3 2.6.5-7.287.3 (SLES 9),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.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 RHEL4 U6 kernel-2.6.9-67.EL.
19 Severity : enhancement
21 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
25 Description: make mgs_setparam() handle fsname containing dash
26 Details : fsname containing a dash does not work with lctl conf_param
28 Severity : enhancement
30 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
32 Severity : enhancement
34 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
38 Frequency : rare, at shutdown
39 Description: access already free / zero obd_namespace.
40 Details : if client_disconnect_export was called without force flag set,
41 and exist connect request in flight, this can produce access to
42 NULL pointer (or already free pointer) when connect_interpret
43 store ocd flags in obd_namespace.
47 Frequency : only at startup
48 Description: not alloc memory with spinlock held.
49 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
52 Severity : enhancement
54 Description: make lustre randomly fail allocating memory
55 Details : Make lustre randomly failed allocating memory for testing purpose.
57 Severity : enhancement
59 Description: lost problems with lov objid file
60 Details : Fixes some scability and access to not inited memory problems
61 in work with lov objdid file.
66 Description: lfs find does not continue on file error
67 Details : Continue other files processing when a file/dir is absent.
71 Description: Inconsistent usage of lustre_pack_reply()
72 Details : Standardize the usage of lustre_pack_reply() such that it
73 always generate a CERROR on failure.
76 Frequency : occasional
78 Description: MDS hang or stay in waiting lock
79 Details : If client receive lock with CBPENDING flag ldlm need send lock
80 cancel as separate rpc, to avoid situation when cancel request
81 can't processed due all i/o threads stay in wait lock.
85 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
89 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
93 Description: Update to SLES9 kernel-2.6.5-7.287.3.
97 Description: Update to RHEL4 latest kernel.
99 Severity : enhancement
101 Description: Build SLES10 patchless client fails
102 Details : The configure was broken by run ./configure with
103 --with-linux-obj=.... argument for patchless client. When the
104 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
105 can't search header adequately. Use absolute path such as
106 -I($LINUX)/include instead.
110 Description: interrupt oig_wait produce painc on resend.
111 Details : brw_redo_request can be used for resend requests from ptlrpcd and
112 private set, and this produce situation when rq_ptlrpcd_data not
113 copyed to new allocated request and triggered LBUG on assert
114 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
115 ptlrpcd set if request is changed and can be safety changed to use
120 Description: LASSERT_{REQ,REP}SWAB macros are buggy
121 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
122 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
123 these macros and replace them with their difinitions instead.
127 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
128 Details : Update which_patch & target file for SLES10 latest kernel.
131 Description: Remove client patches from SLES 10 kernel.
132 Details : This causes SLES 10 clients to behave as patchless clients
133 even on a Lustre-patched (server) kernel.
135 Severity : enhancement
137 Description: self-adjustable client's lru lists
138 Details : use adaptive algorithm for managing client cached locks lru
139 lists according to current server load, other client's work
140 pattern, memory activities, etc. Both, server and client
141 side namespaces provide number of proc tunables for controlling
144 Severity : enhancement
146 Description: light-weight GSS support
147 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
148 and reduce performance overhead.
150 Severity : enhancement
152 Description: Linux keyring support
153 Details : Support using service of Linux keyring for Lustre GSS internal
154 context refresh/cache mechanism.
158 Description: Fix errors in lfs documentation
159 Details : Fixes man pages
163 Description: don't use GFP_* in generic Lustre code.
164 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
168 Description: obdclass is limited by single OBD_ALLOC(idarray)
169 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
173 Description: warnings with build patchless client with vanila 2.6.19 and up
174 Details : change old ctl_table style and replace ctl_table/ctl_table_header
175 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
179 Frequency : Only for SLES
180 Description: sanity_quota fail test_1
181 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
182 file, which makes TSTID[2] inunique.
186 Frequency : Only for SLES9
187 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
188 Details : Update target/ChangeLog/which_patch .
192 Frequency : Only for test_10 in sanity-quota.sh
193 Description: a bug in quota test code
194 Details : checking a test flag in a wrong place
198 Frequency : Only for test_13 in sanity-quota.sh
199 Description: a bug in quota test script
200 Details : shouldn't check overfull allocated quota
204 Description: jbd statistics
205 Details : Port older jbd statistics patch for sles10
207 Severity : enhancement
209 Description: Add printing inode info into message about error in writepage.
214 Description: reply_lock_interpret crash due to race with it and lock cancel.
215 Details : Do not replay locks that are being cancelled. Do not reference
216 locks by their address during replay, just by their handle.
220 Frequency : When flocks are used.
221 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
222 Details : Flock locks might destroy just granted lock if it could be merged
223 with another existing flock, this is done in completion handler,
224 so teach ldlm_cli_enquque_fini that this is a valid case for
230 Description: Oops in read and write path when failing to allocate lock.
231 Details : Check if lock allocation failed and return error back.
235 Description: lstripe command fails for valid OST index
236 Details : The stripe offset is compared to lov->desc.ld_tgt_count
237 instead of lov->desc.ld_active_tgt_count.
241 Description: sanity.sh failed test 103
242 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
247 Description: df doesn't work properly if diskfs blocksize != 4K
248 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
252 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
253 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
257 Description: obd_health_check_timeout too short
258 Details : set obd_health_check_timeout as 1.5x of obd_timeout
262 Description: llapi_file_create() does not allow some changes
263 Details : add llapi_file_open() that allows specifying the mode and
264 open flags, and also returns an open file handle.
268 Description: merge and cleanup kernel patches.
269 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
273 Description: Add journal checksum support.(Kernel part)
274 Details : The journal checksum feature adds two new flags i.e
275 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
276 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
277 indicates that the commit block contains the checksum for
278 the blocks described by the descriptor blocks. Now commit
279 record can be sent to disk without waiting for descriptor
280 blocks to be written to disk. This behavior is controlled
281 using JBD2_FEATURE_ASYNC_COMMIT flag.
285 Description: OSS needs mutliple precreate threads
286 Details : Add ability to start more than one create thread per OSS.
290 Description: Sanity.sh test_65k failed: setstripe should have succeeded
291 Details : execute lfs setstripe on client
295 Description: mds_obd_create error creating tmp object
296 Details : When the user sets quota on root, llog will be affected and can't
297 create files and write files.
300 Frequency : Always on ia64 patchless client, and possibly others.
302 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
303 Details : This allows the patchless client to be loaded on architectures
308 Description: RedHat Update kernel for RHEL5
309 Details : Add the kernel config file for RHEL5.
313 Description: RedHat Update kernel for RHEL5
314 Details : Modify the kernel config file more closer RHEL5.
318 Description: Build failure against Centos5 (RHEL5)
319 Details : Define PAGE_SIZE when it isn't present.
323 Description: test_7 on sanity-quota.sh can't be used on separate machines
324 Details : add facet to handle it
328 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
329 Details : with NFS, the anon dentry's parent was set to itself in
330 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
331 corresponding dentry other than use rec->ur_name.
333 Severity : enhancement
335 Description: lfs setstripe enhancement
336 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
340 Description: enable data checksumming by default
341 Details : enable checksum by default, allow --disable-checksum
342 configure option and "-o nochecksum" mount option. Checksums
343 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
347 Description: lustre support for RHEL5
348 Details : Add support for RHEL5.
352 Description: Client eviction due to failover config
353 Details : after a connection loss, the lustre client should attempt to
354 reconnect to the last active server first before trying the
355 other potential connections.
359 Description: when mds and osts use different quota unit(32bit and 64bit),
360 quota will be released repeatly.
361 Details : void sending multiple quota reqs to mds, which will keep the
362 status between the reqs.
366 Description: osts not allocated evenly to files
367 Details : change the condition to increase offset_idx
370 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
372 Description: Data corruption for OSTs that are formatted larger than 2TB
374 Details : When generating the bio request for lustre file writes the
375 sector number would overflow a temporary variable before being
376 used for the IO. The data reads correctly from Lustre (which
377 will overflow in a similar manner) but other file data or
378 filesystem metadata may be corrupted in some cases.
382 Description: Tune the kernel for good SCSI performance.
383 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
384 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
389 Description: rewrite ext2-derived code in obdclass/uuid.c
390 Details : rewrite inherited code (uuid parsing code from ext2 utils)
391 from scratch preserving functionality.
395 Description: "lfs find -obd UUID" prints directories
396 Details : "lfs find -obd UUID" will return all directory names instead
397 of just file names. It is incorrect because the directories
398 do not reside on the OSTs.
402 Description: use i_size_read and i_size_write in 2.6 port
403 Details : replace inode->i_size access with i_size_read/write()
406 Frequency : when removing large files
408 Description: scheduling issue during removal of large Lustre files
409 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
410 It causes scheduling issues when removing large files (17TB in the
414 Frequency : only with liblustre clients on XT3
416 Description: evictions taking too long
417 Details : allow llrd to evict clients directly on OSTs
420 Frequency : only on ppc
422 Description: /proc/fs/lustre/devices broken on ppc
423 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
424 We should make sure the type of 'index' is loff_t and then cast
425 explicitly as needed below. Do not assign an explicitly cast
429 Frequency : only for rhel5
431 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
432 Details : Modify the target file & which_kernel.
434 Severity : enhancement
436 Description: omit set fsid for export NFS
437 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
438 as FS_REQUIRES_DEV for avoid problems with generate fsid.
442 Frequency : Always, for kernels after 2.6.16
443 Description: Fix warning idr_remove called for id=.. which is not allocated.
444 Details : Last kernels save old s_dev before kill super and not allow
445 to restore from callback - restore it before call kill_anon_super.
449 Description: buffer overruns could theoretically occur
450 Details : llapi_semantic_traverse() modifies the "path" argument by
451 appending values to the end of the origin string, and a
452 overrun may occur. Adding buffer overrun check in liblustreapi.
456 Description: Fix error on 'ls .' at the top of the Lustre mount.
457 Details : Don't revalidate dentry if it is a root dentry.
461 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
462 Details : Modify vm-tunables-rhel4.patch.
466 Description: Kernel config for 2.6.18-vanilla.
467 Details : Modify targets/2.6-vanilla.target.in.
468 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
469 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
470 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
471 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
475 Description: improve handling recoverable errors
476 Details : if request processig with error which can be recoverable on server
477 request should be resend, otherwise page released from cache and
482 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
483 Details : Modify target file & which_patch.
484 A flaw was found in the IA32 system call emulation provided
485 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
486 value could be stored in the %RAX register, which could trigger an
487 out-of-bounds system call table access. An untrusted local user
488 could exploit this flaw to run code in the kernel
489 (ie a root privilege escalation). (CVE-2007-4573).
493 Description: O_DIRECT bypasses client statistics.
494 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
495 read_ahead_stats were updated. copied the stats section from
496 osc_send_oap_rpc() into async_internal().
