to a race condition, liblustre clients were occasionally
659 Severity : enhancement
661 Description: lfs setstripe use optional parameters instead of postional
664 Severity : enhancement
666 Description: Nanosecond timestamp support for ldiskfs
667 Details : The on-disk ldiskfs filesystem has added support for nanosecond
668 resolution timestamps. There is not yet support for this at
669 the Lustre filesystem level.
672 Frequency : during server recovery
674 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
675 Details : request with rq_no_resend flag not awake l_wait_event if they get a
679 Frequency : nfs export on patchless client
681 Description: connectathon hang when test nfs export over patchless client
682 Details : Disconnected dentry cannot be found with lookup, so we do not need
683 to unhash it or make it invalid
686 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
687 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
688 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
693 Description: open req refcounting wrong on reconnect
694 Details : If reconnect happened between getting open reply from server and
695 call to mdc_set_replay_data in ll_file_open, we will schedule
696 replay for unreferenced request that we are about to free.
697 Subsequent close will crash in variety of ways.
698 Check that request is still eligible for replay in
699 mdc_set_replay_data().
704 Description: disable writes to filesystem when reading health_check file
705 Details : the default for reading the health_check proc file has changed
706 to NOT do a journal transaction and write to disk, because this
707 can cause reads of the /proc file to hang and block HA state
708 checking on a healthy but otherwise heavily loaded system. It
709 is possible to return to the previous behaviour during configure
710 with --enable-health-write.
712 Severity : enhancement
714 Description: 64-bit inode version
715 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
716 to the inode. This will be required for version-based recovery.
721 Description: MDS fails to start if a duplicate client export is detected
722 Details : in some rare cases it was possible for a client to connect to
723 an MDS multiple times. Upon recovery the MDS would detect this
724 and fail during startup. Handle this more gracefully.
726 Severity : enhancement
728 Description: Add -o localflock option to simulate old noflock
730 Details : This will achieve local-only flock/fcntl locks
736 Description: log_commit_thread vs filter_destroy race leads to crash
737 Details : Take import reference before releasing llog record semaphore
742 Description: Wrong request locking in request set processing
743 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
744 imp_delayed_list, in this place should be used imp_lock.
747 Frequency : when reconnection
749 Description: Grant Leak when osc reconnect to OST
750 Details : When osc reconnect ost, OST(filter) should check whether it
751 should grant more space to client by comparing fed_grant and
752 cl_avail_grant, and return the granted space to client instead
753 of "new granted" space, because client will call osc_init_grant
754 to update the client grant space info.
757 Frequency : when client reconnect to OST
759 Description: Grant Leak when osc do resend and replay bulk write
760 Details : When osc reconnect to OST, OST(filter)should clear grant info of
761 bulk write request, because the grant info while be sync between
762 OSC and OST when reconnect, and we should ignore the grant info
763 these of resend/replay write req.
768 Description: Grant space more than avaiable left space sometimes.
769 Details : When then OST is about to be full, if two bulk writing from
770 different clients came to OST. Accord the avaliable space of the
771 OST, the first req should be permitted, and the second one
772 should be denied by ENOSPC. But if the seconde arrived before
773 the first one is commited. The OST might wrongly permit second
774 writing, which will cause grant space > avaiable space.
777 Frequency : when client is evicted
779 Description: Grant might be wrongly erased when osc is evicted by OST
780 Details : when the import is evicted by server, it will fork another
781 thread ptlrpc_invalidate_import_thread to invalidate the
782 import, where the grant will be set to 0. While the original
783 thread will update the grant it got when connecting. So if
784 the former happened latter, the grant will be wrongly errased
785 because of this race.
790 Description: Checking Stale with correct fid
791 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
792 to check whether it is stale, because sometimes, we want the
793 enqueue happened anyway, and op_data.fid2 will not be initialized.
795 Severity : enhancement
797 Description: update patchless client
798 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
801 Frequency : only with 2.4 kernel
803 Description: random memory corruption
804 Details : size of struct ll_inode_info is to big for union inode.u and this
805 can be cause of random memory corruption.
810 Description: Memory leak in recovery
811 Details : Lov_mds_md was not free in an error handler in mds_create_object.
812 It should also check obd_fail before fsfilt_start, otherwise if
813 fsfilt_start return -EROFS,(failover mds during mds recovery).
814 then the req will return with repmsg->transno = 0 and rc = EROFS.
815 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
816 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
817 be freed no matter whether fsfilt_commit success or not.
820 Frequency : only with huge count clients
822 Description: Prevents from taking the superblock lock in llap_from_page for
824 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
825 for llap_from_page call in ll_removepage prevents from taking the
826 superblock lock for a soon died page.
831 Description: Not check open intent error before release open handle
832 Details : in some rare cases, the open intent error is not checked before
833 release open handle, which may cause
834 ASSERTION(open_req->rq_transno != 0), because it tries to release
835 the failed open handle.
840 Description: Set cat log bitmap only after create log success.
841 Details : in some rare cases, the cat log bitmap is set too early. and it
842 should be set only after create log success.
846 Description: Accessing a block bevice can re-enable I/O when Lustre is
847 tearing down a device.
848 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
853 Description: service threads may hog cpus when there are a lot of requests
854 Details : Insert cond_resched to give other threads a chance to use some CPU
859 Description: the cat log was not initialized in recovery
860 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
861 unlink log on mds will not be initialized until mds post
862 recovery. And also in mds post recovery, the unlink log will
863 initialization will be done asynchronausly, so there will be race
864 between add unlink log and unlink log initialization.
868 Description: brw_stats were being printed incorrectly
869 Details : brw_stats were being printed as log2 but all of them were not
870 recorded as log2. Also remove some code duplication arising from
871 filter_tally_{read,write}.
875 Frequency : rare, only in recovery.
876 Description: ASSERTION(req->rq_type != LI_POISON) failed
877 Details : imp_lock should be held while iterating over imp_sending_list for
878 prevent destroy request after get timeout in ptlrpc_queue_wait.
882 Description: replay-single.sh test 52 fails
883 Details : A lock's skiplist need to be cleanup when it being unlinked
884 from its resource list.
888 Description: Short directio read returns full requested size rather than
890 Details : Direct I/O operations should return actual amount of bytes
891 transferred rather than requested size.
893 Severity : enhancement
895 Description: metadata RPC reduction (e.g. for rm performance)
896 Details : decrease the amount of synchronous RPC between clients and servers
897 by canceling conflicing lock before the operation on the client
898 and packing thier handles into the main operation RPC to server.
900 Severity : enhancement
902 Description: Async OSC create to avoid the blocking unnecessarily.
903 Details : If a OST has no remain object, system will block on the creating
904 when need to create a new object on this OST. Now, ways use
905 pre-created objects when available, instead of blocking on an
906 empty osc while others are not empty. If we must block, we block
907 for the shortest possible period of time.
909 Severity : enhancement
911 Description: refine locking for avoid write wrong info into lov_objid file
912 Details : fix possible races with add new target and write/update data in
915 --------------------------------------------------------------------------------
917 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
922 Frequency : on some architectures
924 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
925 Details : Uninitialized flags sometimes cause configuration commands to
929 Frequency : patchless clients only
931 Description: missing __iget() symbol export
932 Details : The __iget() symbol export is missing. To avoid the need for
933 this on patchless clients the deathrow inode reaper is turned
934 off, and we depend on the VM to clean up old inodes. This
935 dependency was during via the fix for bug 12181.
937 --------------------------------------------------------------------------------
939 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
941 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
942 INTEROPERATE with older versions automatically. In many cases a
943 special upgrade step is needed. Please read the
944 user documentation before upgrading any part of a live system.
945 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
946 version will not interoperate with 1.6 betas before beta5 (1.5.95).
947 * WARNING: Lustre configuration and startup changes are required with
948 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
954 Severity : enhancement
956 Description: MountConf
957 Details : Lustre configuration is now managed via mkfs and mount
958 commands instead of lmc and lconf. New obd types (MGS, MGC)
959 are added for dynamic configuration management. See
960 https://mail.clusterfs.com/wikis/lustre/MountConf for
963 Severity : enhancement
965 Description: dynamic OST addition
966 Details : OSTs can now be added to a live filesystem
968 Severity : enhancement
970 Description: startup order invariance
971 Details : MDTs and OSTs can be started in any order. Clients only
972 require the MDT to complete startup.
974 Severity : enhancement
976 Description: parallel, asynchronous orphan cleanup
977 Details : orphan cleanup is now performed in separate threads for each
978 OST, allowing parallel non-blocking operation.
980 Severity : enhancement
982 Description: optimized stripe assignment
983 Details : stripe assignments are now made based on ost space available,
984 ost previous usage, and OSS previous usage, in order to try
985 to optimize storage space and networking resources.
987 Severity : enhancement
989 Description: Permanently set tunables
990 Details : All writable /proc/fs/lustre tunables can now be permanently
991 set on a per-server basis, at mkfs time or on a live system.
993 Severity : enhancement
995 Description: Lustre message v2
996 Details : Add lustre message format v2.
998 Severity : enhancement
1000 Description: client OST exclusion list
1001 Details : Clients can be started with a list of OSTs that should be
1002 declared "inactive" for known non-responsive OSTs.
1006 Description: SPEC SFS validation failure on NFS v2 over lustre.
1007 Details : Changes the blocksize for regular files to be 2x RPC size,
1008 and not depend on stripe size.
1010 Severity : enhancement
1012 Description: Multiple MD RPCs in flight.
1013 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1014 To avoid overly-overloading MDS, introduce a limit on number of
1015 MDS RPCs in flight for a single client and add /proc controls
1016 to adjust this limit.
1018 Severity : enhancement
1020 Description: client read/write statistics
1021 Details : Add client read/write call usage stats for performance
1022 analysis of user processes.
1023 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1024 file access. extents_stats shows chunk size distribution.
1025 extents_stats_per_process show chunk size distribution per
1028 Severity : enhancement
1030 Description: mds statistics
1031 Details : Add detailed mds operations statistics in
1032 /proc/fs/lustre/mds/*/stats.
1036 Description: Failure of copying files with lustre special EAs.
1037 Details : Client side always return success for setxattr call for lustre
1038 special xattr (currently only "trusted.lov").
1043 Description: Refcount LNET uuids
1044 Details : The global LNET uuid list grew linearly with every startup;
1045 refcount repeated list entries instead of always adding to
1048 Severity : enhancement
1050 Description: Dynamic service threads
1051 Details : Within a small range, start extra service threads
1052 automatically when the request queue builds up.
1055 Frequency : mixed-endian client/server environments
1057 Description: mixed-endian crashes
1058 Details : The new msg_v2 system had some failures in mixed-endian
1061 Severity : enhancement
1063 Description: Easy OST removal
1064 Details : OSTs can be permanently deactivated with e.g. 'lctl
1065 conf_param lustre-OST0001.osc.active=0'
1067 Severity : enhancement
1069 Description: MGS proc entries
1070 Details : Added basic proc entries for the MGS showing what filesystems
1073 Severity : enhancement
1075 Description: provide MGS failover
1076 Details : Added config lock reacquisition after MGS server failover.
1078 Severity : enhancement
1080 Description: add Linux 2.4 support
1081 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1085 Description: a large application tries to do I/O to the same resource and dies
1086 in the middle of it.
1087 Details : Check the req->rq_arrival time after the call to
1088 ost_brw_lock_get(), but before we do anything about
1089 processing it & sending the BULK transfer request. This
1090 should help move old stale pending locks off the queue as
1091 quickly as obd_timeout.
1094 Frequency : when an incorrect nid is specified during startup
1096 Description: ptlrpc connect to non-existant node causes kernel crash
1097 Details : LNET can't be re-entered from an event callback, which
1098 happened when we expire a message after the export has been
1099 cleaned up. Instead, hand the zombie cleanup off to another
1102 Severity : enhancement
1104 Description: plain/inodebits lock performance improvement
1105 Details : Grouping plain/inodebits in granted list by their request modes
1106 and bits policy, thus improving the performance of search through
1110 Frequency : only if OST filesystem is corrupted
1112 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1113 Details : for a short time RPCs with bulk IO are in the replay list,
1114 but replay of bulk IOs is unimplemented. If the OST filesystem
1115 is corrupted due to disk cache incoherency and then replay is
1116 started it is possible to trip an assertion. Avoid putting
1117 committed RPCs into the replay list at all to avoid this issue.
1120 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1122 Description: System hang on startup
1123 Details : This bug allowed the liblustre (e.g. catamount) client to
1124 return to the app before handling all startup RPCs. This
1125 could leave the node unresponsive to lustre network traffic
1126 and manifested as a server ptllnd timeout.
1128 Severity : enhancement
1130 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1131 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1132 boolean to control whether to print peer debug info when a
1133 client's RPC times out.
1136 Frequency : only for kernels with patches from Lustre below 1.4.3
1138 Description: Remove old rdonly API
1139 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1142 Frequency : only for devices with external journals
1144 Description: Set external device read-only also
1145 Details : During a commanded failover stop, we set the disk device
1146 read-only while the server shuts down. We now also set any
1147 external journal device read-only at the same time.
1150 Frequency : when upgrading from 1.4 while trying to change parameters
1152 Description: The wrong (new) MDC name was used when setting parameters for
1153 upgraded MDT's. Also allows changing of OSC (and MDC)
1154 parameters if --writeconf is specified at tunefs upgrade time.
1157 Frequency : when setting specific ost indicies
1159 Description: QOS code breaks on skipped indicies
1160 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1161 created with --index need not be sequential.
1163 Severity : enhancement
1165 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1166 Details : The uninit_groups feature works in conjunction with the kernel
1167 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1168 up the pass1 processing of e2fsck. This is a read-only feature
1169 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1170 mount filesystems that have had this feature enabled.
1172 Severity : enhancement
1174 Description: Improve multi-block allocation algorithm to avoid fragmentation
1175 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1176 allocation locality and avoid filesystem fragmentation.
1178 ------------------------------------------------------------------------------
1180 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1182 * Support for kernels:
1183 2.4.21-47.0.1.EL (RHEL 3)
1184 2.6.5-7.283 (SLES 9)
1185 2.6.9-42.0.10.EL (RHEL 4)
1186 2.6.12.6 vanilla (kernel.org)
1187 2.6.16.27-0.9 (SLES 10)
1188 * Recommended e2fsprogs version: 1.39.cfs5
1190 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1194 Frequency : occasional, depends on client load and configuration
1195 Bugzilla : 12181, 12203
1196 Description: data loss for recently-modified files
1198 Details : In some cases it is possible that recently written or created
1199 files may not be written to disk in a timely manner (this should
1200 normally be within 30s unless client IO load is very high).
1201 The problem appears as zero-length files or files that are a
1202 multiple of 1MB in size after a client crash or client eviction
1203 that are missing data at the end of the file.