500 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
504 Description: change order of libsysio includes
505 Details : '#include sysio.h' should always come before '#include xtio.h'
509 Description: handle "serious error: objid * already exists" more gracefully
510 Details : If LAST_ID value on disk is smaller than the objects existing in
511 the O/0/d* directories, it indicates disk corruption and causes an
512 LBUG(). If the object is 0-length, then we should use the existing
513 object. This will help to avoid a full fsck in most cases.
515 Severity : enhancement
517 Description: adapt the lustre_config script to support the upgrade case
518 Details : Add "-u" option for lustre_config script to support upgrading 1.4
519 server targets to 1.6 in parallel.
523 Description: To avoid grant space > avaible space when the disk is almost
524 full. Without this patch you might see the error "grant XXXX >
525 available" or some LBUG about grant, when the disk is almost
527 Details : In filter_check_grant, for non_grant cache write, we should
528 check the left space by if (*left > ungranted + bytes), instead
529 of (*left > ungranted), because only we are sure the left space
530 is enough for another "bytes", then the ungrant space should be
531 increase. In client, we should update cl_avail_grant only there
532 is OBD_MD_FLGRANT in the reply.
536 Description: Update RHEL 4 kernel to fix local root privilege escalation.
537 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
538 described in CVE-2007-4573. This problem could allow untrusted
539 local users to gain root access.
542 Frequency : when using O_DIRECT and quotas
544 Description: Incorrect file ownership on O_DIRECT output files
545 Details : block usage reported by 'lfs quota' does not take into account
546 files that have been written with O_DIRECT.
551 Description: touch file failed when fs is not full
552 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
553 otherwise we can get ENOSP while fs is not full.
557 Description: parallel lock callbacks
558 Details : Instead of sending blocking and completion callbacks as separated
559 requests, adding them to a set and sending in parallel.
562 Frequency : only for Cray XT3
563 Bugzilla : 12829/13455
564 Description: Changing primary group doesn't change the group lustre assigns to
566 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
567 overridden with the primary group provided by the group upcall,
568 whereas the supplied fsgid can be trusted if it is in the list of
569 supplementary groups returned by the group upcall.
571 Severity : enhancement
573 Description: Allow masking D_WARNING, D_ERROR messages from console
574 Details : Console messages can now be disabled via lnet.printk.
576 --------------------------------------------------------------------------------
578 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
580 * Support for kernels:
581 2.6.5-7.283 (SLES 9),
582 2.6.9-55.EL (RHEL 4),
583 2.6.16.46-0.14 (SLES 10),
584 2.6.18.8 vanilla (kernel.org)
585 * Client support for unpatched kernels:
586 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
587 2.6.16 - 2.6.22 vanilla (kernel.org)
588 * Due to recently discovered recovery problems, we do not recommend
589 using patchless RHEL 4 clients with this or any earlier release.
590 * Recommended e2fsprogs version: 1.39.cfs8
591 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
592 * Starting with this release, the ldiskfs backing filesystem required
593 by Lustre is now in its own package, lustre-ldiskfs. This package
594 should be installed. It is versioned separately from Lustre and
595 may be released separately in future.
599 Description: block reactivating mgc import until all deactivates complete
600 Details : Fix race when failing back MDT/MGS to itself (testing)
603 Frequency : at statup only
605 Description: mds_lov_synchronize race leads to various problems
606 Details : simultaneous MDT->OST connections at startup can cause the
607 sync to abort, leaving the OSC in a bad state.
609 Severity : enhancement
611 Description: add optional extra BUILD_VERSION info
612 Details : add a new environment variable (namely LUSTRE_VERS) which allows
613 to override the lustre version.
616 Frequency : 2.6.18 servers only
618 Description: ll_kern_mount() doesn't release the module reference
619 Details : The ldiskfs module reference count never drops down to 0
620 because ll_kern_mount() doesn't release the module reference.
625 Description: server LBUG when using old ost_num_threads parameter
626 Details : Accept the old ost_num_threads parameter but warn that it
627 is deprecated, and fix an off-by-one error that caused an LBUG.
632 Description: Transient SCSI error results in persistent IO issue
633 Details : iobuf->dr_error is not reinitialized to 0 between two
637 Frequency : sometimes when underlying device returns I/O errors
639 Description: OSTs not going read-only during write failures
640 Details : OSTs are not remounted read-only when the journal commit threads
641 get I/O errors because fsfilt_ext3 calls journal_start/stop()
642 instead of the ext3 wrappers.
646 Description: poor connect scaling with increasing client count
647 Details : Don't run filter_grant_sanity_check for more than 100 exports
648 to improve scaling for large numbers of clients.
651 Frequency : SLES10 only
653 Description: sanity-quota.sh quotacheck failed: rc = -22
654 Details : Quotas cannot be enabled on SLES10.
657 Frequency : liblustre clients only
659 Description: getdirentries does not give error when run on compute nodes
660 Details : getdirentries does not fail when the size specified as an argument
661 is too small to contain at least one entry
663 Severity : enhancement
665 Description: Add LNET router traceability for debug purposes
666 Details : If a checksum failure occurs with a router as part of the
667 IO path, the NID of the last router that forwarded the bulk data
668 is printed so it can be identified.
673 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
674 Details : Due to a race condition, liblustre clients were occasionally
677 Severity : enhancement
679 Description: lfs setstripe use optional parameters instead of postional
682 Severity : enhancement
684 Description: Nanosecond timestamp support for ldiskfs
685 Details : The on-disk ldiskfs filesystem has added support for nanosecond
686 resolution timestamps. There is not yet support for this at
687 the Lustre filesystem level.
690 Frequency : during server recovery
692 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
693 Details : request with rq_no_resend flag not awake l_wait_event if they get a
697 Frequency : nfs export on patchless client
699 Description: connectathon hang when test nfs export over patchless client
700 Details : Disconnected dentry cannot be found with lookup, so we do not need
701 to unhash it or make it invalid
704 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
705 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
706 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
711 Description: open req refcounting wrong on reconnect
712 Details : If reconnect happened between getting open reply from server and
713 call to mdc_set_replay_data in ll_file_open, we will schedule
714 replay for unreferenced request that we are about to free.
715 Subsequent close will crash in variety of ways.
716 Check that request is still eligible for replay in
717 mdc_set_replay_data().
722 Description: disable writes to filesystem when reading health_check file
723 Details : the default for reading the health_check proc file has changed
724 to NOT do a journal transaction and write to disk, because this
725 can cause reads of the /proc file to hang and block HA state
726 checking on a healthy but otherwise heavily loaded system. It
727 is possible to return to the previous behaviour during configure
728 with --enable-health-write.
730 Severity : enhancement
732 Description: 64-bit inode version
733 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
734 to the inode. This will be required for version-based recovery.
739 Description: MDS fails to start if a duplicate client export is detected
740 Details : in some rare cases it was possible for a client to connect to
741 an MDS multiple times. Upon recovery the MDS would detect this
742 and fail during startup. Handle this more gracefully.
744 Severity : enhancement
746 Description: Add -o localflock option to simulate old noflock
748 Details : This will achieve local-only flock/fcntl locks
754 Description: log_commit_thread vs filter_destroy race leads to crash
755 Details : Take import reference before releasing llog record semaphore
760 Description: Wrong request locking in request set processing
761 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
762 imp_delayed_list, in this place should be used imp_lock.
765 Frequency : when reconnection
767 Description: Grant Leak when osc reconnect to OST
768 Details : When osc reconnect ost, OST(filter) should check whether it
769 should grant more space to client by comparing fed_grant and
770 cl_avail_grant, and return the granted space to client instead
771 of "new granted" space, because client will call osc_init_grant
772 to update the client grant space info.
775 Frequency : when client reconnect to OST
777 Description: Grant Leak when osc do resend and replay bulk write
778 Details : When osc reconnect to OST, OST(filter)should clear grant info of
779 bulk write request, because the grant info while be sync between
780 OSC and OST when reconnect, and we should ignore the grant info
781 these of resend/replay write req.
786 Description: Grant space more than avaiable left space sometimes.
787 Details : When then OST is about to be full, if two bulk writing from
788 different clients came to OST. Accord the avaliable space of the
789 OST, the first req should be permitted, and the second one
790 should be denied by ENOSPC. But if the seconde arrived before
791 the first one is commited. The OST might wrongly permit second
792 writing, which will cause grant space > avaiable space.
795 Frequency : when client is evicted
797 Description: Grant might be wrongly erased when osc is evicted by OST
798 Details : when the import is evicted by server, it will fork another
799 thread ptlrpc_invalidate_import_thread to invalidate the
800 import, where the grant will be set to 0. While the original
801 thread will update the grant it got when connecting. So if
802 the former happened latter, the grant will be wrongly errased
803 because of this race.
808 Description: Checking Stale with correct fid
809 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
810 to check whether it is stale, because sometimes, we want the
811 enqueue happened anyway, and op_data.fid2 will not be initialized.
813 Severity : enhancement
815 Description: update patchless client
816 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
819 Frequency : only with 2.4 kernel
821 Description: random memory corruption
822 Details : size of struct ll_inode_info is to big for union inode.u and this
823 can be cause of random memory corruption.
828 Description: Memory leak in recovery
829 Details : Lov_mds_md was not free in an error handler in mds_create_object.
830 It should also check obd_fail before fsfilt_start, otherwise if
831 fsfilt_start return -EROFS,(failover mds during mds recovery).
832 then the req will return with repmsg->transno = 0 and rc = EROFS.
833 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
834 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
835 be freed no matter whether fsfilt_commit success or not.
838 Frequency : only with huge count clients
840 Description: Prevents from taking the superblock lock in llap_from_page for
842 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
843 for llap_from_page call in ll_removepage prevents from taking the
844 superblock lock for a soon died page.
849 Description: Not check open intent error before release open handle
850 Details : in some rare cases, the open intent error is not checked before
851 release open handle, which may cause
852 ASSERTION(open_req->rq_transno != 0), because it tries to release
853 the failed open handle.
858 Description: Set cat log bitmap only after create log success.
859 Details : in some rare cases, the cat log bitmap is set too early. and it
860 should be set only after create log success.
864 Description: Accessing a block bevice can re-enable I/O when Lustre is
865 tearing down a device.
866 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
871 Description: service threads may hog cpus when there are a lot of requests
872 Details : Insert cond_resched to give other threads a chance to use some CPU
877 Description: the cat log was not initialized in recovery
878 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
879 unlink log on mds will not be initialized until mds post
880 recovery. And also in mds post recovery, the unlink log will
881 initialization will be done asynchronausly, so there will be race
882 between add unlink log and unlink log initialization.
886 Description: brw_stats were being printed incorrectly
887 Details : brw_stats were being printed as log2 but all of them were not
888 recorded as log2. Also remove some code duplication arising from
889 filter_tally_{read,write}.
893 Frequency : rare, only in recovery.
894 Description: ASSERTION(req->rq_type != LI_POISON) failed
895 Details : imp_lock should be held while iterating over imp_sending_list for
896 prevent destroy request after get timeout in ptlrpc_queue_wait.
900 Description: replay-single.sh test 52 fails
901 Details : A lock's skiplist need to be cleanup when it being unlinked
902 from its resource list.
906 Description: Short directio read returns full requested size rather than
908 Details : Direct I/O operations should return actual amount of bytes
909 transferred rather than requested size.
911 Severity : enhancement
913 Description: metadata RPC reduction (e.g. for rm performance)
914 Details : decrease the amount of synchronous RPC between clients and servers
915 by canceling conflicing lock before the operation on the client
916 and packing thier handles into the main operation RPC to server.
918 Severity : enhancement
920 Description: Async OSC create to avoid the blocking unnecessarily.
921 Details : If a OST has no remain object, system will block on the creating
922 when need to create a new object on this OST. Now, ways use
923 pre-created objects when available, instead of blocking on an
924 empty osc while others are not empty. If we must block, we block
925 for the shortest possible period of time.
927 Severity : enhancement
929 Description: refine locking for avoid write wrong info into lov_objid file
930 Details : fix possible races with add new target and write/update data in
933 --------------------------------------------------------------------------------
935 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
940 Frequency : on some architectures
942 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
943 Details : Uninitialized flags sometimes cause configuration commands to
947 Frequency : patchless clients only
949 Description: missing __iget() symbol export
950 Details : The __iget() symbol export is missing. To avoid the need for
951 this on patchless clients the deathrow inode reaper is turned
952 off, and we depend on the VM to clean up old inodes. This
953 dependency was during via the fix for bug 12181.
955 --------------------------------------------------------------------------------
957 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
959 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
960 INTEROPERATE with older versions automatically. In many cases a
961 special upgrade step is needed. Please read the
962 user documentation before upgrading any part of a live system.
963 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
964 version will not interoperate with 1.6 betas before beta5 (1.5.95).
965 * WARNING: Lustre configuration and startup changes are required with
966 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
972 Severity : enhancement
974 Description: MountConf
975 Details : Lustre configuration is now managed via mkfs and mount
976 commands instead of lmc and lconf. New obd types (MGS, MGC)
977 are added for dynamic configuration management. See
978 https://mail.clusterfs.com/wikis/lustre/MountConf for
981 Severity : enhancement
983 Description: dynamic OST addition
984 Details : OSTs can now be added to a live filesystem
986 Severity : enhancement
988 Description: startup order invariance
989 Details : MDTs and OSTs can be started in any order. Clients only
990 require the MDT to complete startup.
992 Severity : enhancement
994 Description: parallel, asynchronous orphan cleanup
995 Details : orphan cleanup is now performed in separate threads for each
996 OST, allowing parallel non-blocking operation.
998 Severity : enhancement
1000 Description: optimized stripe assignment
1001 Details : stripe assignments are now made based on ost space available,
1002 ost previous usage, and OSS previous usage, in order to try
1003 to optimize storage space and networking resources.
1005 Severity : enhancement
1007 Description: Permanently set tunables
1008 Details : All writable /proc/fs/lustre tunables can now be permanently
1009 set on a per-server basis, at mkfs time or on a live system.
1011 Severity : enhancement
1013 Description: Lustre message v2
1014 Details : Add lustre message format v2.
1016 Severity : enhancement
1018 Description: client OST exclusion list
1019 Details : Clients can be started with a list of OSTs that should be
1020 declared "inactive" for known non-responsive OSTs.
1024 Description: SPEC SFS validation failure on NFS v2 over lustre.
1025 Details : Changes the blocksize for regular files to be 2x RPC size,
1026 and not depend on stripe size.
1028 Severity : enhancement
1030 Description: Multiple MD RPCs in flight.
1031 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1032 To avoid overly-overloading MDS, introduce a limit on number of
1033 MDS RPCs in flight for a single client and add /proc controls
1034 to adjust this limit.
1036 Severity : enhancement
1038 Description: client read/write statistics
1039 Details : Add client read/write call usage stats for performance
1040 analysis of user processes.
1041 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1042 file access. extents_stats shows chunk size distribution.
1043 extents_stats_per_process show chunk size distribution per
1046 Severity : enhancement
1048 Description: mds statistics
1049 Details : Add detailed mds operations statistics in
1050 /proc/fs/lustre/mds/*/stats.
1054 Description: Failure of copying files with lustre special EAs.
1055 Details : Client side always return success for setxattr call for lustre
1056 special xattr (currently only "trusted.lov").
1061 Description: Refcount LNET uuids
1062 Details : The global LNET uuid list grew linearly with every startup;
1063 refcount repeated list entries instead of always adding to
1066 Severity : enhancement
1068 Description: Dynamic service threads
1069 Details : Within a small range, start extra service threads
1070 automatically when the request queue builds up.
1073 Frequency : mixed-endian client/server environments
1075 Description: mixed-endian crashes
1076 Details : The new msg_v2 system had some failures in mixed-endian
1079 Severity : enhancement
1081 Description: Easy OST removal
1082 Details : OSTs can be permanently deactivated with e.g. 'lctl
1083 conf_param lustre-OST0001.osc.active=0'
1085 Severity : enhancement
1087 Description: MGS proc entries
1088 Details : Added basic proc entries for the MGS showing what filesystems
1091 Severity : enhancement
1093 Description: provide MGS failover
1094 Details : Added config lock reacquisition after MGS server failover.
1096 Severity : enhancement
1098 Description: add Linux 2.4 support
1099 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1103 Description: a large application tries to do I/O to the same resource and dies
1104 in the middle of it.
1105 Details : Check the req->rq_arrival time after the call to
1106 ost_brw_lock_get(), but before we do anything about
1107 processing it & sending the BULK transfer request. This
1108 should help move old stale pending locks off the queue as
1109 quickly as obd_timeout.
1112 Frequency : when an incorrect nid is specified during startup
1114 Description: ptlrpc connect to non-existant node causes kernel crash
1115 Details : LNET can't be re-entered from an event callback, which
1116 happened when we expire a message after the export has been
1117 cleaned up. Instead, hand the zombie cleanup off to another
1120 Severity : enhancement
1122 Description: plain/inodebits lock performance improvement
1123 Details : Grouping plain/inodebits in granted list by their request modes
1124 and bits policy, thus improving the performance of search through
1128 Frequency : only if OST filesystem is corrupted
1130 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1131 Details : for a short time RPCs with bulk IO are in the replay list,
1132 but replay of bulk IOs is unimplemented. If the OST filesystem
1133 is corrupted due to disk cache incoherency and then replay is
1134 started it is possible to trip an assertion. Avoid putting
1135 committed RPCs into the replay list at all to avoid this issue.
1138 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1140 Description: System hang on startup
1141 Details : This bug allowed the liblustre (e.g. catamount) client to
1142 return to the app before handling all startup RPCs. This
1143 could leave the node unresponsive to lustre network traffic
1144 and manifested as a server ptllnd timeout.
1146 Severity : enhancement
1148 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1149 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1150 boolean to control whether to print peer debug info when a
1151 client's RPC times out.
1154 Frequency : only for kernels with patches from Lustre below 1.4.3
1156 Description: Remove old rdonly API
1157 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1160 Frequency : only for devices with external journals
1162 Description: Set external device read-only also
1163 Details : During a commanded failover stop, we set the disk device
1164 read-only while the server shuts down. We now also set any
1165 external journal device read-only at the same time.
1168 Frequency : when upgrading from 1.4 while trying to change parameters
1170 Description: The wrong (new) MDC name was used when setting parameters for
1171 upgraded MDT's. Also allows changing of OSC (and MDC)
1172 parameters if --writeconf is specified at tunefs upgrade time.
1175 Frequency : when setting specific ost indicies
1177 Description: QOS code breaks on skipped indicies
1178 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1179 created with --index need not be sequential.
1181 Severity : enhancement
1183 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1184 Details : The uninit_groups feature works in conjunction with the kernel
1185 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1186 up the pass1 processing of e2fsck. This is a read-only feature
1187 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1188 mount filesystems that have had this feature enabled.
1190 Severity : enhancement
1192 Description: Improve multi-block allocation algorithm to avoid fragmentation
1193 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1194 allocation locality and avoid filesystem fragmentation.
1196 ------------------------------------------------------------------------------
1198 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1200 * Support for kernels:
1201 2.4.21-47.0.1.EL (RHEL 3)
1202 2.6.5-7.283 (SLES 9)
1203 2.6.9-42.0.10.EL (RHEL 4)
1204 2.6.12.6 vanilla (kernel.org)
1205 2.6.16.27-0.9 (SLES 10)
1206 * Recommended e2fsprogs version: 1.39.cfs5
1208 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1212 Frequency : occasional, depends on client load and configuration
1213 Bugzilla : 12181, 12203
1214 Description: data loss for recently-modified files
1216 Details : In some cases it is possible that recently written or created
1217 files may not be written to disk in a timely manner (this should
1218 normally be within 30s unless client IO load is very high).
1219 The problem appears as zero-length files or files that are a
1220 multiple of 1MB in size after a client crash or client eviction
1221 that are missing data at the end of the file.
1223 This problem is more likely to be hit on clients where files are
1224 repeatedly created and unlinked in the same directory, clients
1225 have a large amount of RAM, have many CPUs, the filesystem has
1226 many OSTs, the clients are rebooted frequently, and/or the files
1227 are not accessed by other nodes after being written.
1229 The presence of the problem can be detected by looking at
1230 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1231 smaller than the second (nr_unused) then dirty files will not
1232 be flushed automatically to disk. "sync; sleep 10" should be
1233 run several times on the node before unmounting it to update
1234 Lustre (this is also safe to run on nodes without this problem).
1236 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1237 that can cause this same problem, so customers using that kernel
1238 also need to update the kernel in addition to Lustre. In order
1239 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1241 It is normal that files written just before a client crash (less
1242 than 30s) may not yet have been flushed to disk, even for local
1246 Frequency : frequent on thin XT3 nodes
1248 Description: UUID collision on thin XT3 Linux nodes
1249 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1250 randomly, since we relied on an insufficiently-seeded PRNG.
1255 Description: OSS hangs after "All ost request buffers busy"
1256 Details : A deadlock between quota and journal operations caused OSS
1257 hangs after printing "All ost request buffers busy."
1260 Frequency : always on liblustre builds
1262 Description: Cleanup compiler warnings on liblustre
1265 Frequency : always on liblustre builds on XT3
1267 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1271 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1272 > 32000 subdirectories
1273 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1274 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1275 subdirectory count crosses 32000. This will aid e2fsck to
1276 correctly handle more than 32000 subdirectories.