1205 This problem is more likely to be hit on clients where files are
1206 repeatedly created and unlinked in the same directory, clients
1207 have a large amount of RAM, have many CPUs, the filesystem has
1208 many OSTs, the clients are rebooted frequently, and/or the files
1209 are not accessed by other nodes after being written.
1211 The presence of the problem can be detected by looking at
1212 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1213 smaller than the second (nr_unused) then dirty files will not
1214 be flushed automatically to disk. "sync; sleep 10" should be
1215 run several times on the node before unmounting it to update
1216 Lustre (this is also safe to run on nodes without this problem).
1218 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1219 that can cause this same problem, so customers using that kernel
1220 also need to update the kernel in addition to Lustre. In order
1221 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1223 It is normal that files written just before a client crash (less
1224 than 30s) may not yet have been flushed to disk, even for local
1228 Frequency : frequent on thin XT3 nodes
1230 Description: UUID collision on thin XT3 Linux nodes
1231 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1232 randomly, since we relied on an insufficiently-seeded PRNG.
1237 Description: OSS hangs after "All ost request buffers busy"
1238 Details : A deadlock between quota and journal operations caused OSS
1239 hangs after printing "All ost request buffers busy."
1242 Frequency : always on liblustre builds
1244 Description: Cleanup compiler warnings on liblustre
1247 Frequency : always on liblustre builds on XT3
1249 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1253 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1254 > 32000 subdirectories
1255 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1256 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1257 subdirectory count crosses 32000. This will aid e2fsck to
1258 correctly handle more than 32000 subdirectories.
1261 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1263 Description: System hang on startup
1264 Details : This bug allowed the liblustre (e.g. catamount) client to
1265 return to the app before handling all startup RPCs. This
1266 could leave the node unresponsive to lustre network traffic
1267 and manifested as a server ptllnd timeout.
1269 Severity : enhancement
1271 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1272 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1273 boolean to control whether to print peer debug info when a
1274 client's RPC times out.
1279 Description: make O_SYNC working on 2.6 kernels
1280 Details : 2.6 kernels use different method for mark pages for write,
1281 so need add a code to lustre for O_SYNC work.
1286 Description: Failure to close file and release space on NFS
1287 Details : Put inode details into lock acquired in ll_intent_file_open.
1288 Use mdc_intent_lock in ll_intent_open to properly
1289 detect all kind of errors unhandled by mdc_enqueue.
1294 Description: proc file read during shutdown sometimes raced obd removal,
1296 Details : Add lock to prevent obd access after proc file removal.
1299 Frequency : Only for files larger than 4GB on 32-bit clients.
1301 Description: improperly doing page alignment of locks
1302 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1303 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1308 Description: RPCs being resent when they shouldn't be
1309 Details : Some RPCs that should not be resent are being resent. This
1310 can cause inconsistencies in the RPC state machine. Do not
1311 resend such requests.
1314 Frequency : rare, only with NFS export
1316 Description: Crash on NFS re-export node
1317 Details : under very unusual load conditions an assertion is hit in
1318 ll_intent_file_open()
1321 Frequency : only if OST filesystem is corrupted
1323 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1324 Details : for a short time RPCs with bulk IO are in the replay list,
1325 but replay of bulk IOs is unimplemented. If the OST filesystem
1326 is corrupted due to disk cache incoherency and then replay is
1327 started it is possible to trip an assertion. Avoid putting
1328 committed RPCs into the replay list at all to avoid this issue.
1333 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1334 Details : Large single O_DIRECT read and write calls can fail to allocate
1335 a sufficiently large buffer to process the request. In case of
1336 allocation failure the allocation is retried with a smaller
1337 buffer and broken into smaller requests.
1339 Severity : enhancement
1341 Description: Add -o localflock option to simulate old noflock behaviour.
1342 Details : This will achieve local-only flock/fcntl locks coherentness.
1347 Description: versioning check is incomplete
1348 Details : Checking the version difference of client vs. server, report
1349 error if the gap is too big.
1354 Description: add support PG_writeback bit
1355 Details : add support for PG_writeback bit for Lustre, for more carefull
1356 work with page cache in 2.6 kernel. This also fix some deadlocks
1357 and remove hack for work O_SYNC with 2.6 kernel.
1359 Severity : enhancement
1361 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1362 Details : The uninit_groups feature works in conjunction with the kernel
1363 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1364 up the pass1 processing of e2fsck. This is a read-only feature
1365 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1366 mount filesystems that have had this feature enabled.
1368 Severity : enhancement
1370 Description: Improve multi-block allocation algorithm to avoid fragmentation
1371 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1372 allocation locality and avoid filesystem fragmentation.
1374 ------------------------------------------------------------------------------
1376 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1378 * Support for kernels:
1379 2.6.9-42.0.3.EL (RHEL 4)
1380 2.6.5-7.276 (SLES 9)
1381 2.4.21-47.0.1.EL (RHEL 3)
1382 2.6.12.6 vanilla (kernel.org)
1383 2.6.16.21-0.8 (SLES10)
1384 * Recommended e2fsprogs version: 1.39.cfs2-0
1386 * The backwards-compatible /proc/sys/portals symlink has been removed
1387 in this release. Before upgrading, please ensure that you change
1388 any configuration scripts or /etc/sysctl.conf files that access
1389 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1390 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1391 in advance of the upgrade on any system running Lustre 1.4.6 or
1392 newer, since /proc/sys/lnet was added in that version.
1393 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1397 Frequency : only when quota is used
1399 Description: avoid scanning export list for quota master
1400 Details : Change the algorithms to avoid scanning export list in order
1401 to improve the efficiency.
1404 Frequency : MDS failover only, very rarely
1406 Description: "went back in time" messages on mds failover
1407 Details : The greatest transno may be lost when the current operation
1408 finishes with an error (transno==0) and the client's last_rcvd
1409 record is over-written. Save the greatest transno in the
1410 mds_last_transno for this case.
1413 Frequency : always for specific kernels and striping counts
1415 Description: client may get "Matching packet too big" without ACL support
1416 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1417 when trying to access files in certain configurations. The
1418 clients should in fact be denied when mounting because they do
1419 not understand ACLs.
1422 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1424 Description: many clients connecting with IO in progress causes connect timeouts
1425 Details : Avoid synchronous journal commits to avoid delays caused by many
1426 clients connecting/disconnecting when bulk IO is in progress.
1427 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1428 OST_IO_PORTAL to avoid delays behind potentially many pending
1432 Frequency : occasionally with multiple writers to a single file
1434 Description: shared writes to file may result in wrong size reported by stat()
1435 Details : Allow growing of kms when extent lock is cancelled
1438 Frequency : always with random mmap IO to multi-striped file
1440 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1441 Details : Only if the hole is at the end of OST object so that kms is too
1442 small. Fix is to increase kms accordingly in ll_nopage.
1445 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1447 Description: writes to a missing object would leak memory on the OST
1448 Details : If there is an inconsistency between the MDS and OST filesystems,
1449 such that the MDS references an object that doesn't exist, writes
1450 to that object will leak memory due to incorrect cleanup in the
1451 error handling path, eventually running out of memory on the OST.
1456 Description: Creating too long symlink causes lustre errors
1457 Details : Check symlink and name lengths before sending requests to MDS.
1460 Frequency : only if flock is enabled (not on by default)
1462 Description: posix locks not released on fd closure on 2.6.9+
1463 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1464 kernels, this caused such locks not to be released on fd close and
1465 then assertions on fs unmount about still used locks.
1468 Frequency : MDS failover only, very rarely
1470 Description: clients may get ASSERTION(granted_lock != NULL)
1471 Details : When request was taking a long time, and a client was resending
1472 a getattr by name lock request. The were multiple lock requests
1473 with the same client lock handle and
1474 mds_getattr_name->fixup_handle_for_resent_request found one of the
1475 lock handles but later failed with ASSERTION(granted_lock != NULL).
1480 Description: handle->h_buffer_credits > 0, assertion failure
1481 Details : h_buffer_credits is zero after truncate, causing assertion
1482 failure. This patch extends the transaction or creates a new
1486 Frequency : NFS re-export or patchless client
1487 Bugzilla : 11179, 10796
1488 Description: Crash on NFS re-export node (__d_move)
1489 Details : We do not want to hash the dentry if we don't have a lock.
1490 But if this dentry is later used in d_move, we'd hit uninitialised
1491 list head d_hash, so we just do this to init d_hash field but
1492 leave dentry unhashed.
1495 Frequency : NFS re-export or patchless client
1497 Description: NFS exports has problem with symbolic link
1498 Details : lustre client didn't properly install dentry when re-exported
1499 to NFS or running patchless client.
1502 Frequency : NFS re-export or patchless client
1504 Description: Various nfs/patchless fixes.
1505 Details : fixes reuse disconected alias for lookup process - this fixes
1506 warning "find_exported_dentry: npd != pd",
1507 fix permission error with open files at nfs.
1508 fix apply umask when do revalidate.
1511 Frequency : occasional
1513 Description: Crash on NFS re-export node
1514 Details : calling clear_page() on the wrong pointer triggered oops in
1515 generic_mapping_read().
1518 Frequency : rarely, using O_DIRECT IO
1520 Description: unaligned directio crashes client with LASSERT
1521 Details : check for unaligned buffers before trying any requests.
1524 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1526 Description: stale data returned from RAID cache
1527 Details : If only a small amount of IO is done to the RAID device before
1528 reading it again it is possible to get stale data from the RAID
1529 cache instead of reading it from disk.
1532 Frequency : always for sles10 kernel
1534 Description: sles10 support
1535 Details : ll_follow_link: compile fixes and using of nd_set_link
1536 under newer kernels.
1539 Frequency : depends on arch, kernel and compiler version, always on sles10
1542 Description: recursive or deep enough symlinks cause stack overflow
1543 Details : getting rid of large stack-allocated variable in
1547 Frequency : depends on hardware
1549 Description: lustre write performance loss in the SLES10 kernel
1550 Details : the performance loss is caused by using of write barriers in the
1551 ext3 code. The SLES10 kernel turns barrier support on by
1552 default. The fix is to undo that change for ldiskfs.
1554 ------------------------------------------------------------------------------
1556 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1558 * Support for kernels:
1559 2.6.9-42.0.3EL (RHEL 4)
1560 2.6.5-7.276 (SLES 9)
1561 2.4.21-47.0.1.EL (RHEL 3)
1562 2.6.12.6 vanilla (kernel.org)
1566 Frequency : quota enabled and large files being deleted
1568 Description: releasing more than 4GB of quota at once hangs OST
1569 Details : If a user deletes more than 4GB of files on a single OST it
1570 will cause the OST to spin in an infinite loop. Release
1571 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1576 Description: statfs data retrieved from /proc may be stale or zero
1577 Details : When reading per-device statfs data from /proc, in the
1578 {kbytes,files}_{total,free,avail} files, it may appear
1579 as zero or be out of date.
1582 Frequency : systems with MD RAID1 external journal devices
1584 Description: lconf's call to blkid is confused by RAID1 journal devices
1585 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1586 of returning all block devices that match the journal UUID.
1589 Frequency : always, for aggregate stripe size over 4GB
1591 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1592 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1593 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1596 Frequency : always if "lfs find" used on a local file/directory
1598 Description: "lfs find" segfaults if used on a local file/directory
1599 Details : The case where a directory component was not specified wasn't
1600 handled correctly. Handle this properly.
1603 Frequency : always on ppc64
1605 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1606 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1607 perform endianness conversion.
1610 Frequency : rarely (truncate to non-zero file size after write under load)
1611 Bugzilla : 10730, 10687
1612 Description: Files padded with zeros to next 4K multiple
1613 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1614 it is possible that files that are truncated to a non-zero size
1615 immediately after being written are filled with zero bytes beyond
1616 the truncated size. No file data is lost.
1618 Severity : enhancement
1620 Description: Allow recovery/failover for liblustre clients.
1621 Details : liblustre clients were unaware of failover configurations until
1624 Severity : enhancement
1626 Description: user file locks should fail when not mounting with flock option
1627 Details : Set up an error-returning stub in ll_file_operations.lock field
1628 to prevent incorrect behaviour when client is mounted without
1629 flock option. Also, set up properly f_op->flock field for
1633 Frequency : always on ia64
1635 Description: "lfs df" loops on printing out MDS statfs information
1636 Details : The obd_ioctl_data was not initialized and in some systems
1637 this caused a failure during the ioctl that did not return
1638 an error. Initialize the struct and return an error on failure.
1641 Frequency : SLES 9 only
1643 Description: Error of copying files with lustre special EAs as root
1644 Details : Client side always return success for setxattr call for lustre
1645 special xattr (currently only "trusted.lov").
1648 Frequency : rarely on clusters with both ia64+i386 clients
1650 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1651 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1652 clients are doing concurrent writes to the same file it is
1653 possible that the ia64 clients may LASSERT because the OST
1654 extent locks are not PAGE_SIZE aligned. Ensure that grown
1655 locks are always aligned on the request boundary.
1658 Frequency : specific use, occasional
1660 Description: Overwriting in use executable truncates on-disk binary image
1661 Details : If one node attempts to overwrite an executable in use by
1662 another node, we now correctly return ETXTBSY instead of
1663 truncating the file.
1665 Severity : enhancement
1667 Description: Async OSC create to avoid the blocking unnecessarily.
1668 Details : If a OST has no remain object, system will block on the creating
1669 when need to create a new object on this OST. Now, ways use
1670 pre-created objects when available, instead of blocking on an
1671 empty osc while others are not empty. If we must block, we block
1672 for the shortest possible period of time.
1677 Description: chmod on Lustre root is propagated to other clients
1678 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1679 invalid by the follow_mount time.
1682 Frequency : liblustre clients only
1684 Description: Race in 'instant cancel' lock handling could lead to such locks
1685 never to be granted in case of SMP MDS
1686 Details : Do not destroy not yet granted but cbpending locks in
1690 Frequency : replay/resend of open
1692 Description: non null lock assetion failure in mds_intent_policy
1693 Details : Trying to replay/resend lockless open requests resulted in
1694 mds_open() returning 0 with no lock. Now it sets a flag if
1695 a lock is going to be returned.
1697 Severity : enhancement
1699 Description: Checksum enhancements
1700 Details : New checksum enhancements allow for resending RPCs that failed
1703 Severity : enhancement
1705 Description: Tunables on number of dirty pages in cacche
1706 Details : Allow to set limit on number of dirty pages cached.
1711 Description: client crash on unmount - lock still has references
1712 Details : In some error handling cases it was possible to leak a lock
1713 reference on a client while accessing a file. This was not
1714 harmful to the client during operation, but would cause the
1715 client to crash when the filesystem is unmounted.
1718 Frequency : specific case, rare
1720 Description: ETXTBSY on mds though file not in use by client
1721 Details : ETXTBSY is no longer incorrectly returned when attempting to
1722 chmod or chown a directory that the user previously tried to
1723 execute or a currently-executing binary.
1726 Frequency : extremely rare except on liblustre-based clients
1728 Description: Lustre space not freed when files are deleted
1729 Details : Clean up open-unlinked files after client eviction. Previously
1730 the unlink was skipped and the files remained as orphans.