1279 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1281 Description: System hang on startup
1282 Details : This bug allowed the liblustre (e.g. catamount) client to
1283 return to the app before handling all startup RPCs. This
1284 could leave the node unresponsive to lustre network traffic
1285 and manifested as a server ptllnd timeout.
1287 Severity : enhancement
1289 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1290 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1291 boolean to control whether to print peer debug info when a
1292 client's RPC times out.
1297 Description: make O_SYNC working on 2.6 kernels
1298 Details : 2.6 kernels use different method for mark pages for write,
1299 so need add a code to lustre for O_SYNC work.
1304 Description: Failure to close file and release space on NFS
1305 Details : Put inode details into lock acquired in ll_intent_file_open.
1306 Use mdc_intent_lock in ll_intent_open to properly
1307 detect all kind of errors unhandled by mdc_enqueue.
1312 Description: proc file read during shutdown sometimes raced obd removal,
1314 Details : Add lock to prevent obd access after proc file removal.
1317 Frequency : Only for files larger than 4GB on 32-bit clients.
1319 Description: improperly doing page alignment of locks
1320 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1321 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1326 Description: RPCs being resent when they shouldn't be
1327 Details : Some RPCs that should not be resent are being resent. This
1328 can cause inconsistencies in the RPC state machine. Do not
1329 resend such requests.
1332 Frequency : rare, only with NFS export
1334 Description: Crash on NFS re-export node
1335 Details : under very unusual load conditions an assertion is hit in
1336 ll_intent_file_open()
1339 Frequency : only if OST filesystem is corrupted
1341 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1342 Details : for a short time RPCs with bulk IO are in the replay list,
1343 but replay of bulk IOs is unimplemented. If the OST filesystem
1344 is corrupted due to disk cache incoherency and then replay is
1345 started it is possible to trip an assertion. Avoid putting
1346 committed RPCs into the replay list at all to avoid this issue.
1351 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1352 Details : Large single O_DIRECT read and write calls can fail to allocate
1353 a sufficiently large buffer to process the request. In case of
1354 allocation failure the allocation is retried with a smaller
1355 buffer and broken into smaller requests.
1357 Severity : enhancement
1359 Description: Add -o localflock option to simulate old noflock behaviour.
1360 Details : This will achieve local-only flock/fcntl locks coherentness.
1365 Description: versioning check is incomplete
1366 Details : Checking the version difference of client vs. server, report
1367 error if the gap is too big.
1372 Description: add support PG_writeback bit
1373 Details : add support for PG_writeback bit for Lustre, for more carefull
1374 work with page cache in 2.6 kernel. This also fix some deadlocks
1375 and remove hack for work O_SYNC with 2.6 kernel.
1377 Severity : enhancement
1379 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1380 Details : The uninit_groups feature works in conjunction with the kernel
1381 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1382 up the pass1 processing of e2fsck. This is a read-only feature
1383 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1384 mount filesystems that have had this feature enabled.
1386 Severity : enhancement
1388 Description: Improve multi-block allocation algorithm to avoid fragmentation
1389 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1390 allocation locality and avoid filesystem fragmentation.
1392 ------------------------------------------------------------------------------
1394 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1396 * Support for kernels:
1397 2.6.9-42.0.3.EL (RHEL 4)
1398 2.6.5-7.276 (SLES 9)
1399 2.4.21-47.0.1.EL (RHEL 3)
1400 2.6.12.6 vanilla (kernel.org)
1401 2.6.16.21-0.8 (SLES10)
1402 * Recommended e2fsprogs version: 1.39.cfs2-0
1404 * The backwards-compatible /proc/sys/portals symlink has been removed
1405 in this release. Before upgrading, please ensure that you change
1406 any configuration scripts or /etc/sysctl.conf files that access
1407 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1408 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1409 in advance of the upgrade on any system running Lustre 1.4.6 or
1410 newer, since /proc/sys/lnet was added in that version.
1411 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1415 Frequency : only when quota is used
1417 Description: avoid scanning export list for quota master
1418 Details : Change the algorithms to avoid scanning export list in order
1419 to improve the efficiency.
1422 Frequency : MDS failover only, very rarely
1424 Description: "went back in time" messages on mds failover
1425 Details : The greatest transno may be lost when the current operation
1426 finishes with an error (transno==0) and the client's last_rcvd
1427 record is over-written. Save the greatest transno in the
1428 mds_last_transno for this case.
1431 Frequency : always for specific kernels and striping counts
1433 Description: client may get "Matching packet too big" without ACL support
1434 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1435 when trying to access files in certain configurations. The
1436 clients should in fact be denied when mounting because they do
1437 not understand ACLs.
1440 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1442 Description: many clients connecting with IO in progress causes connect timeouts
1443 Details : Avoid synchronous journal commits to avoid delays caused by many
1444 clients connecting/disconnecting when bulk IO is in progress.
1445 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1446 OST_IO_PORTAL to avoid delays behind potentially many pending
1450 Frequency : occasionally with multiple writers to a single file
1452 Description: shared writes to file may result in wrong size reported by stat()
1453 Details : Allow growing of kms when extent lock is cancelled
1456 Frequency : always with random mmap IO to multi-striped file
1458 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1459 Details : Only if the hole is at the end of OST object so that kms is too
1460 small. Fix is to increase kms accordingly in ll_nopage.
1463 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1465 Description: writes to a missing object would leak memory on the OST
1466 Details : If there is an inconsistency between the MDS and OST filesystems,
1467 such that the MDS references an object that doesn't exist, writes
1468 to that object will leak memory due to incorrect cleanup in the
1469 error handling path, eventually running out of memory on the OST.
1474 Description: Creating too long symlink causes lustre errors
1475 Details : Check symlink and name lengths before sending requests to MDS.
1478 Frequency : only if flock is enabled (not on by default)
1480 Description: posix locks not released on fd closure on 2.6.9+
1481 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1482 kernels, this caused such locks not to be released on fd close and
1483 then assertions on fs unmount about still used locks.
1486 Frequency : MDS failover only, very rarely
1488 Description: clients may get ASSERTION(granted_lock != NULL)
1489 Details : When request was taking a long time, and a client was resending
1490 a getattr by name lock request. The were multiple lock requests
1491 with the same client lock handle and
1492 mds_getattr_name->fixup_handle_for_resent_request found one of the
1493 lock handles but later failed with ASSERTION(granted_lock != NULL).
1498 Description: handle->h_buffer_credits > 0, assertion failure
1499 Details : h_buffer_credits is zero after truncate, causing assertion
1500 failure. This patch extends the transaction or creates a new
1504 Frequency : NFS re-export or patchless client
1505 Bugzilla : 11179, 10796
1506 Description: Crash on NFS re-export node (__d_move)
1507 Details : We do not want to hash the dentry if we don't have a lock.
1508 But if this dentry is later used in d_move, we'd hit uninitialised
1509 list head d_hash, so we just do this to init d_hash field but
1510 leave dentry unhashed.
1513 Frequency : NFS re-export or patchless client
1515 Description: NFS exports has problem with symbolic link
1516 Details : lustre client didn't properly install dentry when re-exported
1517 to NFS or running patchless client.
1520 Frequency : NFS re-export or patchless client
1522 Description: Various nfs/patchless fixes.
1523 Details : fixes reuse disconected alias for lookup process - this fixes
1524 warning "find_exported_dentry: npd != pd",
1525 fix permission error with open files at nfs.
1526 fix apply umask when do revalidate.
1529 Frequency : occasional
1531 Description: Crash on NFS re-export node
1532 Details : calling clear_page() on the wrong pointer triggered oops in
1533 generic_mapping_read().
1536 Frequency : rarely, using O_DIRECT IO
1538 Description: unaligned directio crashes client with LASSERT
1539 Details : check for unaligned buffers before trying any requests.
1542 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1544 Description: stale data returned from RAID cache
1545 Details : If only a small amount of IO is done to the RAID device before
1546 reading it again it is possible to get stale data from the RAID
1547 cache instead of reading it from disk.
1550 Frequency : always for sles10 kernel
1552 Description: sles10 support
1553 Details : ll_follow_link: compile fixes and using of nd_set_link
1554 under newer kernels.
1557 Frequency : depends on arch, kernel and compiler version, always on sles10
1560 Description: recursive or deep enough symlinks cause stack overflow
1561 Details : getting rid of large stack-allocated variable in
1565 Frequency : depends on hardware
1567 Description: lustre write performance loss in the SLES10 kernel
1568 Details : the performance loss is caused by using of write barriers in the
1569 ext3 code. The SLES10 kernel turns barrier support on by
1570 default. The fix is to undo that change for ldiskfs.
1572 ------------------------------------------------------------------------------
1574 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1576 * Support for kernels:
1577 2.6.9-42.0.3EL (RHEL 4)
1578 2.6.5-7.276 (SLES 9)
1579 2.4.21-47.0.1.EL (RHEL 3)
1580 2.6.12.6 vanilla (kernel.org)
1584 Frequency : quota enabled and large files being deleted
1586 Description: releasing more than 4GB of quota at once hangs OST
1587 Details : If a user deletes more than 4GB of files on a single OST it
1588 will cause the OST to spin in an infinite loop. Release
1589 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1594 Description: statfs data retrieved from /proc may be stale or zero
1595 Details : When reading per-device statfs data from /proc, in the
1596 {kbytes,files}_{total,free,avail} files, it may appear
1597 as zero or be out of date.
1600 Frequency : systems with MD RAID1 external journal devices
1602 Description: lconf's call to blkid is confused by RAID1 journal devices
1603 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1604 of returning all block devices that match the journal UUID.
1607 Frequency : always, for aggregate stripe size over 4GB
1609 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1610 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1611 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1614 Frequency : always if "lfs find" used on a local file/directory
1616 Description: "lfs find" segfaults if used on a local file/directory
1617 Details : The case where a directory component was not specified wasn't
1618 handled correctly. Handle this properly.
1621 Frequency : always on ppc64
1623 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1624 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1625 perform endianness conversion.
1628 Frequency : rarely (truncate to non-zero file size after write under load)
1629 Bugzilla : 10730, 10687
1630 Description: Files padded with zeros to next 4K multiple
1631 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1632 it is possible that files that are truncated to a non-zero size
1633 immediately after being written are filled with zero bytes beyond
1634 the truncated size. No file data is lost.
1636 Severity : enhancement
1638 Description: Allow recovery/failover for liblustre clients.
1639 Details : liblustre clients were unaware of failover configurations until
1642 Severity : enhancement
1644 Description: user file locks should fail when not mounting with flock option
1645 Details : Set up an error-returning stub in ll_file_operations.lock field
1646 to prevent incorrect behaviour when client is mounted without
1647 flock option. Also, set up properly f_op->flock field for
1651 Frequency : always on ia64
1653 Description: "lfs df" loops on printing out MDS statfs information
1654 Details : The obd_ioctl_data was not initialized and in some systems
1655 this caused a failure during the ioctl that did not return
1656 an error. Initialize the struct and return an error on failure.
1659 Frequency : SLES 9 only
1661 Description: Error of copying files with lustre special EAs as root
1662 Details : Client side always return success for setxattr call for lustre
1663 special xattr (currently only "trusted.lov").
1666 Frequency : rarely on clusters with both ia64+i386 clients
1668 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1669 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1670 clients are doing concurrent writes to the same file it is
1671 possible that the ia64 clients may LASSERT because the OST
1672 extent locks are not PAGE_SIZE aligned. Ensure that grown
1673 locks are always aligned on the request boundary.
1676 Frequency : specific use, occasional
1678 Description: Overwriting in use executable truncates on-disk binary image
1679 Details : If one node attempts to overwrite an executable in use by
1680 another node, we now correctly return ETXTBSY instead of
1681 truncating the file.
1683 Severity : enhancement
1685 Description: Async OSC create to avoid the blocking unnecessarily.
1686 Details : If a OST has no remain object, system will block on the creating
1687 when need to create a new object on this OST. Now, ways use
1688 pre-created objects when available, instead of blocking on an
1689 empty osc while others are not empty. If we must block, we block
1690 for the shortest possible period of time.
1695 Description: chmod on Lustre root is propagated to other clients
1696 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1697 invalid by the follow_mount time.
1700 Frequency : liblustre clients only
1702 Description: Race in 'instant cancel' lock handling could lead to such locks
1703 never to be granted in case of SMP MDS
1704 Details : Do not destroy not yet granted but cbpending locks in
1708 Frequency : replay/resend of open
1710 Description: non null lock assetion failure in mds_intent_policy
1711 Details : Trying to replay/resend lockless open requests resulted in
1712 mds_open() returning 0 with no lock. Now it sets a flag if
1713 a lock is going to be returned.
1715 Severity : enhancement
1717 Description: Checksum enhancements
1718 Details : New checksum enhancements allow for resending RPCs that failed
1721 Severity : enhancement
1723 Description: Tunables on number of dirty pages in cacche
1724 Details : Allow to set limit on number of dirty pages cached.
1729 Description: client crash on unmount - lock still has references
1730 Details : In some error handling cases it was possible to leak a lock
1731 reference on a client while accessing a file. This was not
1732 harmful to the client during operation, but would cause the
1733 client to crash when the filesystem is unmounted.
1736 Frequency : specific case, rare
1738 Description: ETXTBSY on mds though file not in use by client
1739 Details : ETXTBSY is no longer incorrectly returned when attempting to
1740 chmod or chown a directory that the user previously tried to
1741 execute or a currently-executing binary.
1744 Frequency : extremely rare except on liblustre-based clients
1746 Description: Lustre space not freed when files are deleted
1747 Details : Clean up open-unlinked files after client eviction. Previously
1748 the unlink was skipped and the files remained as orphans.
1753 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1754 Details : In some cases it was possible to send a blocking callback to
1755 a client doing a glimpse, even though that client didn't get
1756 a lock granted. When the glimpse lock is cancelled on the OST
1757 the freed lock is left on the waiting list and corrupted the list.
1760 Frequency : all core dumps
1762 Description: Broke core dumps to lustre
1763 Details : Negative dentry may be unhashed if parent does not have UPDATE
1764 lock, but some callers, e.g. do_coredump, expect dentry to be
1765 hashed after successful create, hash it in ll_create_it.
1767 ------------------------------------------------------------------------------
1769 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1771 * Support for kernels:
1772 2.6.9-42.0.2.EL (RHEL 4)
1773 2.6.5-7.276 (SLES 9)
1774 2.4.21-40.EL (RHEL 3)
1775 2.6.12.6 vanilla (kernel.org)
1779 Frequency : always on RHEL 3
1781 Description: Number of open files grows over time
1782 Details : The number of open files grows over time, whether or not
1783 Lustre is started. This was due to a filp leak introduced
1784 by one of our kernel patches.
1786 ------------------------------------------------------------------------------
1788 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1790 * Support for kernels:
1791 2.6.9-42.EL (RHEL 4)
1792 2.6.5-7.276 (SLES 9)
1793 2.4.21-40.EL (RHEL 3)
1794 2.6.12.6 vanilla (kernel.org)
1799 Bugzilla : 5719, 9635, 9792, 9684
1800 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1801 Details : If a server is under heavy load and cannot reply to new
1802 connection requests before the client resends the (re)connect,
1803 the connection handling code can behave badly if two service
1804 threads are concurrently handing separate (re)connections from
1805 the same client. Add better locking to the connection handling
1806 code, and ensure that only a single connection will be processed
1807 for a given client UUID, even if the lock is dropped.
1809 Severity : enhancement
1811 Description: add TCP zero-copy support to kernel
1812 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1813 sends if the hardware supports scatter-gather and checksumming.
1814 This allows socklnd to do client-write and server-read more
1815 efficiently and reduce CPU utilization from skbuf copying.
1818 Frequency : only if NFS exporting from client
1820 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1821 Details : If setting attributes on a file created under NFS that had
1822 never been opened it would be possible to oops the client
1823 if the file had no objects.
1826 Frequency : always for liblustre
1828 Description: liblustre client does MDS+OSTs setattr RPC for each write
1829 Details : When doing a write from a liblustre client, the client
1830 incorrectly issued an RPC to the MDS and each OST the file was
1831 striped over in order to update the timestamps. When writing
1832 with small chunks and many clients this could overwhelm the MDS
1833 with RPCs. In all cases it would slow down the write because
1834 these RPCs are unnecessary.
1836 Severity : enhancement
1838 Description: allow number of MDS service threads to be changed at module load
1839 Details : It is now possible to change the number of MDS service threads
1840 running. Adding "options mds mds_num_threads={N}" to the MDS's
1841 /etc/modprobe.conf will set the number of threads for the next
1842 time Lustre is restarted (assuming the "mds" module is also
1843 reloaded at that time). The default number of threads will
1844 stay the same, 32 for most systems.
1849 Description: OST crash if filesystem is unformatted or corrupt
1850 Details : If an OST is started on a device that has never been formatted
1851 or if the filesystem is corrupt and cannot even mount then the
1852 error handling cleanup routines would dereference a NULL pointer.
1857 Description: NULL pointer deref in llap_from_page.
1858 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1859 incorrect) mapping if the page was truncated/reclaimed while it was
1860 searched for. Check for this condition and skip such pages when
1861 doing readahead. Introduce extra check to llap_from_page() to
1862 verify page->mapping->host is non-NULL (so page is not anonymous).
1865 Frequency : Sometimes when using sys_sendfile
1867 Description: "page not covered by a lock" warnings from ll_readpage
1868 Details : sendfile called ll_readpage without right page locks present.
1869 Now we introduced ll_file_sendfile that does necessary locking
1870 around call to generic_file_sendfile() much like we do in
1874 Frequency : with certain MDS communication failures at client mount time
1876 Description: NULL pointer deref after failed client mount
1877 Details : a client connection request may delayed by the network layer
1878 and not be sent until after the PTLRPC layer has timed out the
1879 request. If the client fails the mount immediately it will try
1880 to clean up before the network times out the request. Add a
1881 reference from the request import to the obd device and delay
1882 the cleanup until the network drops the request.
1885 Frequency : occasionally during client (re)connect
1887 Description: assertion failure during client (re)connect
1888 Details : processing a client connection request may be delayed by the
1889 client or server longer than the client connect timeout. This
1890 causes the client to resend the connection request. If the
1891 original connection request is replied in this interval, the
1892 client may trip an assertion failure in ptlrpc_connect_interpret()
1893 which thought it would be the only running connect process.
1896 Frequency : only with obd_echo servers and clients that are rebooted
1898 Description: kernel BUG accessing uninitialized data structure
1899 Details : When running an obd_echo server it did not start the ping_evictor
1900 thread, and when a client was evicted an uninitialized data
1901 structure was accessed. Start the ping_evictor in the RPC
1902 service startup instead of the OBD startup.
1904 Severity : enhancement
1905 Bugzilla : 10193 (patchless)
1906 Description: Remove dependency on various unexported kernel interfaces.
1907 Details : No longer need reparent_to_init, exit_mm, exit_files,
1908 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1911 Frequency : rare (only users of deprecated and unsupported LDAP config)
1913 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1914 Details : LDAP apparently contains 'lustreName' attributes instead of
1915 'name'. A simple remapping of the name is sufficient.
1918 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1920 Description: waiting_locks_callback trips kernel BUG if client is evicted
1921 Details : Running with the dump_on_timeout debug flag turned on makes
1922 it possible that the waiting_locks_callback() can try to dump
1923 the Lustre kernel debug logs from an interrupt handler. Defer
1924 this log dumping to the expired_lock_main() thread.
1926 Severity : enhancement
1928 Description: Support NFS exporting on 2.6 kernels.
1929 Details : Implement non-rawops metadata methods for NFS server to use without
1930 changing NFS server code.
1933 Frequency : very rare (synthetic metadata workload only)
1935 Description: two racing renames might cause an MDS thread to deadlock
1936 Details : Running the "racer" program may cause one MDS thread to rename
1937 a file from being the source of a rename to being the target of
1938 a rename at exactly the same time that another thread is doing
1939 so, and the second thread has already enqueued these locks after
1940 doing a lookup of the target and is trying to relock them in
1941 order. Ensure that we don't try to re-lock the same resource.
1944 Frequency : only very large systems with liblustre clients
1946 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1947 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1948 all OSTs in parallel. This allows the eviction of stale liblustre
1949 clients to proceed much faster than if they were done in series,
1950 and also offers similar improvements for other set_info RPCs.
1955 Description: excessive CPU usage during initial read phase on client
1956 Details : During the initial read phase on a client, it would agressively
1957 retry readahead on the file, consuming too much CPU and impacting
1958 performance (since 1.4.5.8). Improve the readahead algorithm
1959 to avoid this, and also improve some other common cases (read
1960 of small files in particular, where "small" is files smaller than
1961 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1966 Description: MDS crash when receiving packet with unknown intent.
1967 Details : Do not LBUG in unknown intent case, just return -EFAULT
1969 Severity : enhancement
1970 Bugzilla : 9293, 9385
1971 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1972 Details : Do not serialize getattr (non-intent version) and statfs.
1975 Frequency : occasional, when OST network is overloaded/intermittent
1977 Description: client evicted by OST after bulk IO timeout
1978 Details : If a client sends a bulk IO request (read or write) the OST
1979 may evict the client if it is unresposive to its data GET/PUT
1980 request. This is incorrect if the network is overloaded (takes
1981 too long to transfer the RPC data) or dropped the OST GET/PUT
1982 request. There is no need to evict the client at all, since
1983 the pinger and/or lock callbacks will handle this, and the
1984 client can restart the bulk request.