1735 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1736 Details : In some cases it was possible to send a blocking callback to
1737 a client doing a glimpse, even though that client didn't get
1738 a lock granted. When the glimpse lock is cancelled on the OST
1739 the freed lock is left on the waiting list and corrupted the list.
1742 Frequency : all core dumps
1744 Description: Broke core dumps to lustre
1745 Details : Negative dentry may be unhashed if parent does not have UPDATE
1746 lock, but some callers, e.g. do_coredump, expect dentry to be
1747 hashed after successful create, hash it in ll_create_it.
1749 ------------------------------------------------------------------------------
1751 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1753 * Support for kernels:
1754 2.6.9-42.0.2.EL (RHEL 4)
1755 2.6.5-7.276 (SLES 9)
1756 2.4.21-40.EL (RHEL 3)
1757 2.6.12.6 vanilla (kernel.org)
1761 Frequency : always on RHEL 3
1763 Description: Number of open files grows over time
1764 Details : The number of open files grows over time, whether or not
1765 Lustre is started. This was due to a filp leak introduced
1766 by one of our kernel patches.
1768 ------------------------------------------------------------------------------
1770 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1772 * Support for kernels:
1773 2.6.9-42.EL (RHEL 4)
1774 2.6.5-7.276 (SLES 9)
1775 2.4.21-40.EL (RHEL 3)
1776 2.6.12.6 vanilla (kernel.org)
1781 Bugzilla : 5719, 9635, 9792, 9684
1782 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1783 Details : If a server is under heavy load and cannot reply to new
1784 connection requests before the client resends the (re)connect,
1785 the connection handling code can behave badly if two service
1786 threads are concurrently handing separate (re)connections from
1787 the same client. Add better locking to the connection handling
1788 code, and ensure that only a single connection will be processed
1789 for a given client UUID, even if the lock is dropped.
1791 Severity : enhancement
1793 Description: add TCP zero-copy support to kernel
1794 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1795 sends if the hardware supports scatter-gather and checksumming.
1796 This allows socklnd to do client-write and server-read more
1797 efficiently and reduce CPU utilization from skbuf copying.
1800 Frequency : only if NFS exporting from client
1802 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1803 Details : If setting attributes on a file created under NFS that had
1804 never been opened it would be possible to oops the client
1805 if the file had no objects.
1808 Frequency : always for liblustre
1810 Description: liblustre client does MDS+OSTs setattr RPC for each write
1811 Details : When doing a write from a liblustre client, the client
1812 incorrectly issued an RPC to the MDS and each OST the file was
1813 striped over in order to update the timestamps. When writing
1814 with small chunks and many clients this could overwhelm the MDS
1815 with RPCs. In all cases it would slow down the write because
1816 these RPCs are unnecessary.
1818 Severity : enhancement
1820 Description: allow number of MDS service threads to be changed at module load
1821 Details : It is now possible to change the number of MDS service threads
1822 running. Adding "options mds mds_num_threads={N}" to the MDS's
1823 /etc/modprobe.conf will set the number of threads for the next
1824 time Lustre is restarted (assuming the "mds" module is also
1825 reloaded at that time). The default number of threads will
1826 stay the same, 32 for most systems.
1831 Description: OST crash if filesystem is unformatted or corrupt
1832 Details : If an OST is started on a device that has never been formatted
1833 or if the filesystem is corrupt and cannot even mount then the
1834 error handling cleanup routines would dereference a NULL pointer.
1839 Description: NULL pointer deref in llap_from_page.
1840 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1841 incorrect) mapping if the page was truncated/reclaimed while it was
1842 searched for. Check for this condition and skip such pages when
1843 doing readahead. Introduce extra check to llap_from_page() to
1844 verify page->mapping->host is non-NULL (so page is not anonymous).
1847 Frequency : Sometimes when using sys_sendfile
1849 Description: "page not covered by a lock" warnings from ll_readpage
1850 Details : sendfile called ll_readpage without right page locks present.
1851 Now we introduced ll_file_sendfile that does necessary locking
1852 around call to generic_file_sendfile() much like we do in
1856 Frequency : with certain MDS communication failures at client mount time
1858 Description: NULL pointer deref after failed client mount
1859 Details : a client connection request may delayed by the network layer
1860 and not be sent until after the PTLRPC layer has timed out the
1861 request. If the client fails the mount immediately it will try
1862 to clean up before the network times out the request. Add a
1863 reference from the request import to the obd device and delay
1864 the cleanup until the network drops the request.
1867 Frequency : occasionally during client (re)connect
1869 Description: assertion failure during client (re)connect
1870 Details : processing a client connection request may be delayed by the
1871 client or server longer than the client connect timeout. This
1872 causes the client to resend the connection request. If the
1873 original connection request is replied in this interval, the
1874 client may trip an assertion failure in ptlrpc_connect_interpret()
1875 which thought it would be the only running connect process.
1878 Frequency : only with obd_echo servers and clients that are rebooted
1880 Description: kernel BUG accessing uninitialized data structure
1881 Details : When running an obd_echo server it did not start the ping_evictor
1882 thread, and when a client was evicted an uninitialized data
1883 structure was accessed. Start the ping_evictor in the RPC
1884 service startup instead of the OBD startup.
1886 Severity : enhancement
1887 Bugzilla : 10193 (patchless)
1888 Description: Remove dependency on various unexported kernel interfaces.
1889 Details : No longer need reparent_to_init, exit_mm, exit_files,
1890 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1893 Frequency : rare (only users of deprecated and unsupported LDAP config)
1895 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1896 Details : LDAP apparently contains 'lustreName' attributes instead of
1897 'name'. A simple remapping of the name is sufficient.
1900 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1902 Description: waiting_locks_callback trips kernel BUG if client is evicted
1903 Details : Running with the dump_on_timeout debug flag turned on makes
1904 it possible that the waiting_locks_callback() can try to dump
1905 the Lustre kernel debug logs from an interrupt handler. Defer
1906 this log dumping to the expired_lock_main() thread.
1908 Severity : enhancement
1910 Description: Support NFS exporting on 2.6 kernels.
1911 Details : Implement non-rawops metadata methods for NFS server to use without
1912 changing NFS server code.
1915 Frequency : very rare (synthetic metadata workload only)
1917 Description: two racing renames might cause an MDS thread to deadlock
1918 Details : Running the "racer" program may cause one MDS thread to rename
1919 a file from being the source of a rename to being the target of
1920 a rename at exactly the same time that another thread is doing
1921 so, and the second thread has already enqueued these locks after
1922 doing a lookup of the target and is trying to relock them in
1923 order. Ensure that we don't try to re-lock the same resource.
1926 Frequency : only very large systems with liblustre clients
1928 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1929 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1930 all OSTs in parallel. This allows the eviction of stale liblustre
1931 clients to proceed much faster than if they were done in series,
1932 and also offers similar improvements for other set_info RPCs.
1937 Description: excessive CPU usage during initial read phase on client
1938 Details : During the initial read phase on a client, it would agressively
1939 retry readahead on the file, consuming too much CPU and impacting
1940 performance (since 1.4.5.8). Improve the readahead algorithm
1941 to avoid this, and also improve some other common cases (read
1942 of small files in particular, where "small" is files smaller than
1943 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1948 Description: MDS crash when receiving packet with unknown intent.
1949 Details : Do not LBUG in unknown intent case, just return -EFAULT
1951 Severity : enhancement
1952 Bugzilla : 9293, 9385
1953 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1954 Details : Do not serialize getattr (non-intent version) and statfs.
1957 Frequency : occasional, when OST network is overloaded/intermittent
1959 Description: client evicted by OST after bulk IO timeout
1960 Details : If a client sends a bulk IO request (read or write) the OST
1961 may evict the client if it is unresposive to its data GET/PUT
1962 request. This is incorrect if the network is overloaded (takes
1963 too long to transfer the RPC data) or dropped the OST GET/PUT
1964 request. There is no need to evict the client at all, since
1965 the pinger and/or lock callbacks will handle this, and the
1966 client can restart the bulk request.
1969 Frequency : Always when mmapping file with no objects
1971 Description: client crashes when mmapping file with no objects
1972 Details : Check that we actually have objects in a file before doing any
1973 operations on objects in ll_vm_open, ll_vm_close and
1979 Description: Request leak when working with deleted CWD
1980 Details : Introduce advanced request refcount tracking for requests
1981 referenced from lustre intent.
1983 Severity : Enhancement
1985 Description: Cache open file handles on client.
1986 Details : MDS now will return special lock along with openhandle, if
1987 requested and client is allowed to hold openhandle, even if unused,
1988 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1989 closing files for every read/write openration.
1991 Severity : Enhancement
1993 Description: Cache open negative dentries on client when possible.
1994 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1995 negative dentries on lock revocation.
2000 Description: Remounting a client read-only wasn't possible with a zconf mount
2001 Details : It wasn't possible to remount a client read-only with llmount.
2003 Severity : enhancement
2004 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2005 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2006 MPICH is included to add Lustre-specific ADIO interfaces.
2007 This is based closely on the UFS ADIO layer and only differs
2008 in file creation, in order to allow the OST striping to be set.
2009 This is user-contributed code and not supported by CFS.
2014 Description: extended inode attributes (immutable, append-only) work improperly
2015 when 2.4 and 2.6 kernels are used on client/server or vice versa
2016 Details : Introduce kernel-independent values for these flags.
2018 Severity : enhancement
2021 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2022 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2023 granularity. Now, Fractional MB granularity is be supported,
2024 this is very useful for low memory system.
2026 Severity : enhancement
2028 Description: Getattr by fid
2029 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2030 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2031 have locks and avoids getattr rpc requests for it on every stat.
2034 Frequency : Always, for filesystems larger than 2TB
2036 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2037 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2038 filesystems larger than 2TB to be mounted with mballoc enabled.
2041 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2043 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2044 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2045 filesystems larger than 2TB is mounted then the kernel will
2046 silently corrupt the start of the filesystem. CONFIG_LBD is
2047 enabled for all CFS-supported kernels, but the possibility of
2048 this happening with a modified kernel config exists.
2050 Severity : enhancement
2052 Description: add client O_DIRECT support for 2.6 kernels
2053 Details : It is now possible to do O_DIRECT reads and writes to files
2054 in the Lustre client mountpoint on 2.6 kernel clients.
2056 Severity : enhancement
2058 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2059 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2060 parallel, not waiting for response from every OST before sending
2061 a rpc to the next OST.
2066 Description: setattr vs write race when updating file timestamps
2067 Details : Client processes that update a file timestamp into the past
2068 right after writing to the file (e.g. tar) it is possible that
2069 the updated file modification time can be reset to the current
2070 time due to a race between processing the setattr and write RPC.
2072 Severity : enhancement
2074 Description: Bring 'lfs find' closer in line with regular Linux find.
2075 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2076 -print0 options and obtains all the needed info through the lustre
2079 Severity : enhancement
2081 Description: support up to 1024 configured devices on one node
2082 Details : change obd_dev array from statically allocated to dynamically
2083 allocated structs as they are first used to reduce memory usage
2088 Description: Flush dirty partially truncated pages during truncate
2089 Details : Immediatelly flush partially truncated pages in filter_setattr,
2090 this way we completely avoid having any pages in page cache on OST
2091 and can retire ugly workarounds during writes to flush such pages.
2096 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2097 Details : protect inode from truncation within vfs_unlink() context
2098 just take a reference before calling vfs_unlink() and release it
2099 when parent's i_sem is free.
2104 Description: last_id value checked outside lock on OST caused LASSERT failure
2105 Details : If there were multiple MDS->OST object precreate requests in
2106 flight, it was possible that the OST's last object id was checked
2107 outside a lock and incorrectly tripped an assertion. Move checks
2108 inside locks, and discard old precreate requests.
2111 Frequency : always, if extents are used on OSTs
2113 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2114 Details : OSTs using the extents format would not zero the high 16 bits of
2115 the index physical block number. This is not a problem for any
2116 OST filesystems smaller than 16TB, and no kernels support ext3
2117 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2118 new/modified files) and can be fixed for existing filesystems
2119 with e2fsprogs-1.39-cfs1.
2124 Description: import connection selection may be incorrect if timer wraps
2125 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2126 import connections to be ignored if the 32-bit jiffies counter
2127 wraps. Use a 64-bit jiffies counter.
2130 Frequency : very large clusters immediately after boot
2132 Description: LNET request buffers exhausted under heavy short-term load
2133 Details : If a large number of client requests are generated on a service
2134 that has previously never seen so many requests it is possible
2135 that the request buffer growth cannot keep up with the spike in
2136 demand. Instead of dropping incoming requests, they are held in
2137 the LND until the RPC service can accept more requests.
2140 Frequency : Sometimes during replay
2142 Description: Assertion failure in ll_local_open after replay.
2143 Details : If replay happened on an open request reply before we were able
2144 to set replay handler, reply will become not swabbed tripping the
2145 assertion in ll_local_open. Now we set the handler right after
2146 recognising of open request
2149 Frequency : very rare
2151 Description: kernel reports "badness in vsnprintf"
2152 Details : Reading from the "recovery_status" /proc file in small chunks
2153 may cause a negative length in lprocfs_obd_rd_recovery_status()
2154 call to vsnprintf() (which is otherwise harmless). Exit early
2155 if there is no more space in the output buffer.
2157 Severity : enhancement
2159 Description: clear OBD RPC statistics by writing to them
2160 Details : It is now possible to clear the OBD RPC statistics by writing
2161 to the "stats" file.
2166 Description: Client mtime is not the same on different clients after utimes
2167 Details : In some cases, the client was using the utimes() syscall on
2168 a file cached on another node. The clients now validate the
2169 ctime from the MDS + OSTs to determine which one is right.
2174 Description: Inability to activate failout mode
2175 Details : lconf script incorrectly assumed that in pythong string's numeric
2176 value is used in comparisons.
2179 Frequency : always with multiple stripes per file
2181 Description: Inefficient object allocation for mutli-stripe files
2182 Details : When selecting which OSTs to stripe files over, for files with
2183 a stripe count that divides evenly into the number of OSTs,
2184 the MDS is always picking the same starting OST for each file.
2185 Return the OST selection heuristic to the original design.
2190 Description: mount failures may take full timeout to return an error
2191 Details : Under some heavy load conditions it is possible that a
2192 failed mount can wait for the full obd_timeout interval,
2193 possibly several minutes, before reporting an error.
2194 Instead return an error as soon as the status is known.
2196 Frequency : quota enabled and large files being deleted
2198 Description: releasing more than 4GB of quota at once hangs OST
2199 Details : If a user deletes more than 4GB of files on a single OST it
2200 will cause the OST to spin in an infinite loop. Release
2201 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2206 Description: statfs data retrieved from /proc may be stale or zero
2207 Details : When reading per-device statfs data from /proc, in the
2208 {kbytes,files}_{total,free,avail} files, it may appear
2209 as zero or be out of date.