1987 Frequency : Always when mmapping file with no objects
1989 Description: client crashes when mmapping file with no objects
1990 Details : Check that we actually have objects in a file before doing any
1991 operations on objects in ll_vm_open, ll_vm_close and
1997 Description: Request leak when working with deleted CWD
1998 Details : Introduce advanced request refcount tracking for requests
1999 referenced from lustre intent.
2001 Severity : Enhancement
2003 Description: Cache open file handles on client.
2004 Details : MDS now will return special lock along with openhandle, if
2005 requested and client is allowed to hold openhandle, even if unused,
2006 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2007 closing files for every read/write openration.
2009 Severity : Enhancement
2011 Description: Cache open negative dentries on client when possible.
2012 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2013 negative dentries on lock revocation.
2018 Description: Remounting a client read-only wasn't possible with a zconf mount
2019 Details : It wasn't possible to remount a client read-only with llmount.
2021 Severity : enhancement
2022 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2023 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2024 MPICH is included to add Lustre-specific ADIO interfaces.
2025 This is based closely on the UFS ADIO layer and only differs
2026 in file creation, in order to allow the OST striping to be set.
2027 This is user-contributed code and not supported by CFS.
2032 Description: extended inode attributes (immutable, append-only) work improperly
2033 when 2.4 and 2.6 kernels are used on client/server or vice versa
2034 Details : Introduce kernel-independent values for these flags.
2036 Severity : enhancement
2039 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2040 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2041 granularity. Now, Fractional MB granularity is be supported,
2042 this is very useful for low memory system.
2044 Severity : enhancement
2046 Description: Getattr by fid
2047 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2048 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2049 have locks and avoids getattr rpc requests for it on every stat.
2052 Frequency : Always, for filesystems larger than 2TB
2054 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2055 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2056 filesystems larger than 2TB to be mounted with mballoc enabled.
2059 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2061 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2062 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2063 filesystems larger than 2TB is mounted then the kernel will
2064 silently corrupt the start of the filesystem. CONFIG_LBD is
2065 enabled for all CFS-supported kernels, but the possibility of
2066 this happening with a modified kernel config exists.
2068 Severity : enhancement
2070 Description: add client O_DIRECT support for 2.6 kernels
2071 Details : It is now possible to do O_DIRECT reads and writes to files
2072 in the Lustre client mountpoint on 2.6 kernel clients.
2074 Severity : enhancement
2076 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2077 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2078 parallel, not waiting for response from every OST before sending
2079 a rpc to the next OST.
2084 Description: setattr vs write race when updating file timestamps
2085 Details : Client processes that update a file timestamp into the past
2086 right after writing to the file (e.g. tar) it is possible that
2087 the updated file modification time can be reset to the current
2088 time due to a race between processing the setattr and write RPC.
2090 Severity : enhancement
2092 Description: Bring 'lfs find' closer in line with regular Linux find.
2093 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2094 -print0 options and obtains all the needed info through the lustre
2097 Severity : enhancement
2099 Description: support up to 1024 configured devices on one node
2100 Details : change obd_dev array from statically allocated to dynamically
2101 allocated structs as they are first used to reduce memory usage
2106 Description: Flush dirty partially truncated pages during truncate
2107 Details : Immediatelly flush partially truncated pages in filter_setattr,
2108 this way we completely avoid having any pages in page cache on OST
2109 and can retire ugly workarounds during writes to flush such pages.
2114 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2115 Details : protect inode from truncation within vfs_unlink() context
2116 just take a reference before calling vfs_unlink() and release it
2117 when parent's i_sem is free.
2122 Description: last_id value checked outside lock on OST caused LASSERT failure
2123 Details : If there were multiple MDS->OST object precreate requests in
2124 flight, it was possible that the OST's last object id was checked
2125 outside a lock and incorrectly tripped an assertion. Move checks
2126 inside locks, and discard old precreate requests.
2129 Frequency : always, if extents are used on OSTs
2131 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2132 Details : OSTs using the extents format would not zero the high 16 bits of
2133 the index physical block number. This is not a problem for any
2134 OST filesystems smaller than 16TB, and no kernels support ext3
2135 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2136 new/modified files) and can be fixed for existing filesystems
2137 with e2fsprogs-1.39-cfs1.
2142 Description: import connection selection may be incorrect if timer wraps
2143 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2144 import connections to be ignored if the 32-bit jiffies counter
2145 wraps. Use a 64-bit jiffies counter.
2148 Frequency : very large clusters immediately after boot
2150 Description: LNET request buffers exhausted under heavy short-term load
2151 Details : If a large number of client requests are generated on a service
2152 that has previously never seen so many requests it is possible
2153 that the request buffer growth cannot keep up with the spike in
2154 demand. Instead of dropping incoming requests, they are held in
2155 the LND until the RPC service can accept more requests.
2158 Frequency : Sometimes during replay
2160 Description: Assertion failure in ll_local_open after replay.
2161 Details : If replay happened on an open request reply before we were able
2162 to set replay handler, reply will become not swabbed tripping the
2163 assertion in ll_local_open. Now we set the handler right after
2164 recognising of open request
2167 Frequency : very rare
2169 Description: kernel reports "badness in vsnprintf"
2170 Details : Reading from the "recovery_status" /proc file in small chunks
2171 may cause a negative length in lprocfs_obd_rd_recovery_status()
2172 call to vsnprintf() (which is otherwise harmless). Exit early
2173 if there is no more space in the output buffer.
2175 Severity : enhancement
2177 Description: clear OBD RPC statistics by writing to them
2178 Details : It is now possible to clear the OBD RPC statistics by writing
2179 to the "stats" file.
2184 Description: Client mtime is not the same on different clients after utimes
2185 Details : In some cases, the client was using the utimes() syscall on
2186 a file cached on another node. The clients now validate the
2187 ctime from the MDS + OSTs to determine which one is right.
2192 Description: Inability to activate failout mode
2193 Details : lconf script incorrectly assumed that in pythong string's numeric
2194 value is used in comparisons.
2197 Frequency : always with multiple stripes per file
2199 Description: Inefficient object allocation for mutli-stripe files
2200 Details : When selecting which OSTs to stripe files over, for files with
2201 a stripe count that divides evenly into the number of OSTs,
2202 the MDS is always picking the same starting OST for each file.
2203 Return the OST selection heuristic to the original design.
2208 Description: mount failures may take full timeout to return an error
2209 Details : Under some heavy load conditions it is possible that a
2210 failed mount can wait for the full obd_timeout interval,
2211 possibly several minutes, before reporting an error.
2212 Instead return an error as soon as the status is known.
2214 Frequency : quota enabled and large files being deleted
2216 Description: releasing more than 4GB of quota at once hangs OST
2217 Details : If a user deletes more than 4GB of files on a single OST it
2218 will cause the OST to spin in an infinite loop. Release
2219 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2224 Description: statfs data retrieved from /proc may be stale or zero
2225 Details : When reading per-device statfs data from /proc, in the
2226 {kbytes,files}_{total,free,avail} files, it may appear
2227 as zero or be out of date.
2230 Frequency : systems with MD RAID1 external journal devices
2232 Description: lconf's call to blkid is confused by RAID1 journal devices
2233 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2234 of returning all block devices that match the journal UUID.
2237 Frequency : always, for aggregate stripe size over 4GB
2239 Description: assertion fails when trying to use 4GB stripe size
2240 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2241 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2244 Frequency : always on ppc64
2246 Description: the first write on an ext3 filesystem with mballoc got stuck
2247 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2248 perform endianness conversion.
2250 ------------------------------------------------------------------------------
2252 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2254 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2255 INTEROPERATE with older versions automatically. Please read the
2256 user documentation before upgrading any part of a live system.
2257 * WARNING: Lustre networking configuration changes are required with
2258 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2261 * Support for kernels:
2262 2.6.9-22.0.2.EL (RHEL 4)
2263 2.6.5-7.244 (SLES 9)
2264 2.6.12.6 vanilla (kernel.org)
2267 Severity : enhancement
2268 Bugzilla : 7981/8208
2269 Description: Introduced Lustre Networking (LNET)
2270 Details : LNET is new networking infrastructure for Lustre, it includes
2271 a reorganized network configuration mode (see the user
2272 documentation for full details) as well as support for routing
2273 between different network fabrics. Lustre Networking Devices
2274 (LNDs) for the supported network fabrics have also been
2275 created for this new infrastructure.
2277 Severity : enhancement
2278 Description: Introduced Access control lists
2279 Details : clients can set ACLs on files and directories in order to have
2280 more fine-grained permissions than the standard Unix UGO+RWX.
2281 The MDS must be started with the "-o acl" mount option.
2283 Severity : enhancement
2284 Description: Introduced filesystem quotas
2285 Details : Administrators may now establish per-user quotas on the
2288 Severity : enhancement
2290 Description: Configuration change for the XT3
2291 The PTLLND is now used to run Lustre over Portals on the XT3
2292 The configure option(s) --with-cray-portals are no longer used.
2293 Rather --with-portals=<path-to-portals-includes> is used to
2294 enable building on the XT3. In addition to enable XT3 specific
2295 features the option --enable-cray-xt3 must be used.
2300 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2301 Details : A race between allocating a new llcd and re-getting the llcd_lock
2302 allowed another thread to grab newly-allocated llcd.
2304 Severity : enhancement
2306 Description: 2.6 OST async journal commit and locking fix to improve performance
2307 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2308 2.6 kernels are now async as they already were in 2.4 kernels so
2309 that they can commit concurrently with the network bulk transfer.
2310 For block-allocated files the filter allocation semaphore is held
2311 to avoid filesystem fragmentation during allocation. BKL lock
2312 removed for 2.6 xattr operations where it is no longer needed.
2317 Description: lconf incorrectly determined whether two IP networks could talk
2318 Details : In some more complicated routing and multiple-network
2319 configurations, lconf will avoid trying to make a network
2320 connection to a disjoint part of the IP space. It was doing the
2321 math incorrectly for one set of cases.
2326 Description: Fix for potential infinite loop processing records in an llog.
2327 Details : If an llog record is corrupted/zeroed, it is possible to loop
2328 forever in llog_process(). Validate the llog record length
2329 and skip the remainder of the block on error.
2332 Frequency : occasional (liblustre only)
2334 Description: liblustre could not open files whose last component is a symlink
2335 Details : sysio_path_walk() would incorrectly pass the open intent to
2336 intermediate path components.
2339 Frequency : rare (liblustre only with non-standard tuning)
2340 Bugzilla : 7201 (7350)
2341 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2342 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2343 after it has been released, only do this for referenced locks
2345 Severity : enhancement
2347 Description: specifying an (invalid) directory default stripe_size of -1
2348 would reset the directory default striping
2349 Details : stripe_size -1 was used internally to signal directory stripe
2350 removal, now use "all default" to signal dir stripe removal
2351 as a directory striping of "all default" is not useful
2354 Frequency : common for large clusters running liblustre clients
2356 Description: doing an ls when liblustre clients are running is slow
2357 Details : sending a glimpse AST to a liblustre client waits for every AST
2358 to time out, as liblustre clients will not respond. Since they
2359 cannot cache data we refresh the OST lock LVB from disk instead.
2361 Severity : enhancement
2363 Description: doing an ls at the same time as file IO can be slow
2364 Details : enqueue and other "small" requests can be blocked behind many
2365 large IO requests. Create a new OST IO portal for non-IO
2366 requests so they can be processed faster.
2369 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2371 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2372 created file with mode 000
2373 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2374 allow an ftruncate() as a non-root user to a file with mode 000.
2375 Lustre now respects this flag to disable mode checks when
2376 truncating a file owned by the user
2379 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2382 Description: Revoking locks from clients that went dead or catatonic might take
2384 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2385 cancellation of such locks instant on servers without waiting for
2386 any reply from clients. Clients drops these locks when cancel
2387 notification from server is received without replying.
2390 Frequency : liblustre-only, when liblustre client dies or becomes busy
2392 Description: Doing ls on Linux clients can take a long time with active
2394 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2395 granting such locks to it in the first place if possible. Locks
2396 are taken by proxy on the OST during the read or write and
2397 dropped immediately afterward. Add connect flags handling, do
2398 not grant locks to liblustre clients for glimpse ASTs.
2400 Severity : enhancement
2402 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2403 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2404 sequential reads and behaves badly with applications doing random
2405 reads. Improve it by reading ahead at least read region, and
2406 avoiding excessive large RPC for small reads.
2408 Severity : enhancement
2410 Description: Creating more than 1000 files for a single job may cause a load
2411 imbalance on the OSTs if there are also a large number of OSTs.
2412 Details : qos_prep_create() uses an OST index reseed value that is an
2413 even multiple of the number of available OSTs so that if the
2414 reseed happens in the middle of the object allocation it will
2415 still utilize the OSTs as uniformly as possible.
2420 Description: OST or MDS may oops in ping_evictor_main()
2421 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2422 but doesn't restart at beginning of obd_exports_timed list
2425 Severity : enhancement
2427 Description: improve by-nid export eviction on the MDS and OST
2428 Details : allow multiple exports with the same NID to be evicted at one
2429 time without re-searching the exports list.
2432 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2434 Description: MDS may oops in groups_free()
2435 Details : in rare race conditions a newly allocated group_info struct is
2436 freed again, and this can be NULL. The 2.4 compatibility code
2437 for groups_free() checked for a NULL pointer, but 2.6 did not.
2440 Frequency : common for liblustre clients doing little filesystem IO
2441 Bugzilla : 9352, 7313
2442 Description: server may evict liblustre clients accessing contended locks
2443 Details : if a client is granted a lock or receives a completion AST
2444 with a blocking AST already set it would not reply to the AST
2445 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2448 Frequency : lfs setstripe, only systems with more than 160 OSTs
2450 Description: unable to set striping with a starting offset beyond OST 160
2451 Details : llapi_create_file() incorrectly limited the starting stripe
2452 index to the maximum single-file stripe count.
2455 Frequency : LDAP users only
2457 Description: lconf did not handle in-kernel recovery with LDAP properly
2458 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2460 Severity : enhancement
2462 Description: bind OST threads to NUMA nodes to improve performance
2463 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2464 node and do their allocations there to localize memory access
2466 Severity : enhancement
2468 Description: llmount can determine client NID directly from Myrinet (GM)
2469 Details : the client NID code from gmnalnid was moved directly into
2470 llmount, removing the need to use this or specifying the
2471 client NID explicitly when mounting GM clients with zeroconf
2474 Frequency : if client is started with down MDS
2476 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2477 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2478 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2479 ptlrpc_import_delayed_req() because we didn't check if the
2480 request was interrupted, and we also didn't break out of the
2481 event loop if there was no timeout
2486 Description: data loss during non-page-aligned writes to a single file from
2487 both multiple nodes and multiple threads on one node at same time
2488 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2489 inconsistency led to false short-reads, that were cached and later
2490 used by ->prepare_write() to fill in partially written page,
2491 leading to data loss.
2494 Frequency : always, if lconf --abort_recovery used
2496 Description: lconf --abort_recovery fails with 'Operation not supported'
2497 Details : lconf was attempting to abort recovery on the MDT device and not
2500 Severity : enhancement
2502 Description: remove cleanup logs
2503 Details : replace lconf-generated cleanup logs with lustre internal
2504 cleanup routines. Eliminates the need for client-cleanup and
2507 Severity : enhancement
2509 Description: add support for EAs (user and system) on lustre filesystems
2510 Details : it is now possible to store extended attributes in the Lustre
2511 client filesystem, and with the user_xattr mount option it
2512 is possible to allow users to store EAs on their files also
2514 Severity : enhancement
2516 Description: Add possibility (config option) to show minimal available OST free
2518 Details : When compiled with --enable-mindf configure option, statfs(2)
2519 (and so, df) will return least minimal free space available from
2520 all OSTs as amount of free space on FS, instead of summary of
2521 free spaces of all OSTs.
2523 Severity : enhancement
2525 Description: do not expand extent locks acquired on OST-side
2526 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2527 by server: they are not cached anyway.
2530 Frequency : when mmap is used/binaries executed from Lustre
2532 Description: Unmmap pages before throwing them away from read cache.
2533 Details : llap_shrink cache now attempts to unmap pages before discarding
2534 them (if unmapping failed - do not discard). SLES9 kernel has
2535 extra checks that trigger if this unmapping is not done first.
2540 Description: lconf didn't resolve symlinks before checking to see whether a
2541 given mountpoint was already in use
2544 Frequency : when migrating failover services
2545 Bugzilla : 6395, 9514
2546 Description: When migrating a subset of services from a node (e.g. failback
2547 from a failover service node) the remaining services would
2548 time out and evict clients.
2549 Details : lconf --force (implied by --failover) sets the global obd_timeout
2550 to 5 seconds in order to quickly disconnect, but this caused
2551 other RPCs to time out too quickly. Do not change the global
2552 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2554 Severity : enhancement
2555 Frequency : if MDS is started with down OST
2556 Bugzilla : 9439,5706
2557 Description: Allow startup/shutdown of an MDS without depending on the
2558 availability of the OSTs.
2559 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2560 Add appropriate locking and lov-osc refcounts for safe
2561 cleaning. Add osc abort_inflight calls in case the
2562 synchronize never started.
2565 Frequency : occasional (Cray XT3 only)
2567 Description: root not authorized to access files in CRAY_PORTALS environment
2568 Details : The client process capabilities were not honoured on the MDS in
2569 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2570 been accessed by an authorized user then root was able to access
2571 the file on the local client also. The root user capabilities
2572 are now allowed on the MDS, as this environment has secure UID.
2575 Frequency : occasional
2577 Description: ldiskfs "too long searching" message happens too often
2578 Details : A debugging message (otherwise harmless) prints too often on
2579 the OST console. This has been reduced to only happen when
2580 there are fragmentation problems on the filesystem.
2585 Description: Division by zero in statfs when all OSCs are inactive
2586 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2587 lov_statfs divides by value returned by lov_get_stripecnt().
2591 Bugzilla : 9489, 3273
2592 Description: First write from each client to each OST was only 4kB in size,
2593 to initialize client writeback cache, which caused sub-optimal
2594 RPCs and poor layout on disk for the first writen file.
2595 Details : Clients now request an initial cache grant at (re)connect time
2596 and so that they can start streaming writes to the cache right
2597 away and always do full-sized RPCs if there is enough data.
2598 If the OST is rebooted the client also re-establishes its grant
2599 so that client cached writes will be honoured under the grant.
2604 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2605 Details : Now I/O RPCs go to different portal number and (presumably) fast
2606 lock requests (and glimses) and other RPCs get their own service
2607 threads pool that should be able to service those RPCs
2610 Severity : enhancement
2612 Description: Ability to exchange lustre version between client and servers and
2613 issue warnings at client side if client is too old. Also for
2614 liblustre clients there is ability to refuse connection of too old
2616 Details : New 'version' field is added to connect data structure that is
2617 filled with version info. That info is later checked by server and
2621 Frequency : rare, liblustre only.
2622 Bugzilla : 9296, 9581
2623 Description: Two simultaneous writes from liblustre at offset within same page
2624 might proceed at the same time overwriting eachother with stale
2626 Details : I/O lock withing llu_file_prwv was released too early, before data
2627 actually was hitting the wire. Extended lock-holding time until
2628 server acknowledges receiving data.
2631 Frequency : extremely rare. Never observed in practice.
2633 Description: avoid generating lustre_handle cookie of 0.
2634 Details : class_handle_hash() generates handle cookies by incrementing
2635 global counter, and can hit 0 occasionaly (this is unlikely, but
2636 not impossible, because initial value of cookie counter is
2637 selected randonly). Value of 0 is used as a sentinel meaning
2638 "unassigned handle" --- avoid it. Also coalesce two critical
2639 sections in this function into one.
2641 Severity : enhancement
2643 Description: allow liblustre clients to delegate truncate locking to OST
2644 Details : To avoid overhead of locking, liblustre client instructs OST to
2645 take extent lock in ost_punch() on client's behalf. New connection
2646 flag is added to handle backward compatibility.
2648 Severity : enhancement
2649 Bugzilla : 4928, 7341, 9758
2650 Description: allow number of OST service threads to be specified
2651 Details : a module parameter allows the number of OST service threads
2652 to be specified via "options ost ost_num_threads={N}" in the
2653 OSS's /etc/modules.conf or /etc/modprobe.conf.
2657 Bugzilla : 6146, 9635, 9895
2658 Description: servers crash with bad pointer in target_handle_connect()
2659 Details : In rare cases when a client is reconnecting it was possible that
2660 the connection request was the last reference for that export.
2661 We would temporarily drop the export reference and get a new
2662 one, but this may have been the last reference and the export
2663 was just destroyed. Get new reference before dropping old one.
2665 Severity : enhancement
2666 Frequency : if client is started with failover MDS
2668 Description: Allow multiple MDS hostnames in the mount command
2669 Details : Try to read the configuration from all specified MDS
2670 hostnames during a client mount in case the "primary"
2673 Severity : enhancement
2675 Description: Stop sending data to evicted clients as soon as possible.
2676 Details : Check if the client we are about to send or are sending data to
2677 was evicted already. (Check is done every second of waiting,
2678 for which l_wait_event interface was extended to allow checking
2679 of exit condition at specified intervals).
2682 Frequency : rare, normally only when NFS exporting is done from client
2684 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2686 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2689 Frequency : always, if srand() is called before liblustre initialization
2691 Description: Liblustre uses system PRNG disturbing its usage by user application
2692 Details : Introduce internal to lustre fast and high-quality PRNG for
2693 lustre usage and make liblustre and some other places in generic
2694 lustre code to use it.