2212 Frequency : systems with MD RAID1 external journal devices
2214 Description: lconf's call to blkid is confused by RAID1 journal devices
2215 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2216 of returning all block devices that match the journal UUID.
2219 Frequency : always, for aggregate stripe size over 4GB
2221 Description: assertion fails when trying to use 4GB stripe size
2222 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2223 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2226 Frequency : always on ppc64
2228 Description: the first write on an ext3 filesystem with mballoc got stuck
2229 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2230 perform endianness conversion.
2232 ------------------------------------------------------------------------------
2234 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2236 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2237 INTEROPERATE with older versions automatically. Please read the
2238 user documentation before upgrading any part of a live system.
2239 * WARNING: Lustre networking configuration changes are required with
2240 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2243 * Support for kernels:
2244 2.6.9-22.0.2.EL (RHEL 4)
2245 2.6.5-7.244 (SLES 9)
2246 2.6.12.6 vanilla (kernel.org)
2249 Severity : enhancement
2250 Bugzilla : 7981/8208
2251 Description: Introduced Lustre Networking (LNET)
2252 Details : LNET is new networking infrastructure for Lustre, it includes
2253 a reorganized network configuration mode (see the user
2254 documentation for full details) as well as support for routing
2255 between different network fabrics. Lustre Networking Devices
2256 (LNDs) for the supported network fabrics have also been
2257 created for this new infrastructure.
2259 Severity : enhancement
2260 Description: Introduced Access control lists
2261 Details : clients can set ACLs on files and directories in order to have
2262 more fine-grained permissions than the standard Unix UGO+RWX.
2263 The MDS must be started with the "-o acl" mount option.
2265 Severity : enhancement
2266 Description: Introduced filesystem quotas
2267 Details : Administrators may now establish per-user quotas on the
2270 Severity : enhancement
2272 Description: Configuration change for the XT3
2273 The PTLLND is now used to run Lustre over Portals on the XT3
2274 The configure option(s) --with-cray-portals are no longer used.
2275 Rather --with-portals=<path-to-portals-includes> is used to
2276 enable building on the XT3. In addition to enable XT3 specific
2277 features the option --enable-cray-xt3 must be used.
2282 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2283 Details : A race between allocating a new llcd and re-getting the llcd_lock
2284 allowed another thread to grab newly-allocated llcd.
2286 Severity : enhancement
2288 Description: 2.6 OST async journal commit and locking fix to improve performance
2289 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2290 2.6 kernels are now async as they already were in 2.4 kernels so
2291 that they can commit concurrently with the network bulk transfer.
2292 For block-allocated files the filter allocation semaphore is held
2293 to avoid filesystem fragmentation during allocation. BKL lock
2294 removed for 2.6 xattr operations where it is no longer needed.
2299 Description: lconf incorrectly determined whether two IP networks could talk
2300 Details : In some more complicated routing and multiple-network
2301 configurations, lconf will avoid trying to make a network
2302 connection to a disjoint part of the IP space. It was doing the
2303 math incorrectly for one set of cases.
2308 Description: Fix for potential infinite loop processing records in an llog.
2309 Details : If an llog record is corrupted/zeroed, it is possible to loop
2310 forever in llog_process(). Validate the llog record length
2311 and skip the remainder of the block on error.
2314 Frequency : occasional (liblustre only)
2316 Description: liblustre could not open files whose last component is a symlink
2317 Details : sysio_path_walk() would incorrectly pass the open intent to
2318 intermediate path components.
2321 Frequency : rare (liblustre only with non-standard tuning)
2322 Bugzilla : 7201 (7350)
2323 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2324 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2325 after it has been released, only do this for referenced locks
2327 Severity : enhancement
2329 Description: specifying an (invalid) directory default stripe_size of -1
2330 would reset the directory default striping
2331 Details : stripe_size -1 was used internally to signal directory stripe
2332 removal, now use "all default" to signal dir stripe removal
2333 as a directory striping of "all default" is not useful
2336 Frequency : common for large clusters running liblustre clients
2338 Description: doing an ls when liblustre clients are running is slow
2339 Details : sending a glimpse AST to a liblustre client waits for every AST
2340 to time out, as liblustre clients will not respond. Since they
2341 cannot cache data we refresh the OST lock LVB from disk instead.
2343 Severity : enhancement
2345 Description: doing an ls at the same time as file IO can be slow
2346 Details : enqueue and other "small" requests can be blocked behind many
2347 large IO requests. Create a new OST IO portal for non-IO
2348 requests so they can be processed faster.
2351 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2353 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2354 created file with mode 000
2355 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2356 allow an ftruncate() as a non-root user to a file with mode 000.
2357 Lustre now respects this flag to disable mode checks when
2358 truncating a file owned by the user
2361 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2364 Description: Revoking locks from clients that went dead or catatonic might take
2366 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2367 cancellation of such locks instant on servers without waiting for
2368 any reply from clients. Clients drops these locks when cancel
2369 notification from server is received without replying.
2372 Frequency : liblustre-only, when liblustre client dies or becomes busy
2374 Description: Doing ls on Linux clients can take a long time with active
2376 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2377 granting such locks to it in the first place if possible. Locks
2378 are taken by proxy on the OST during the read or write and
2379 dropped immediately afterward. Add connect flags handling, do
2380 not grant locks to liblustre clients for glimpse ASTs.
2382 Severity : enhancement
2384 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2385 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2386 sequential reads and behaves badly with applications doing random
2387 reads. Improve it by reading ahead at least read region, and
2388 avoiding excessive large RPC for small reads.
2390 Severity : enhancement
2392 Description: Creating more than 1000 files for a single job may cause a load
2393 imbalance on the OSTs if there are also a large number of OSTs.
2394 Details : qos_prep_create() uses an OST index reseed value that is an
2395 even multiple of the number of available OSTs so that if the
2396 reseed happens in the middle of the object allocation it will
2397 still utilize the OSTs as uniformly as possible.
2402 Description: OST or MDS may oops in ping_evictor_main()
2403 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2404 but doesn't restart at beginning of obd_exports_timed list
2407 Severity : enhancement
2409 Description: improve by-nid export eviction on the MDS and OST
2410 Details : allow multiple exports with the same NID to be evicted at one
2411 time without re-searching the exports list.
2414 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2416 Description: MDS may oops in groups_free()
2417 Details : in rare race conditions a newly allocated group_info struct is
2418 freed again, and this can be NULL. The 2.4 compatibility code
2419 for groups_free() checked for a NULL pointer, but 2.6 did not.
2422 Frequency : common for liblustre clients doing little filesystem IO
2423 Bugzilla : 9352, 7313
2424 Description: server may evict liblustre clients accessing contended locks
2425 Details : if a client is granted a lock or receives a completion AST
2426 with a blocking AST already set it would not reply to the AST
2427 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2430 Frequency : lfs setstripe, only systems with more than 160 OSTs
2432 Description: unable to set striping with a starting offset beyond OST 160
2433 Details : llapi_create_file() incorrectly limited the starting stripe
2434 index to the maximum single-file stripe count.
2437 Frequency : LDAP users only
2439 Description: lconf did not handle in-kernel recovery with LDAP properly
2440 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2442 Severity : enhancement
2444 Description: bind OST threads to NUMA nodes to improve performance
2445 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2446 node and do their allocations there to localize memory access
2448 Severity : enhancement
2450 Description: llmount can determine client NID directly from Myrinet (GM)
2451 Details : the client NID code from gmnalnid was moved directly into
2452 llmount, removing the need to use this or specifying the
2453 client NID explicitly when mounting GM clients with zeroconf
2456 Frequency : if client is started with down MDS
2458 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2459 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2460 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2461 ptlrpc_import_delayed_req() because we didn't check if the
2462 request was interrupted, and we also didn't break out of the
2463 event loop if there was no timeout
2468 Description: data loss during non-page-aligned writes to a single file from
2469 both multiple nodes and multiple threads on one node at same time
2470 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2471 inconsistency led to false short-reads, that were cached and later
2472 used by ->prepare_write() to fill in partially written page,
2473 leading to data loss.
2476 Frequency : always, if lconf --abort_recovery used
2478 Description: lconf --abort_recovery fails with 'Operation not supported'
2479 Details : lconf was attempting to abort recovery on the MDT device and not
2482 Severity : enhancement
2484 Description: remove cleanup logs
2485 Details : replace lconf-generated cleanup logs with lustre internal
2486 cleanup routines. Eliminates the need for client-cleanup and
2489 Severity : enhancement
2491 Description: add support for EAs (user and system) on lustre filesystems
2492 Details : it is now possible to store extended attributes in the Lustre
2493 client filesystem, and with the user_xattr mount option it
2494 is possible to allow users to store EAs on their files also
2496 Severity : enhancement
2498 Description: Add possibility (config option) to show minimal available OST free
2500 Details : When compiled with --enable-mindf configure option, statfs(2)
2501 (and so, df) will return least minimal free space available from
2502 all OSTs as amount of free space on FS, instead of summary of
2503 free spaces of all OSTs.
2505 Severity : enhancement
2507 Description: do not expand extent locks acquired on OST-side
2508 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2509 by server: they are not cached anyway.
2512 Frequency : when mmap is used/binaries executed from Lustre
2514 Description: Unmmap pages before throwing them away from read cache.
2515 Details : llap_shrink cache now attempts to unmap pages before discarding
2516 them (if unmapping failed - do not discard). SLES9 kernel has
2517 extra checks that trigger if this unmapping is not done first.
2522 Description: lconf didn't resolve symlinks before checking to see whether a
2523 given mountpoint was already in use
2526 Frequency : when migrating failover services
2527 Bugzilla : 6395, 9514
2528 Description: When migrating a subset of services from a node (e.g. failback
2529 from a failover service node) the remaining services would
2530 time out and evict clients.
2531 Details : lconf --force (implied by --failover) sets the global obd_timeout
2532 to 5 seconds in order to quickly disconnect, but this caused
2533 other RPCs to time out too quickly. Do not change the global
2534 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2536 Severity : enhancement
2537 Frequency : if MDS is started with down OST
2538 Bugzilla : 9439,5706
2539 Description: Allow startup/shutdown of an MDS without depending on the
2540 availability of the OSTs.
2541 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2542 Add appropriate locking and lov-osc refcounts for safe
2543 cleaning. Add osc abort_inflight calls in case the
2544 synchronize never started.
2547 Frequency : occasional (Cray XT3 only)
2549 Description: root not authorized to access files in CRAY_PORTALS environment
2550 Details : The client process capabilities were not honoured on the MDS in
2551 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2552 been accessed by an authorized user then root was able to access
2553 the file on the local client also. The root user capabilities
2554 are now allowed on the MDS, as this environment has secure UID.
2557 Frequency : occasional
2559 Description: ldiskfs "too long searching" message happens too often
2560 Details : A debugging message (otherwise harmless) prints too often on
2561 the OST console. This has been reduced to only happen when
2562 there are fragmentation problems on the filesystem.
2567 Description: Division by zero in statfs when all OSCs are inactive
2568 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2569 lov_statfs divides by value returned by lov_get_stripecnt().
2573 Bugzilla : 9489, 3273
2574 Description: First write from each client to each OST was only 4kB in size,
2575 to initialize client writeback cache, which caused sub-optimal
2576 RPCs and poor layout on disk for the first writen file.
2577 Details : Clients now request an initial cache grant at (re)connect time
2578 and so that they can start streaming writes to the cache right
2579 away and always do full-sized RPCs if there is enough data.
2580 If the OST is rebooted the client also re-establishes its grant
2581 so that client cached writes will be honoured under the grant.
2586 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2587 Details : Now I/O RPCs go to different portal number and (presumably) fast
2588 lock requests (and glimses) and other RPCs get their own service
2589 threads pool that should be able to service those RPCs
2592 Severity : enhancement
2594 Description: Ability to exchange lustre version between client and servers and
2595 issue warnings at client side if client is too old. Also for
2596 liblustre clients there is ability to refuse connection of too old
2598 Details : New 'version' field is added to connect data structure that is
2599 filled with version info. That info is later checked by server and
2603 Frequency : rare, liblustre only.
2604 Bugzilla : 9296, 9581
2605 Description: Two simultaneous writes from liblustre at offset within same page
2606 might proceed at the same time overwriting eachother with stale
2608 Details : I/O lock withing llu_file_prwv was released too early, before data
2609 actually was hitting the wire. Extended lock-holding time until
2610 server acknowledges receiving data.
2613 Frequency : extremely rare. Never observed in practice.
2615 Description: avoid generating lustre_handle cookie of 0.
2616 Details : class_handle_hash() generates handle cookies by incrementing
2617 global counter, and can hit 0 occasionaly (this is unlikely, but
2618 not impossible, because initial value of cookie counter is
2619 selected randonly). Value of 0 is used as a sentinel meaning
2620 "unassigned handle" --- avoid it. Also coalesce two critical
2621 sections in this function into one.
2623 Severity : enhancement
2625 Description: allow liblustre clients to delegate truncate locking to OST
2626 Details : To avoid overhead of locking, liblustre client instructs OST to
2627 take extent lock in ost_punch() on client's behalf. New connection
2628 flag is added to handle backward compatibility.
2630 Severity : enhancement
2631 Bugzilla : 4928, 7341, 9758
2632 Description: allow number of OST service threads to be specified
2633 Details : a module parameter allows the number of OST service threads
2634 to be specified via "options ost ost_num_threads={N}" in the
2635 OSS's /etc/modules.conf or /etc/modprobe.conf.
2639 Bugzilla : 6146, 9635, 9895
2640 Description: servers crash with bad pointer in target_handle_connect()
2641 Details : In rare cases when a client is reconnecting it was possible that
2642 the connection request was the last reference for that export.
2643 We would temporarily drop the export reference and get a new
2644 one, but this may have been the last reference and the export
2645 was just destroyed. Get new reference before dropping old one.
2647 Severity : enhancement
2648 Frequency : if client is started with failover MDS
2650 Description: Allow multiple MDS hostnames in the mount command
2651 Details : Try to read the configuration from all specified MDS
2652 hostnames during a client mount in case the "primary"
2655 Severity : enhancement
2657 Description: Stop sending data to evicted clients as soon as possible.
2658 Details : Check if the client we are about to send or are sending data to
2659 was evicted already. (Check is done every second of waiting,
2660 for which l_wait_event interface was extended to allow checking
2661 of exit condition at specified intervals).
2664 Frequency : rare, normally only when NFS exporting is done from client
2666 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2668 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2671 Frequency : always, if srand() is called before liblustre initialization
2673 Description: Liblustre uses system PRNG disturbing its usage by user application
2674 Details : Introduce internal to lustre fast and high-quality PRNG for
2675 lustre usage and make liblustre and some other places in generic
2676 lustre code to use it.
2678 Severity : enhancement
2679 Bugzilla : 9477, 9557, 9870
2680 Description: Verify that the MDS configuration logs are updated when xml is
2681 Details : Check if the .xml configuration logs are newer than the config
2682 logs stored on the MDS and report an error if this is the case.
2683 Request --write-conf, or allow starting with --old_conf.
2685 Severity : enhancement
2687 Description: Handle symlinks in the path when checking if Lustre is mounted.
2688 Details : Resolve intermediate symlinks when checking if a client has
2689 mounted a filesystem to avoid duplicate client mounts.
2694 Description: lconf can hit an error exception but still return success.
2695 Details : The lconf command catches the Command error exception at the top
2696 level script context and will exit with the associated exit
2697 status, but doesn't ensure that this exit status is non-zero.
2702 Description: failure of ptlrpc thread startup can cause oops
2703 Details : Starting a ptlrpc service thread can fail if there are a large
2704 number of threads or the server memory is very fragmented.
2705 Handle this without oopsing.
2708 Frequency : always, only if liblustre and non-default acceptor port was used
2710 Description: liblustre cannot connect to servers with non-default acceptor port
2711 Details : tcpnal_set_default_params() was not called and was therefore
2712 ignoring the environment varaible TCPNAL_PORT, as well as other
2713 TCPNAL_ environment variables
2718 Description: two objects could be created on the same OST for a single file
2719 Details : If an OST is down, in some cases it was possible to create two
2720 objects on a single OST for a single file. No problems other
2721 than potential performance impact and spurious error messages.
2725 Bugzilla : 5681, 9562
2726 Description: Client may oops in ll_unhash_aliases
2727 Details : Client dcache may become inconsistent in race condition.
2728 In some cases "getcwd" can fail if the current directory is
2734 Description: Inode refcounting problems in NFS export code
2735 Details : link_raw functions used to call d_instantiate without obtaining
2736 extra inode reference first.
2740 Bugzilla : 9942, 9903
2741 Description: Referencing freed requests leading to crash, memleaks with NFS.
2742 Details : We used to require that call to ll_revalidate_it was always
2743 followed by ll_lookup_it. Also with revalidate_special() it is
2744 possible to call ll_revalidate_it() twice for the same dentry
2745 even if first occurence returned success. This fix changes semantic
2746 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2747 reference on a request referred from the intent.
2748 ll_intent_release() then releases such a request.
2751 Frequency : rare, normally benchmark loads only
2753 Description: unlinked inodes were kept in memory on the client
2754 Details : If a client is repeatedly creating and unlinking files it
2755 can accumulate a lot of stale inodes in the inode slab cache.
2756 If there is no other client load running this can cause the
2757 client node to run out of memory. Instead flush old inodes
2758 from client cache that have the same inode number as a new inode.
2761 Frequency : SLES9 2.6.5 kernel and long filenames only
2762 Bugzilla : 9969, 10379
2763 Description: utime reports stale NFS file handle
2764 Details : SLES9 uses out-of-dentry names in some cases, which confused
2765 the lustre dentry revalidation. Change it to always use the
2769 Frequency : rare, unless heavy write-truncate concurrency is continuous
2770 Bugzilla : 4180, 6984, 7171, 9963, 9331
2771 Description: OST becomes very slow and/or deadlocked during object unlink
2772 Details : filter_destroy() was holding onto the parent directory lock
2773 while truncating+unlinking objects. For very large objects this
2774 may block other threads for a long time and slow overall OST
2775 responsiveness. It may also be possible to get a lock ordering
2776 deadlock in this case, or run out of journal credits because of
2777 the combined truncate+unlink. Solution is to do object truncate
2778 first in one transaction without parent lock, and then do the
2779 final unlink in a new transaction with the parent lock. This
2780 reduces the lock hold time dramatically.
2783 Frequency : rare, 2.4 kernels only
2785 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2786 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2787 before unmounting their filesystem in order to avoid deadlock.
2788 The kernel_locked() function in 2.4 kernels only checks whether
2789 the kernel lock is held, not whether it is this process that is
2790 holding it as 2.6 kernels do.
2795 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2796 Details : The client ptlrpc code may be trying to reconnect to a down
2797 server before a previous connection attempt has timed out.
2798 Increase the reconnect interval to be longer than the connection
2799 timeout interval to avoid sending duplicate connections to
2803 Frequency : echo_client brw_test command
2805 Description: fix echo_client to work with OST preallocated code
2806 Details : OST preallocation code (5137) didn't take echo_client IO path
2807 into account: echo_client calls filter methods outside of any
2808 OST thread and, hence, there is no per-thread preallocated
2809 pages and buffers to use. Solution: hijack pga pages for IO. As
2810 a byproduct, this avoids unnecessary data copying.
2814 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2815 Description: Client can oops in mdc_commit_close() after open replay
2816 Details : It was possible for the MDS to return an open request with no
2817 transaction number in mds_finish_transno() if the client was
2818 evicted, but without actually returning an error. Clients
2819 would later try to replay that open and may trip an assertion
2820 Simplify the client close codepath, and always return an error
2821 from the MDS in case the open is not successful.
2824 Frequency : rare, 2.6 OSTs only
2826 Description: OST may deadlock under high load on fragmented files
2827 Details : If there was a heavy load and highly-fragmented OST filesystems
2828 it was possible to have all the OST threads deadlock waiting on
2829 allocation of biovecs, because the biovecs were not released
2830 until the entire RPC IO was completed. Instead, release biovecs
2831 as soon as they are complete to ensure forward IO progress.
2833 Severity : enhancement
2835 Description: Support for specifying external journal device at mount
2836 Details : If an OST or MDS device is formatted with an external journal
2837 device, this device major/minor is stored in the ext3 superblock
2838 and may not be valid for failover. Allow detecting and
2839 specifying the external journal at mount time.
2844 Description: Mounting an MDS with pending unlinked files may cause oops
2845 Details : target_finish_recovery() calls mds_postrecov() which returned
2846 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2847 considers this an error and immediately begins cleaning up the
2848 lov, just after starting the mds_lov process
2850 Severity : enhancement
2852 Description: Implement 'lfs df' to report actual free space on per-OST basis
2853 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2854 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2855 usage of objects; '-h' to report in human readable format.
2857 ------------------------------------------------------------------------------
2859 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2866 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2867 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2868 at startup during orphan unlinking. Instead, skip buddy update
2869 before setup, as it will be regenerated after recovery is complete.
2874 Description: If an OST is inactive, its locks might reference stale inodes.
2875 Details : lov_change_cbdata() must iterate over all namespaces, even if
2876 they are inactive to clear inode references from the lock.
2878 Severity : enhancement
2879 Frequency : occasional, if non-standard max_dirty_mb used
2881 Description: Client will block write RPCs if not enough grant
2882 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2883 then the client will block writes while waiting for another RPC
2884 to complete instead of consuming its dirty limit. With change
2885 we get improved performance when max_dirty_mb is small.
2887 Severity : enhancement
2888 Bugzilla : 3389, 6253
2889 Description: Add support for supplementary groups on the MDS.
2890 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2891 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2892 lookups for user supplementary groups into a cache.
2896 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2897 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2898 for liblustre O_EXCL is enforced by MDS.
2902 Description: readdir never returns NULL in liblustre.
2903 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2904 dentry in struct dirent.
2908 Frequency : liblustre only, depends on application IO pattern
2909 Description: liblustre clients evicted if not contacting servers
2910 Details : Don't put liblustre clients into the ping_evictor list, so
2911 they will not be evicted by the pinger ever.
2913 Severity : enhancement
2915 Description: Add ability to evict clients by NID from MDS.
2916 Details : By echoing "nid:$NID" string into
2917 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2918 $NID would be instantly evicted from this MDS and from all active
2919 OSTs connected to it.
2923 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2924 Details : lookup_it_finish() used to query file size from OSTs that was not
2929 Description: service threads change working directory to that of init
2930 Details : Starting lustre service threads may pin the working directory
2931 of the parent thread, making that filesystem busy. Threads
2932 now change to the working directory of init to avoid this.
2936 Frequency : during shutdown only
2937 Description: shutdown with a failed MDS or OST can cause unmount to hang
2938 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2943 Frequency : 2.6 only
2944 Description: chmod/chown may include an extra supplementary group
2945 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2946 supplementary group and if none is specified this is used.
2949 Bugzilla : 5479 (6816)
2951 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2952 Details : If lookup is in progress on a file that is unlinked we might try
2953 to revalidate the inode and fail in revalidate after lookup is
2954 complete and ll_file_open() enqueues the open again but
2955 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2958 Frequency : always, if lconf --abort_recovery used
2960 Description: lconf --abort_recovery fails with 'Operation not supported'
2961 Details : lconf was attempting to abort recovery on the MDT device and not
2964 ------------------------------------------------------------------------------
2966 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2971 Frequency : rare (only unsupported configurations with a node running as an
2973 Bugzilla : 6514, 5137
2974 Description: Mounting a Lustre file system on a node running as an OST could
2976 Details : OSTs now preallocates memory needed to write out data at
2977 startup, instead of when needed, to avoid having to
2978 allocate memory in possibly low memory situations.
2979 Specifically, if the file system is mounted on on OST,
2980 memory pressure could force it to try to write out data,
2981 which it needed to allocate memory to do. Due to the low
2982 memory, it would be unable to do so and the node would
2983 become unresponsive.
2985 Severity : enhancement
2987 Description: Addition of lconf --service command line option
2988 Details : lconf now accepts a '--service <arg>' option, which is
2989 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2991 Severity : enhancement
2993 Description: Failover mode is now the default for OSTs.
2994 Details : By default, OSTs will now run in failover mode. To return to
2995 the old behaviour, add '--failout' to the lmc line for OSTs.
2997 Severity : enhancement
2999 Description: Health checks are now provided for MDS and OSTs
3000 Details : Additional detailed health check information on MSD and OSTs
3001 is now provided through the procfs health_check value.
3004 Frequency : occasional, depends on IO load
3006 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3007 after numerous create/delete cycles
3008 Details : The ext3 inode allocation policy would not allocate new inodes
3009 very well on the OSTs because there are no new directories
3010 being created. Instead we look for groups with free space if
3011 the parent directories are nearly full.
3016 Description: Network or server problems during mount may cause partially
3017 mounted clients instead of returning an error.
3018 Details : The config llog parsing code may overwrite the error return
3019 code during mount error handling, returning success instead
3025 Description: MDS can fail to allocate large reply buffers
3026 Details : After long uptimes the MDS can fail to allocate large reply
3027 buffers (e.g. zconf client mount config records) due to memory
3028 fragmentation or consumption by the buffer cache. Preallocate
3029 some large reply buffers so that these replies can be sent even
3030 under memory pressure.
3034 Frequency : rare (liblustre)
3035 Description: fsx running with liblustre complained that using truncate() to
3036 extend the file doesn't work. This patch corrects that issue.
3037 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3038 ATTR_SIZE and lsm use in llu_setattr_raw.
3042 Frequency : rare, only 2.6 kernels
3043 Description: Unusual file access patterns on the MDS may result in inode
3044 data being lost in very rare circumstances.
3045 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3046 code in the 2.6 kernel caused the inode and/or EA data not to
3047 be read from disk, causing single-file corruption.
3051 Frequency : rare, only 2.6 filesystems using extents
3052 Description: Heavy concurrent write and delete load may cause data corruption.
3053 Details : It was possible under high-load situations to have an extent
3054 metadata block in the block device cache from a just-unlinked
3055 file overwrite a newly-allocated data block. We now unmap any
3056 metadata buffers that alias just-allocated data blocks.
3060 Frequency : filesystems with default stripe_count larger than 77
3061 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3062 Details : lconf specifies an inode size of 4096 bytes when the default
3063 stripe_count is larger than 77. This conflicts with the default
3064 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3065 this case to avoid pinning too much memory for large EAs.
3067 ------------------------------------------------------------------------------
3069 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3074 Frequency : rare (extremely heavy IO load with hundreds of clients)
3076 Description: Client is evicted, gets IO error writing to file
3077 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3078 caused clients to be evicted for AST timeouts. The fixes in
3079 bug 5192 mean we no longer need to have such short AST timeouts
3080 so ldlm_timeout has been increased.
3083 Frequency : occasional during --force or --failover shutdown under load
3084 Bugzilla : 5949, 4834
3085 Description: Server oops/LBUG if stopped with --force or --failover under load
3086 Details : a collection of import/export refcount and cleanup ordering
3087 issues fixed for safer force cleanup
3090 Frequency : only filesystems larger than 120 OSTs
3091 Bugzilla : 5990, 6223
3092 Description: lfs getstripe would oops on a very large filesystem
3093 Details : lov_getconfig used kfree on vmalloc'd memory
3096 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3097 Bugzilla : 6242, 6243
3098 Description: reading from files that had been truncated to a non-zero size
3099 but never opened returned no data
3100 Details : ll_file_read() reads zeros from no-object files to EOF
3105 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3106 Details : There's little harm in aborting MDS/OSS recovery and letting it
3107 try again, so I removed the LASSERT and return an error instead.
3109 Severity : enhancement
3111 Description: New debugging infrastructure for tracking down data corruption
3112 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3113 (b) cover more of the client-side code path, and (c) try to narrow
3114 down where problems occurred
3118 Bugzilla : 3819, 4364, 4397, 6313
3119 Description: Racing close and eviction MDS could cause assertion in mds_close
3120 Details : It was possible to get multiple mfd references during close and
3121 client eviction, leading to one thread referencing a freed mfd.
3123 Severity: : enhancement
3124 Bugzilla : 3262, 6359
3125 Description: Attempts to reconnect to servers are now more aggressive.
3126 Details : This builds on the enhanced upcall-less recovery that was added
3127 in 1.4.2. When trying to reconnect to servers, clients will
3128 now try each server in the failover group every 10 seconds. By
3129 default, clients would previously try one server every 25 seconds.
3134 Description: After recovery, certain operations trigger a failed
3135 assertion on a client.
3136 Details : Failing over an mds, using lconf -d --failover, while a
3137 client was doing a readdir() call would cause the client to
3138 LBUG after recovery completed and the readdir() was resent.
3140 Severity : enhancement
3142 Description: Default groups are now added by lconf
3143 Details : You can now run lconf --group <servicename> without having to
3144 manually add groups with lmc.
3147 Frequency : occasional
3149 Description: Nodes with an elan id of 0 trigger a failed assertion
3152 Frequency : always when accessing e.g. tty/console device nodes
3154 Description: tty and some other devices nodes cannot be used on lustre
3155 Details : file's private_data field is used by device data and lustre
3156 values in there got lost. New field was added to struct file to
3157 store fs-specific private data.
3160 Frequency : when exporting Lustre via NFS
3162 Description: NFSD failed occasionally when looking up a path component
3163 Details : NFSD is looking up ".." which was broken in ext3 directories
3164 that had grown large enough to become hashed.
3167 Frequency : Clusters with multiple interfaces not on the same subnet
3169 Description: Nodes will repeatedly try to reconnect to an interface which it
3170 cannot reach and report an error to the log.
3171 Details : Extra peer list entries will be created by lconf with some peers
3172 unreachable. lconf now validates the peer before adding it.
3175 Frequency : Only if a default stripe is set on the filesystem root.
3177 Description: Setting a default stripe on the filesystem root prevented the
3178 filesystem from being remounted.
3179 Details : The client was sending extra request flags in the root getattr
3180 request and did not allocate a reply buffer for the dir EA.
3183 Frequency : occasional, higher if lots of files are accessed by one client
3184 Bugzilla : 6159, 6097
3185 Description: Client trips assertion regarding lsm mismatch/magic
3186 Details : While revalidating inodes the VFS looks up inodes with ifind()
3187 and in rare cases can find an inode that is being freed.
3188 The ll_test_inode() code will free the lsm during ifind()
3189 when it finds an existing inode and then the VFS later attaches
3190 this free lsm to a new inode.
3194 Bugzilla : 6422, 7030
3195 Description: MDS deadlock between mkdir and client eviction
3196 Details : Creating a new file via mkdir or mknod (starting a transaction
3197 and getting the ns lock) can deadlock with client eviction
3198 (gets ns lock and trying to finish a synchronous transaction).
3201 Frequency : occasional
3202 Description: While starting a server, the fsfilt_ext3 module could not be
3204 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3205 kernels. Previously, lconf would still use the ext3 name
3206 when trying to load modules. Now, it will correctly use
3207 ext3 on 2.4 and ldiskfs on 2.6.
3209 Severity : enhancement
3210 Description: The default stripe count has been changed to 1
3211 Details : The interpretation of the default stripe count (0, to lfs
3212 or lmc) has been changed to mean striping across a single
3213 OST, rather than all available. For general usage we have
3214 found a stripe count of 1 or 2 works best.
3216 Severity : enhancement
3217 Description: Add support for compiling against Cray portals.
3218 Details : Conditional compiling for some areas that are different
3222 Frequency : occasional
3223 Bugzilla : 6409, 6834
3224 Description: Creating files with an explicit stripe count may lead to
3225 a failed assertion on the MDS
3226 Details : If some OSTs are full or unavailable, creating files may
3227 trigger a failed assertion on the MDS. Now, Lustre will
3228 try to use other servers or return an error to the
3232 Frequency : occasional
3234 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3235 caused lock timeout during readahead (since 1.4.2).
3236 Details : Processes doing readahead might match a lock that hasn't been
3237 granted yet if there are overlapping and conflicting lock
3238 requests. The readahead process waits on ungranted lock
3239 (original lock is CBPENDING), while OST waits for that process
3240 to cancel CBPENDING read lock and eventually evicts client.
3242 Severity : enhancement
3244 Description: Initial enabling of flock support for clients
3245 Details : Implements fcntl advisory locking and file status functions.
3246 This feature is provided as an optional mount flag (default
3247 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3248 locking are implemented yet, and those that are are not guaranteed
3249 to be completely correct in production environments.
3250 mount -t lustre -o [flock|noflock] ...
3253 Frequency : occasional
3255 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3256 assertion in the ext3 JBD (journaling) layer.
3257 Details : The b_committed_data struct is protected by the big kernel lock
3258 in 2.4 kernels, serializing journal_commit_transaction() and
3259 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3260 access to this struct. In 2.6 kernels there is finer grained
3261 locking to improve SMP performance of the JBD layer.
3265 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3270 Description: merge and cleanup kernel patches.
3273 -----------------------------------------------------------------------------
3275 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3277 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3278 versions of Lustre. Please update all servers and clients to
3279 version 1.4.2 or later at the same time. You must also run
3280 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3281 to update the configuration logs.
3283 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3284 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3285 - swab llog records of type '0' so we get proper header size/idx (5861)
3286 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3287 - fix rename of one directory over another leaking an inode (5953)
3288 - avoid SetPageDirty on 2.6 (5981)
3289 - don't re-add just-being-destroyed locks to the waiting list (5653)
3290 - when creating new directories, inherit the parent's custom
3291 striping settings if present parent (3048)
3292 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3293 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3294 - don't hold client locks on temporary worklist from l_lru (5666)
3295 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3296 - automatically evict dead clients (5921)
3297 - Update file size properly in create+truncate+fstat case (6196)
3298 - Do not unhash mountpoint dentries, do not allow removal of
3300 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3301 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3302 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3303 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3304 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3305 - Correctly handle reads of files with no objects (6243)
3306 - lctl recover will also mark a device active if deactivate used (5933)
3308 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3309 - allow --write-conf on an MDS with different nettype than client (5619)
3310 - don't write config llogs to MDS for mounts not from that MDS (5617)
3311 - lconf should create multiple TCP connections from a client (5201)
3312 - init scripts are now turned off by default; run chkconfig --on
3313 lustre and chkconfig --on lustrefs to use them
3314 - upcalls are no longer needed for clients to recover to failover
3316 - add --abort-recovery option to lconf to abort recovery on device
3318 - add support for an arbitrary number of OSTs (3026)
3319 - Quota support protocol changes.
3320 - forward compatibility changes to wire structs (6007)
3321 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3322 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3323 - improved "lustre status" script
3324 - initialize blocksize for non-regular files (6062)
3325 - added --disable-server and --disable-client configure options (5782)
3326 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3327 - Vanilla 2.4.29 support
3328 - increase maximum number of obd devices to 520 (6242)
3329 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3330 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3331 SLES 9 2.6.5 kernels
3332 - limit stripes per file to 160 (the maximum EA size) (6093)
3334 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3337 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3338 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3339 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3340 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3341 - small ext3 extents cleanups and fixes (5733)
3342 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3343 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3344 - avoid needless client->OST connect, fix handle mismatch (5317)
3345 - fix DLM error path that led to out-of-sync client, long delays (5779)
3346 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3347 - fix several locking issues related to i_size (5492,5624,5654,5672)
3348 - don't move pending lock onto export if it is already evicted (5683)
3349 - fix kernel oops when creating .foo in unlinked directory (5548)
3350 - fix deadlock in obdfilter statistics vs. object create (5811)
3351 - use time_{before,after} to avoid timer jiffies wrap (5882)
3352 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3353 - Do not leak request if server was not able to process it (5154)
3354 - If mds_open unable to find parent dir, make that negative lookup(5154)
3355 - don't create new directories with extent-mapping (5909, 5936)
3357 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3358 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3359 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3360 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3361 - support for RHEL4 kernel on i686 (5773)
3362 - provide error messages when incompatible logs are encountered (5898)
3364 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3365 * version 1.4.0.10 (1.4.1 release candidate 1)
3367 - don't keep a lock reference when lock is not granted (4238)
3368 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3369 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3370 - drop import inflight refcount on signal_completed_replay error (5255)
3371 - unlock page after async write error during send (3677)
3372 - handle missing objects in filter_preprw_read properly (5265)
3373 - no transno return for symlink open, don't save no-trasno open (3440)
3374 - don't try to complete elan receive that already failed (4012)
3375 - free RPC server reply state on error (5406)
3376 - clean up thread from ptlrpc_start_thread() on error (5160)
3377 - readahead could read extra page into cache that wasn't ejected (5388)
3378 - prevent races in class_attach/setup/cleanup/detach (5260)
3379 - don't dereference de->d_inode after l_dput of de (5458)
3380 - use "int" for stripe value returned from lock_to_stripe (5544)
3381 - mballoc allocation and error-checking fixes in 2.6 (5504)
3382 - block device patches to fix I/O request sizes in 2.6 (5482)
3383 - look up hostnames for IB nals (5602)
3384 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3385 - don't start multiple acceptors for the same port (5277)
3386 - fix incorrect LASSERT in mds_getattr_name (5635)
3387 - export a proc file for general "ping" checking (5628)
3388 - fix "lfs check" to not block when the MDS is down (5628)
3390 - service request history (4965)
3391 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3392 - create an "evict_client" /proc entry on OSTs, like the MDS has
3393 - fix mount usage message, return errors per mount(8) (5168)
3394 - change grep [] to grep "[]" in tests so they work in more UMLs
3395 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3396 - remove ancient LOV_MAGIC_V0 EA support (5047)
3397 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3398 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3399 - allow readahead allocations to fail when low on memory (5383)
3400 - mmap locking landed again, after considerable improvement (2828)
3401 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3403 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3406 - send OST transaction number in read/write reply to free req (4966)
3407 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3408 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3409 - the watchdog thread now runs as interruptible (5246)
3410 - flock/lockf fixes (but it's still disabled, pending 5135)
3411 - don't use EXT3 constants in llite code (5094)
3412 - memory shortage at startup could cause assertion (5176)
3414 - reorganization of lov code
3415 - single portals codebase
3417 - add extents/mballoc support (5025)
3418 - direct I/O reads in the obdfilter (4048)
3419 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3421 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3424 - send OST transaction number in read/write reply to free req (4966)
3425 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3426 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3427 - the watchdog thread now runs as interruptible (5246)
3428 - handle missing objects in filter_preprw_read properly (5265)
3429 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3430 - drop import inflight refcount on signal_completed_replay error (5255)
3431 - unlock page after async write error during send (3677)
3432 - return original error code on reconstructed replies (3761)
3433 - no transno return for symlink open, don't save no-trasno open (3440)
3435 - add pid to ldlm debugging output (4922)
3436 - bump the watchdog timeouts -- we can't handle 30sec yet
3437 - extra debugging for orphan dentry/inode bug (5259)
3439 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3442 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3443 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3444 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3445 - fix NULL dereference in /proc/sys/portals/routes (4827)
3446 - allow failed mdc_close() operations to be interrupted (4561)
3447 - stop precreate on OST before MDS would time out on it (4778)
3448 - don't send partial-page writes before EOF from client (4410)
3449 - discard client grant for sub-page writes on large-page clients (4520)
3450 - don't free dentries not owned by NFS code, check generation (4806)
3451 - fix lsm leak if mds_create_objects() fails (4801)
3452 - limit debug_daemon file size, always print CERROR messages (4789)
3453 - use transno after validating reply (3892)
3454 - process timed out requests if import state changes (3754)
3455 - update mtime on OST during writes, return in glimpse (4829)
3456 - add mkfsoptions to LDAP (4679)
3457 - use ->max_readahead method instead of zapping global ra (5039)
3458 - don't interrupt __l_wait_event() during strace
3460 - add software watchdogs to catch hung threads quickly (4941)
3461 - make lustrefs init script start after nfs is mounted
3462 - fix CWARN/ERROR duplication (4930)
3463 - return async write errors to application if possible (2248)
3464 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3465 - print NAL number in %x format (4645)
3466 - update barely-supported suse-2.4.21-171 series (4842)
3467 - support for sles 9 %post scripts
3468 - support for building 2.6 kernel-source packages
3469 - support for sles km_* packages
3471 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3474 - ignore -ENOENT errors in osc_destroy (3639)
3475 - notify osc create thread that OSC is being cleaned up (4600)
3476 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3477 - reconstruct ost_handle() like mds_handle() (4657)
3478 - create a new thread to do import eviction to avoid deadlock (3969)
3479 - let lconf resolve symlinked-to devices (4629)
3480 - don't unlink "objects" from directory with default EA (4554)
3481 - hold socknal file ref over connect in case target is down (4394)
3482 - allow more than 32000 subdirectories in a single directory (3244)
3483 - fix blocks count for O_DIRECT writes (3751)
3484 - OST returns ENOSPC from object create when no space left (4539)
3485 - don't send truncate RPC if file size isn't changing (4410)
3486 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3487 - bind to privileged port in socknal and tcpnal (3689)
3489 - rate limit CERROR/CWARN console message to avoid overload (4519)
3490 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3491 - basic mmap support (3918)
3492 - kernel patch series update from b1_4 (4711)
3494 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3497 - avoid crash during MDS cleanup with OST shut down (2775)
3498 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3499 - don't use bad inodes on the MDS (3744)
3500 - dynamic object preallocation to improve recovery speed (4236)
3501 - don't hold spinlock over lock dumping or change debug flags (4401)
3502 - don't zero obd_dev when it is force cleaned (3651)
3503 - print grants to console if they go negative (4431)
3504 - "lctl deactivate" will stop automatic recovery attempts (3406)
3505 - look for existing locks in ldlm_handle_enqueue() (3764)
3506 - don't resolve lock handle twice in recovery avoiding race (4401)
3507 - revalidate should check working dir is a directory (4134)
3509 - don't always mark "slow" obdfilter messages as errors (4418)
3511 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3514 - don't close LustreDB during write_conf until it is done (3860)
3515 - fix typo in lconf for_each_profile (3821)
3516 - allow dumping logs from multiple threads at one time (3820)
3517 - don't allow multiple threads in OSC recovery (3812)
3518 - fix debug_size parameters (3864)
3519 - fix mds_postrecov to initialize import for llog ctxt (3121)
3520 - replace config semaphore with spinlock (3306)
3521 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3522 - don't allow enqueue to complete on a destroyed export (3822)
3523 - down write_lock before checking llog header bitmap (3825)
3524 - recover from lock replay timeout (3764)
3525 - up llog sem before sending rpc (3652)
3526 - reduce ns lock hold times when setting kms (3267)
3527 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3528 - fix NULL deref and obd_dev leak on setup error (3312)
3529 - replace some LBUG about llog ops with error handling (3841)
3530 - don't match INVALID dentries from d_lookup and spin (3784)
3531 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3532 - fix invalid assertion in ptlrpc_set_wait (3880)
3534 - add libwrap support for the TCP acceptor (3996)
3535 - add /proc/sys/portals/routes for non-root route listing (3994)
3536 - allow setting MDS UUID in .xml (2580)
3537 - print the stack of a process that LBUGs (4228)
3539 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3542 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3543 - make sure to unset replay flag from failed open requests (3440)
3544 - if default stripe count is 0, use OST count for inode size (3636)
3545 - update parent mtime/ctime on client for create/unlink (2611)
3546 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3547 - free recovery state on server during a forced cleanup (3571)
3548 - unregister_reply for resent reqs (3063)
3549 - loop back devices mounting and status check on 2.6 (3563)
3550 - fix resource-creation race that can provoke i_size == 0 (3513)
3551 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3552 - more debugging for page-accounting assertion (3746)
3553 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3554 - avoid deadlock after precreation failure (3758)
3555 - fix race and lock order deadlock in orphan handling (3450, 3750)
3556 - add validity checks when grabbing inodes from l_ast_data (3599)
3558 - add /proc/.../recovery_status to obdfilter (3428)
3559 - lightweight CDEBUG infrastructure, debug daemon (3668)
3560 - change default OSC RPC parameters to be better on small clusters
3561 - turn off OST read cache for files smaller than 32MB
3562 - install man pages and include them in rpms (3100)
3563 - add new init script for (un)mounting lustre filesystems (2593)
3564 - run chkconfig in %post for init scripts (3701)
3565 - drop scimac NAL (unmaintained)
3567 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3570 - clean kiobufs before and after use (3485)
3571 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3572 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3573 - change init script to fail more clearly if not run as root (1528)
3574 - allow clients to reconnect during replay (1742)
3575 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3576 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3577 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3578 - fix cond_resched (3554)
3579 - only evict unfinished clients after recovery (3515)
3580 - allow bulk resend, prevent data loss (3570)
3581 - dynamic ptlrpc request buffer allocation (2102)
3582 - don't allow unlinking open directory if it isn't empty (2904)
3583 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3584 - remove extraneous obd dereference causing LASSERT failure (3334)
3585 - don't use get_cycles() when creating temp. files on the mds (3156)
3586 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3587 - handle EEXIST for set-stripe, set proper directory name (3336)
3589 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3590 - fix ksocknal_fmb_callback() error messages (2918)
3592 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3595 - don't copy lvb into (possibly NULL) reply on error (2983)
3596 - don't deref dentry after dput, don't free lvb on error (2922)
3597 - use the kms to determine writeback rpc length (2947)
3598 - increment oti_logcookies when osc is inactive (2948)
3599 - update client's i_blocks count via lvb messages (2543)
3600 - handle intent open/close of special files properly (1557)
3601 - mount MDS with errors=remount-ro, like obdfilter (2009)
3602 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3603 - don't use cancelling-locks' kms values (2947)
3604 - use highest lock extent for kms, not last one (2925)
3605 - don't dereference ERR_PTR() dentry in error handling path (3107)
3606 - fix thread race in portals_debug_dumplog() (3122)
3607 - create lprocfs device entries at setup instead of at attach (1519)
3608 - common AST error handler, don't evict client on completion race (3145)
3609 - zero nameidata in detach_mnt in 2.6 (3118)
3610 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3611 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3612 - initialize RPC timeout timer earlier for 2.6 (3219)
3613 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3614 - print nal/nid for unknown nid (3258)
3615 - additional checks for oscc recovery before doing precreate (3284)
3616 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3617 - don't crash in mdc_close for bad permissions on open (3285)
3618 - zero i_rdev for non-device files (3147)
3619 - clear page->private before handing to FS, better assertion (3119)
3620 - tune the read pipeline (3236)
3621 - fix incorrect decref of invalidated dentry (2350)
3622 - provide read-ahead stats and refine rpc in flight stats (3328)
3623 - don't hold journal transaction open across create RPC (3313)
3624 - update atime on MDS at close time (3265)
3625 - close LDAP connection when recovering to avoid server load (3315)
3626 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3627 - don't leak open file on MDS after open resend (3325)
3628 - serialize filter_precreate and filter_destroy_precreated (3329)
3629 - loop device shouldn't call sync_dev() for nul device (3092)
3630 - clear page cache after eviction (2766)
3631 - resynchronize MDS->OST in background (2824)
3632 - refuse to mount the same filesystem twice on same mountpoint (3394)
3633 - allow llmount to create routes for mounting behind routers (3320)
3634 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3635 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3636 - fix rare problem with rename on htree directories (3417)
3638 - allow default OST striping configuration per directory (1414)
3639 - fix compilation for qswnal for 2.6 kernels (3125)
3640 - increase maximum number of MDS request buffers for large systems
3641 - change liblustreapi to be useful for external progs like lfsck (3098)
3642 - increase local configuration timeout for slow disks (3353)
3643 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3645 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3648 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3649 - don't overwrite extent policy data in reply if lock was blocked (2901)
3650 - drop filter export grants atomically with removal from device (2663)
3651 - del obd_self_export from work_list in class_disconnect_exports (2908)
3652 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3653 - swab reply message in mdc_close, other PPC fixes (2464)
3654 - fix destroying of named logs (2325)
3655 - overwrite old logs when running lconf --write_conf (2264)
3656 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3657 - fix race in target_handle_connect (2898)
3658 - mds_reint_create() should take same inode create lock (2926)
3659 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3660 - don't close files for self_export to avoid uninitialized obd (2936)
3661 - allow MDS with the same name as client node (2939)
3662 - hold dentry reference for closed log files for unlink (2325)
3663 - reserve space for all logs during transactions (2059)
3664 - don't evict page beyond end of stripe extent (2925)
3665 - don't oops on a deleted current working directory (2399)
3666 - handle hard links to targets without a parent properly (2517)
3667 - don't dereference NULL lock when racing during eviction (2867)
3668 - don't grow lock extents when lots of conflicting locks (2919)
3670 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3673 - account for cache space usage on clients to avoid data loss (974)
3674 - lfsck support in lustre kernel code (2349)
3675 - reduce journal credits needed for BRW writes (2370)
3676 - orphan handling to avoid losing space on client/server crashes
3677 - ptlrpcd can be blocked, stopping ALL progress (2477)
3678 - use lock value blocks to assist in proper KMS, faster stat (1021)
3679 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3680 - recovery for initial connections (2355)
3681 - fixes for mds_cleanup_orphans (1934)
3682 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3683 - block all file creations until orphan recovery completes (1901)
3684 - client remove rq_connection from request struct (2423)
3685 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3686 - recovery timer race (2670)
3687 - mdc_close recovey bug (2532)
3688 - ptlrpc cleanup bug (2710)
3689 - mds timeout on local locks (2588)
3690 - namespace lock held during RPCs (2431)
3691 - handle interrupted sync write properly (2503)
3692 - don't try to handle a message that hasn't been replied to (2699)
3693 - client assert failure during cleanup after abort recovery (2701)
3694 - leak mdc device after failed mount (2712)
3695 - ptlrpc_check_set allows timedout requests to complete (2714)
3696 - wait for inflight reqs when ptlrpcd finishes (2710)
3697 - make sure unregistered services are removed from the srv_list
3698 - reset bulk XID's when resending them (caught by 1138 test)
3699 - unregister_bulk after timeout
3700 - fix lconf error (2694)
3701 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3702 - readahead locks pages, leaves pending causing memory pressure (2673)
3703 - increase OST request buffers to 4096 on large machines (2729)
3704 - fix up permission of existing directories in simple_mkdir (2661)
3705 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3706 - don't assign transno to errored transactions (2742)
3707 - don't delete objects on OST if given a bogus objid from MDS (2751)
3708 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3709 - if rq_no_resend, then timeout request after recovery (2432)
3710 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3711 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3712 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3713 - MDS recovery shouldn't delete objects that it has given out (2730)
3714 - if enqueue arrives after completion, don't clobber LVB (2819)
3715 - don't unlock pages twice when trigger_group_io returns error (2814)
3716 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3717 - don't write pages to disk if there was an error (1450)
3718 - don't ping imports that have recovery disabled (2676)
3719 - take buffered bytes into account when balancing socknal conn (2817)
3720 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3721 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3722 - remove little-endian swabbing of llog records (1987)
3723 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3724 - retry reposting request buffers if they fail (1191)
3725 - grow extent at grant time to avoid granting a revoked lock (2809)
3726 - lock revoke doesn't evict page if covered by a second lock (2765)
3727 - disable VM readahead to avoid reading outside lock extents (2805)
3729 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3730 - updated kernel patches for hp-2.4.20 kernel (2681)
3732 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3735 - fix truncated write corruption (2366)
3736 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3738 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3739 - ensure len(uuid) < 37 in lmc (1171)
3740 - fix ia64 OOPS in llog_test (2255)
3741 - zero end of page at obdfilter for partial page writes (2648)
3742 - don't leave stale dentries around after renames (bug 2428)
3743 - fix timeouts when evicting a client with a single lock held (2642)
3744 - set deadline for the initial HELLO message to drain (2634)
3745 - print out dotted-quad IP addresses in the socknal (2302)
3747 - additional debugging for MDS client eviction problem (2443)
3748 - fix mkfsoptions support for osts (2603, 2604)
3750 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3753 - add series for the vanilla 2.6.0 kernel
3754 - add series for the vanilla 2.4.24 kernel
3755 - add series for a cray x86/64 UL kernel drop
3756 - fix xattr patches for the vanilla 2.4.19 series
3758 - generate true UUIDs in lmc (1171)
3759 - have portals stack dumping break in UML (2466)
3760 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3761 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3762 - wake callback waiting threads on client eviction (2460)
3763 - Add --ptldebug and --subsystem to lmc (1719)
3764 - update assertion to allow safe interrupt allocation
3765 - set rq_no_resend for cancel requests (2432)
3766 - recalculate ptlrpcd timeout after resend (2494)
3767 - call vfs_rmdir when removing pending directories (2368)
3768 - fix renaming a file to itself (2429)
3769 - lmc creates a default one-stripe lov (2454)
3770 - expand procfs space to handle large clusters (2326)
3771 - increase UML stack to avoid overflow
3772 - update lconf's list of debug and subsystem masks
3773 - fix lfs find --obd (2510)
3774 - /proc tunable for disabling filter read caching (2591)
3775 - stop rpm packages from altering slapd.conf (2301)
3776 - disable nagle in the socknal under 0conf (2578)
3777 - choose mds inode size based on stripe count (2572)
3778 - fix kernel-source rpm problems (2516)
3780 - add --disable-doc to avoid pdf generation (2421)
3781 - update documentation, tests, type-os, comments
3782 - avoid format warnings on ia64
3783 - remove the TOE NAL
3784 - tiny code cleanups by removing unused fields
3786 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3789 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3790 - remove the most blatant lies from BUILDING file (2371)
3791 - change default debug level to reasonable production setting
3792 - reduce client side cache size to reduce cache flush time
3793 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3794 - make TCP zerocopy and pinger support enabled by default (2476)
3795 - sync writes completed after process exits caused crashes (2319)
3796 - maintain correct mount count on the MDS (2356)
3797 - backout 1557, because 2316 wasn't really fixed
3798 - better file I/O statistics gathering in /proc
3799 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3800 - another kernel patch to fix zero-copy TCP function export
3801 - don't take duplicate lock when processing re-sent getattr (2420)
3802 - lctl uses obd_self_export instead of creating new conn (2353)
3803 - MDS/OST recovery case which requires object creation asserted (2425)
3804 - move lfs from /usr/sbin to /usr/bin in packages
3805 - fix race between mds_client_add and mds_client_free (2417)
3806 - use kmalloc instead of slabs in portals (2430)
3807 - don't create duplicate records when a failover MDS is present (2442)
3808 - remove unnecessary mount age check (2332)
3809 - don't remove directory inodes from locks prematurely (2451)
3810 - don't break if MDS service name is the same as hostname (2103)
3811 - fix races in client write RPC generation when cache full (2482)
3813 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3816 - remove now-unused request->rq_obd (278)
3817 - if an allocation fails, print out how much memory we've used (1933)
3818 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3819 - add the "configurable stack size" patch to most series files (1256)
3820 - ability to write large log records, for 100+ OST configs (2306)
3821 - fix NULL deref when filter_prep fails (2314)
3822 - fix operator precedence error in filter_sync
3823 - dynamic allocation of socknal TX descriptors (2315)
3824 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3825 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3826 - ensure that CWARN messages in Portals always get to the syslog
3827 - __init/__exit are not for prototype decls (ldlm_init/exit)
3828 - x86-64 compile warning fixes
3829 - fix gateway LMC keyword conflict (2318)
3830 - fix MDS lock inversions in getattr/reint paths (1844)
3831 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3832 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3833 - DLM race condition prevented some lock evictions (2328)
3834 - ENOMEM detection and retry on socknal sends (2230)
3835 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3836 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3837 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3838 - better error messages when a client is rejected during recovery (1505)
3839 - avoid cancelling locks which were never granted, after failure (2330)
3840 - fix i_sem/journal inversion in mds_client_add (2333)
3841 - fix truncate/getattr lock cycle deadlock (2334)
3842 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3843 - fix two rare exit paths which could leak an l_lock() ref (2321)
3844 - fencepost error in MDS/OST orphan recovery (2226)
3845 - make log record alignment 8 bytes (1988)
3846 - lstripe now fails when requested offset > ost_count (2237)
3847 - ensure that all kernel series have a complete list.h (1607)
3848 - fix crashes in special-file operations (2316)
3849 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3851 - allow configurable automake binary, for testing new versions
3852 - small update to the lfs documentation
3854 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3856 * fix negative export reference count in fsfilt_sync (2312)
3858 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3859 * release candidate 0.9.1
3861 - orphans are moved into the PENDING directory for possible recovery
3862 - replayed opens now open by fid for orphan/rename safety (1042)
3863 - last close of an orphan inode generates a transno (683)
3864 - chdir() and mount() now pin the directory entry (1020)
3865 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3866 - discard very old requests without processing them (1502)
3867 - remove some common, well-understood CERRORs (1505)
3868 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3869 - clear "grant" flags in OST replies until OST grant code lands (1644)
3870 - fix read performance by not clobbering i_blksize on client (1598)
3871 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3872 - make LRU size a /proc tunable, clears locks when reduced (707)
3873 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3874 - server locks take references on exports now (1558)
3875 - build fixes for 2.4.20-rh trees (1663)
3876 - return an error from lov_create if all OSCs are inactive (1751)
3877 - fix import levels when a reconnect happens without a timeout (1597)
3878 - exit early from mds_open if we get a lookup error (1749)
3879 - partial page read at EOF wouldn't wait for disk before sending (1642)
3880 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3881 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3882 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3883 - write/truncate lock inversion (1639)
3884 - Don't auto-load obdclass, portals modules during cleanup (1495)
3885 - fix timestamps from jumping to "now" (1763)
3886 - extra journal assertions (1648)
3887 - add an extra multiunlink test (1771)
3888 - fix read_record/write_record API (1776)
3889 - fix leak of offset_extent, possible incorrect i_size later (1772)
3890 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3891 - Debugging for the kqswnal_get_idle_tx problems (1820)
3892 - Allow recovery to be attempted multiple times (1536)
3893 - Write out MDS last_rcvd file after it is first created (1600)
3894 - Fix tx_descriptor leak in failed transmit situations (1827)
3895 - ext3 journaling fixes for assertion failure after IO error (1871)
3896 - class_export_put() on freed export after completion AST error (1896)
3897 - Fix revalidate looping in VFS (1322)
3898 - Don't access a freed export during MDS_REINT timeout (1521)
3899 - Add open-unlink recovery support on the MDS (1673,1764)
3900 - Return an error if no MDS data was read from last_rcvd (1946)
3901 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3902 - Don't setup a disk device that doesn't match exported UUID (317)
3903 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3904 - avoid committing NULL handle in force close
3905 - local.sh is now a one-stripe LOV configuration
3906 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3907 - fix ext3 htree duplicate directory entry corruption (1516)
3908 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3909 - update to new LOV EA format (2097)
3910 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3912 2003-06-15 Phil Schwan <phil@clusterfs.com>
3915 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3916 - per-import/export recovery handling (958, 931, 959)
3917 - multiple last-rcvd slots, for serving multiple FSes (949)
3918 - connections are again shared between multiple imp/exports (963, 964)
3919 - "umount -f" would hang if any requests needed to be sent (393, 978)
3920 - avoid pinning large req buffer by copying for queued messages (989)
3921 - add "uuid" to "lctl device" command to help upcalls (991)
3922 - "open" RPCs with transnos would confuse recovery counters (1037)
3923 - do proper endian conversion of all wire messages (288, 340, 891)
3924 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3925 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3926 - fix ptlrpc_connection leak in target_handle_connect (1174)
3927 - fix import refcounting bug in OST and MDS cleanup (1134)
3928 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3929 - fix very unlikely obd_types race condition (501)
3930 - remove osc_open hack for echo_client (1187)
3931 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3932 - a failure in read_inode2 leads to deadlock (1139)
3933 - cancel ack-locks as soon as transaction is committed (1072)
3934 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3935 - make sure to commitrw after any preprw to avoid deadlock (1162)
3936 - failing to execute a file in a lustre FS would lock inode (1203)
3937 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3938 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3939 - obd_uuid2tgt would walk off the end of the list (1255)
3940 - on IA64 the osc would give portals incorrect bulk size (1258)
3941 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3942 - fix lock inversion caused by new llite matching code (1282)
3943 - limit the number of dirty pages on a client to 10MB (1286)
3944 - timed out locks were not being corrected cancelled (1289)
3945 - fix O_DIRECT above 4GB on IA-32 (1292)
3946 * major user-visible changes
3947 - fail out/fail over policy now controlled by the upcall (993)
3949 - add OBD_PING to check server availability and failure (954)
3950 - lustre messages are now sent in sending host order (288, 340, 891)
3951 - add eadatalen to MDS getattr reply (340)
3952 - OST read replies may contain second buffer, with per-page status (593)
3954 2003-03-11 Phil Schwan <phil@clusterfs.com>
3957 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3958 - failed open()s could cause deadlock; fixed (867, 869)
3959 - stop cancelling OST locks when files are closed (481)
3960 - overlapping XID spaces caused network corruption (851, 853)
3961 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3962 - setattr_raw added, to do single-RPC, server-side setattrs
3963 - lmc/lconf syntax change for OST UUIDs
3964 - fix crashy race condition between ptlrpc_free_req and osc_close
3965 - don't use request in mdc_enqueue if we hit a timeout (889)
3966 - don't set the inode i_size for regular files from the MDS (896)
3967 - handle out of order completion AST (842)
3968 - don't LBUG if a lock request times out after receiving AST (913)
3969 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3970 - if a bad lock AST arrives, send an error instead of dropping entirely
3971 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3972 - fix leak in bulk IO when only partially completed (899, 900, 926)
3973 - fix O_DIRECT for ia64 (55)
3974 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3975 - C-z after timeout could hang a process forever; fixed (977)
3977 - client-side I/O cache (678, 924, 929, 941, 970)
3979 - READPAGE and SETATTRs which don't take server-side locks get
3982 2003-02-11 Phil Schwan <phil@clusterfs.com>
3985 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3986 - Fix fsfilt_extN_readpage() to read a full page of directory
3987 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3988 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3989 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3990 - Most intent operations were removed, in favour of a new RPC mode
3991 that does a single RPC to the server and bypasses most of the VFS
3992 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3993 - Aggressively cancel local locks on DLM servers
3994 - mds_reint_unlink sends EA to the client if it's the last nlink.
3995 client uses that EA to unlink OST objects.
3996 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3997 - recursive symlinks were fixed (439)
3998 - fixed NULL deref in DEBUG_REQ
3999 - filter_update_lastobjid no longer calls sync, which annoyed extN
4000 - fixed multi-client small-writes to a single file problem (445)
4001 - fixed mtime updates during file writes (607)
4002 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4003 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4004 - fixed Linux OST size reporting problem (444, 656)
4005 - OST now updates object mtime with writes or setattr (607, 619)
4006 - client verifies file size before zeroing page past EOF (445)
4007 - OST now writes last allocated objid to disk with allocation (108)
4008 - LOV on echo now works (409)
4010 - mds_reint_unlink sends a new buffer, with the EA included. this
4011 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4012 set if a regular file was being unlinked, and it was the last link
4013 - use PtlGet from the target for bulk writes (315)
4014 - OST now updates object mtime with writes or setattr (607, 619)
4015 - LDLM now has a grant-time callback to revalidate locked items, if
4017 - Many MDS operations were reorganized to combat race conditions
4019 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4020 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4022 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4025 - Fully reactivate OST imports after reconnection (512, others)
4026 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4027 - More graceful error handling for truncating on dead OST (515)
4028 - Don't error out unless we're actually accessing dead stripes (474)
4029 - Fix garbage sizes when stripes are missing (410)
4030 - LRU counters were broken, causing constant lock purge (433, 432)
4031 - garbage on read from stripes with failed OSTs (441)
4032 - mark OSCs as active before reconnecting during recovery (438)
4033 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4034 - lfind did not preserve OST order in output (443)
4035 - symlinks cause hung clients, incorrect data (439)
4036 - stop dereferencing request after dropping refcount (457)
4037 - don't LASSERT(spin_is_locked) on non-SMP (455)
4038 - fixes for many rename() bugs
4039 - fstat didn't correctly synchronize attributes (399)
4040 - server must handle lock cancellation during blocking AST prep (487)
4041 - bulk descriptors were free()d too soon (511)
4042 - fix paths in lconf, which would load incorrect modules (451, 507)
4043 - fix confusing lconf 'host not found' error message (386)
4044 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4045 - fix race condition in mdc_blocking_ast() for inode access (526)
4046 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4047 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4048 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4049 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4050 - fix lstripe in several ways (kernel side) (527)
4051 - fix request leak in ldlm_cli_enqueue (262)
4052 - incorrect OSC was marked inactive after OST failure
4053 - call mds_fs_cleanup before unmounting filesystem (524)
4054 - fix races between taking ns_lock and ldlm_lock_change_resource
4055 - fix races updating LOV export open file list
4056 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4057 - fix recovery NULL deref in ldlm_cli_cancel_unused
4058 - fix some DLM races by using new hash table for lock handles (419)
4059 - permit the client to specify desired inodes, at replay
4060 - duplicate requests when we queue them for replay reintegration
4061 - fix last_rcvd offset calculation
4062 - sync after each recovered transaction, so we always make progress
4063 - never, not always, ERESTART requests without transnos
4064 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4066 - skip replay if the MDS says that the client is already connected
4067 - don't check for a recovery-enabled export to match lctl's UUID
4068 - don't INC_USE_COUNT for phantom exports
4069 - don't crash when cleaning up phantom exports (567)
4070 - don't double-finish or set replay data for errored mdc_open requests
4071 - abort requests when they time out, so we don't get old replies
4072 - send/receive replies for AST messages again
4073 - if the client says that it doesn't have the lock, cancel it on the
4075 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4076 mark it as destroyed, it will all work out when decref is called
4077 - fix module use counts (22, 581)
4079 - ASTs now expect a reply (server cancels lock on error reply)
4081 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4084 - fix many simultaneous client startup (392)
4085 - fix dentry->d_it clobbering
4086 - credentials weren't being shipped for readdir/getattr operations
4087 - remove invalid assertions triggered during some concurrent MD
4089 - proper Lustre versions added (336, 389)
4090 - fix memory leak for create error case (398)
4091 - fix LOV locking bug that would get cli/srv out of sync
4092 - fix echo client over LOV (409)
4093 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4094 - fix double-O_EXCL intent crash (424)
4095 - avoid sending multiple lock CANCELs (352)
4097 - MDS can do multi-client recovery (modulo bugs in new code)
4099 - many updates, edits, cleanups
4101 2002-11-18 Phil Schwan <phil@clusterfs.com>
4104 - fix null d_it dereference (346)
4105 - fix full OST/dbench hang (333)
4106 - fix permission problem with file removal (286)
4107 - fix removal of OSCs from LOV when they fail
4108 - fix NULL deref during bulk timeout (214)
4109 - fix problems related to multiple filesystems on one MDS (241)
4110 - fixed serious subtle metadata locking bugs
4111 - free locks on clients when inodes are removed due to memory
4113 - fix inode pointer in lock data (285)
4114 - partial support for multiple MDS on a single host (241)
4115 - data locks weren't cancelled at clear_inode time (290, 311)
4116 - intent locks could lead to unbounded lock growth (205)
4117 - added a maximum lock count, an LRU list, and a flusher
4118 - fix multiple rename (365)
4119 - properly abstracted the echo client
4120 - OSC locked 1 byte too many; fixed
4121 - rewrote brw callback code:
4122 - fixed recovery bugs related to LOVs (306)
4123 - fixed too-many-pages-in-one-write crash (191)
4124 - fixed (again) crash in sync_io_timeout (214)
4125 - probably fixed callback-related race (385)
4127 - Add capability to MDS protocol
4128 - LDLM cancellations and callbacks on different portals
4130 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4133 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4134 - defer open object destruction to close time (601981, 138)
4135 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4136 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4137 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4138 - don't call obd_setattr in ll_file_release for destroyed objects
4139 * protocol change to lustre_msg: move |version| and add |flags|
4140 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4141 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4142 * added replay of create, unlink, link and rename operations during
4143 MDS failover; recovery should be much more robust now
4144 * remove failed OSCs from LOVs (only lov_create uses this so far)
4145 * the lustre-HOWTO was brought (more) up to date (582544)
4147 2002-10-23 Phil Schwan <phil@clusterfs.com>
4150 - in-use dentries weren't being reused properly (617851)
4151 - prevent multiple LDLM setup (599178)
4152 - fix LOV size calculations for truncate (617853)
4153 - fix client handling of MDS intent errors (POSIX)
4154 - fix permission bug in lovstripe.c test (624321)
4155 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4156 - truncate past end of file could corrupt data
4157 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4158 - a race in recovery could return ETIMEDOUT to apps (623947)
4159 - building outside the source directory was fixed
4160 * the lustre-HOWTO was brought (more) up to date (582544)
4161 * major progress was made on recovery functionality
4163 2002-10-10 Phil Schwan <phil@clusterfs.com>
4166 - recovery deadlock fix
4167 - rm -rf causes LBUG fix (617817)
4168 - file open by multiple tasks fix (618962)
4169 - directory permissions bugs (602707 and 620007)
4170 - journal_stop fixed with locking (611313)
4171 - O_APPEND failures resolved (618273, perhaps 614459)
4172 - lconf PATH fix (619770)
4173 - IA64 build fix (621450)
4174 - RPC buffer sizes scale with amount of memory
4176 2002-10-01 Phil Schwan <phil@clusterfs.com>
4179 - locks would be cancelled without throwing away data pages,
4180 resulting in inconsistent data (605627)
4181 - inode attributes were not always being refreshed (605627, 612449)
4182 - lconf now continues to cleanup after lctl reports an error
4183 - MDS now enforces user permissions (602707)
4184 - lprocfs cleanup fixed, but not yet enabled (614157)
4185 - fixed infinite server hang, should a client not respond to an AST
4186 - avoid going into recovery if user calls readlink() with a buffer
4187 that's too small (613941)
4188 - AST RPCs no longer require replies (614867) -- this may be changed
4189 - don't crash server if client sends an IOV that's too big (611336)
4190 - fixed lock conversion deadlock (611892)
4191 - fixed the following of symlinks (614622)
4192 * recovery: the server can remove locks from a client that dies, other
4193 clients can make progress
4194 * more extN patch fixes
4195 * compile-time configurable ptlrpc buffer allocations
4197 - collaborative read cache document
4198 - Lustre Lite Performance CDR document-in-progress
4200 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4203 - fix typo in patch-2.4.18
4205 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4208 - clear ptlrpc request each time in handle_incoming_request()
4209 - unlink of files now destroys the object on the OST
4211 2002-09-19 Peter Braam <braam@clusterfs.com>
4213 * add hard link support
4214 * change obdfile creation method
4215 * kernel patch changed
4217 2002-09-19 Peter Braam <braam@clusterfs.com>
4220 - stack overflow bug in extN fixed
4222 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4224 * documentation updates
4225 - add man pages for config tools
4226 - update tests/README to describe testing with new config tools
4227 - finish metadata API descriptions
4228 * bug fixes and cleanups
4229 - statfs workaround for 16TB limit
4230 - LOV stripe allocation improved, can stripe on subset of OSTs
4231 - LOV file size/IO offset was wrong for files > 4GB in size
4232 - object EA data was being dropped, caused files to be unreadable
4233 - memory overflow with non-LOV OST caused memory corruption
4234 - fixed regression tests to work with new config tools, obdfilter
4235 - fixed bug when directory size became larger than 1 block
4236 - fixed bug (for single client case) when PWD was deleted
4237 - invalidate local directory pages when doing intent-based ops
4238 - avoid LDLM oops when lock callback contained bad data
4240 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4242 * documentation updates
4243 * bug fixes and cleanups
4244 - configuration tools
4247 - 64-bit compile warnings
4248 - 64-bit internal statfs data
4250 * test_brw on persistent OST devices
4252 * lprocfs (disabled)
4254 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4256 * documentation updates
4257 * bug fixes and cleanups
4258 * configuration tools
4260 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4263 * many small fixes to 0_5_4
4264 * io/network handling
4265 * thinkos in MDS operations
4267 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4270 * crucial basic fixes to 0.5.3
4271 * IOR, Iozone work over Elan
4274 2002-08-07 Phil Schwan <phil@clusterfs.com>
4275 * version 0_5_3, our first alpha
4276 * we use the new Portals iovs
4277 * documentation updates
4278 * bug fixes and cleanups
4279 * small changes in the DLM wire protocol
4281 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4282 * version 0_5_1 with some initial stability,
4283 * locking on MD and file I/O.
4284 * documentation updates
4285 * several bug fixes since 0.5.0
4286 * small changes in wire protocol
4288 2002-07-18 Phil Schwan <phil@clusterfs.com>
4290 * delivered as Lustre Light Alpha
4291 * fixed a crash after handling invalid MDS requests
4292 * fixed directory pages for architectures with non-4k pages sizes
4294 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4295 * release version v0_4_4
4296 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4298 * Fixes a number of interruption problems with OST operations.
4299 * Update documentation for portals header changes
4300 * Move all wire protocol structs/defines to lustre_idl.h
4301 * Fixes symlink length bug.
4302 * Add tcpdump to repository.
4304 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4305 * release version v0_4_3
4306 * Fixes statfs for inodes on extN.
4307 * Fixes bug in runtests which would delete /etc/hosts.
4308 * Use 64-bit object IDs wherever possible (not into VFS though)
4309 Remove ost_get_info, which is unused by lustre, and out of date.
4311 2002-07-03 Peter Braam <braam@clusterfs.com>
4312 * release version v0_4_2 Fixes a lookup error (type not passed)
4313 * move forward to head of Portals
4314 * move forward to latest Lustre kernel
4316 2002-06-25 Peter Braam <braam@clusterfs.com>
4317 * release version v0_4_1. Hopefully stable on single node use.