2696 Severity : enhancement
2697 Bugzilla : 9477, 9557, 9870
2698 Description: Verify that the MDS configuration logs are updated when xml is
2699 Details : Check if the .xml configuration logs are newer than the config
2700 logs stored on the MDS and report an error if this is the case.
2701 Request --write-conf, or allow starting with --old_conf.
2703 Severity : enhancement
2705 Description: Handle symlinks in the path when checking if Lustre is mounted.
2706 Details : Resolve intermediate symlinks when checking if a client has
2707 mounted a filesystem to avoid duplicate client mounts.
2712 Description: lconf can hit an error exception but still return success.
2713 Details : The lconf command catches the Command error exception at the top
2714 level script context and will exit with the associated exit
2715 status, but doesn't ensure that this exit status is non-zero.
2720 Description: failure of ptlrpc thread startup can cause oops
2721 Details : Starting a ptlrpc service thread can fail if there are a large
2722 number of threads or the server memory is very fragmented.
2723 Handle this without oopsing.
2726 Frequency : always, only if liblustre and non-default acceptor port was used
2728 Description: liblustre cannot connect to servers with non-default acceptor port
2729 Details : tcpnal_set_default_params() was not called and was therefore
2730 ignoring the environment varaible TCPNAL_PORT, as well as other
2731 TCPNAL_ environment variables
2736 Description: two objects could be created on the same OST for a single file
2737 Details : If an OST is down, in some cases it was possible to create two
2738 objects on a single OST for a single file. No problems other
2739 than potential performance impact and spurious error messages.
2743 Bugzilla : 5681, 9562
2744 Description: Client may oops in ll_unhash_aliases
2745 Details : Client dcache may become inconsistent in race condition.
2746 In some cases "getcwd" can fail if the current directory is
2752 Description: Inode refcounting problems in NFS export code
2753 Details : link_raw functions used to call d_instantiate without obtaining
2754 extra inode reference first.
2758 Bugzilla : 9942, 9903
2759 Description: Referencing freed requests leading to crash, memleaks with NFS.
2760 Details : We used to require that call to ll_revalidate_it was always
2761 followed by ll_lookup_it. Also with revalidate_special() it is
2762 possible to call ll_revalidate_it() twice for the same dentry
2763 even if first occurence returned success. This fix changes semantic
2764 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2765 reference on a request referred from the intent.
2766 ll_intent_release() then releases such a request.
2769 Frequency : rare, normally benchmark loads only
2771 Description: unlinked inodes were kept in memory on the client
2772 Details : If a client is repeatedly creating and unlinking files it
2773 can accumulate a lot of stale inodes in the inode slab cache.
2774 If there is no other client load running this can cause the
2775 client node to run out of memory. Instead flush old inodes
2776 from client cache that have the same inode number as a new inode.
2779 Frequency : SLES9 2.6.5 kernel and long filenames only
2780 Bugzilla : 9969, 10379
2781 Description: utime reports stale NFS file handle
2782 Details : SLES9 uses out-of-dentry names in some cases, which confused
2783 the lustre dentry revalidation. Change it to always use the
2787 Frequency : rare, unless heavy write-truncate concurrency is continuous
2788 Bugzilla : 4180, 6984, 7171, 9963, 9331
2789 Description: OST becomes very slow and/or deadlocked during object unlink
2790 Details : filter_destroy() was holding onto the parent directory lock
2791 while truncating+unlinking objects. For very large objects this
2792 may block other threads for a long time and slow overall OST
2793 responsiveness. It may also be possible to get a lock ordering
2794 deadlock in this case, or run out of journal credits because of
2795 the combined truncate+unlink. Solution is to do object truncate
2796 first in one transaction without parent lock, and then do the
2797 final unlink in a new transaction with the parent lock. This
2798 reduces the lock hold time dramatically.
2801 Frequency : rare, 2.4 kernels only
2803 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2804 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2805 before unmounting their filesystem in order to avoid deadlock.
2806 The kernel_locked() function in 2.4 kernels only checks whether
2807 the kernel lock is held, not whether it is this process that is
2808 holding it as 2.6 kernels do.
2813 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2814 Details : The client ptlrpc code may be trying to reconnect to a down
2815 server before a previous connection attempt has timed out.
2816 Increase the reconnect interval to be longer than the connection
2817 timeout interval to avoid sending duplicate connections to
2821 Frequency : echo_client brw_test command
2823 Description: fix echo_client to work with OST preallocated code
2824 Details : OST preallocation code (5137) didn't take echo_client IO path
2825 into account: echo_client calls filter methods outside of any
2826 OST thread and, hence, there is no per-thread preallocated
2827 pages and buffers to use. Solution: hijack pga pages for IO. As
2828 a byproduct, this avoids unnecessary data copying.
2832 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2833 Description: Client can oops in mdc_commit_close() after open replay
2834 Details : It was possible for the MDS to return an open request with no
2835 transaction number in mds_finish_transno() if the client was
2836 evicted, but without actually returning an error. Clients
2837 would later try to replay that open and may trip an assertion
2838 Simplify the client close codepath, and always return an error
2839 from the MDS in case the open is not successful.
2842 Frequency : rare, 2.6 OSTs only
2844 Description: OST may deadlock under high load on fragmented files
2845 Details : If there was a heavy load and highly-fragmented OST filesystems
2846 it was possible to have all the OST threads deadlock waiting on
2847 allocation of biovecs, because the biovecs were not released
2848 until the entire RPC IO was completed. Instead, release biovecs
2849 as soon as they are complete to ensure forward IO progress.
2851 Severity : enhancement
2853 Description: Support for specifying external journal device at mount
2854 Details : If an OST or MDS device is formatted with an external journal
2855 device, this device major/minor is stored in the ext3 superblock
2856 and may not be valid for failover. Allow detecting and
2857 specifying the external journal at mount time.
2862 Description: Mounting an MDS with pending unlinked files may cause oops
2863 Details : target_finish_recovery() calls mds_postrecov() which returned
2864 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2865 considers this an error and immediately begins cleaning up the
2866 lov, just after starting the mds_lov process
2868 Severity : enhancement
2870 Description: Implement 'lfs df' to report actual free space on per-OST basis
2871 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2872 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2873 usage of objects; '-h' to report in human readable format.
2875 ------------------------------------------------------------------------------
2877 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2884 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2885 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2886 at startup during orphan unlinking. Instead, skip buddy update
2887 before setup, as it will be regenerated after recovery is complete.
2892 Description: If an OST is inactive, its locks might reference stale inodes.
2893 Details : lov_change_cbdata() must iterate over all namespaces, even if
2894 they are inactive to clear inode references from the lock.
2896 Severity : enhancement
2897 Frequency : occasional, if non-standard max_dirty_mb used
2899 Description: Client will block write RPCs if not enough grant
2900 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2901 then the client will block writes while waiting for another RPC
2902 to complete instead of consuming its dirty limit. With change
2903 we get improved performance when max_dirty_mb is small.
2905 Severity : enhancement
2906 Bugzilla : 3389, 6253
2907 Description: Add support for supplementary groups on the MDS.
2908 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2909 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2910 lookups for user supplementary groups into a cache.
2914 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2915 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2916 for liblustre O_EXCL is enforced by MDS.
2920 Description: readdir never returns NULL in liblustre.
2921 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2922 dentry in struct dirent.
2926 Frequency : liblustre only, depends on application IO pattern
2927 Description: liblustre clients evicted if not contacting servers
2928 Details : Don't put liblustre clients into the ping_evictor list, so
2929 they will not be evicted by the pinger ever.
2931 Severity : enhancement
2933 Description: Add ability to evict clients by NID from MDS.
2934 Details : By echoing "nid:$NID" string into
2935 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2936 $NID would be instantly evicted from this MDS and from all active
2937 OSTs connected to it.
2941 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2942 Details : lookup_it_finish() used to query file size from OSTs that was not
2947 Description: service threads change working directory to that of init
2948 Details : Starting lustre service threads may pin the working directory
2949 of the parent thread, making that filesystem busy. Threads
2950 now change to the working directory of init to avoid this.
2954 Frequency : during shutdown only
2955 Description: shutdown with a failed MDS or OST can cause unmount to hang
2956 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2961 Frequency : 2.6 only
2962 Description: chmod/chown may include an extra supplementary group
2963 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2964 supplementary group and if none is specified this is used.
2967 Bugzilla : 5479 (6816)
2969 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2970 Details : If lookup is in progress on a file that is unlinked we might try
2971 to revalidate the inode and fail in revalidate after lookup is
2972 complete and ll_file_open() enqueues the open again but
2973 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2976 Frequency : always, if lconf --abort_recovery used
2978 Description: lconf --abort_recovery fails with 'Operation not supported'
2979 Details : lconf was attempting to abort recovery on the MDT device and not
2982 ------------------------------------------------------------------------------
2984 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2989 Frequency : rare (only unsupported configurations with a node running as an
2991 Bugzilla : 6514, 5137
2992 Description: Mounting a Lustre file system on a node running as an OST could
2994 Details : OSTs now preallocates memory needed to write out data at
2995 startup, instead of when needed, to avoid having to
2996 allocate memory in possibly low memory situations.
2997 Specifically, if the file system is mounted on on OST,
2998 memory pressure could force it to try to write out data,
2999 which it needed to allocate memory to do. Due to the low
3000 memory, it would be unable to do so and the node would
3001 become unresponsive.
3003 Severity : enhancement
3005 Description: Addition of lconf --service command line option
3006 Details : lconf now accepts a '--service <arg>' option, which is
3007 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3009 Severity : enhancement
3011 Description: Failover mode is now the default for OSTs.
3012 Details : By default, OSTs will now run in failover mode. To return to
3013 the old behaviour, add '--failout' to the lmc line for OSTs.
3015 Severity : enhancement
3017 Description: Health checks are now provided for MDS and OSTs
3018 Details : Additional detailed health check information on MSD and OSTs
3019 is now provided through the procfs health_check value.
3022 Frequency : occasional, depends on IO load
3024 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3025 after numerous create/delete cycles
3026 Details : The ext3 inode allocation policy would not allocate new inodes
3027 very well on the OSTs because there are no new directories
3028 being created. Instead we look for groups with free space if
3029 the parent directories are nearly full.
3034 Description: Network or server problems during mount may cause partially
3035 mounted clients instead of returning an error.
3036 Details : The config llog parsing code may overwrite the error return
3037 code during mount error handling, returning success instead
3043 Description: MDS can fail to allocate large reply buffers
3044 Details : After long uptimes the MDS can fail to allocate large reply
3045 buffers (e.g. zconf client mount config records) due to memory
3046 fragmentation or consumption by the buffer cache. Preallocate
3047 some large reply buffers so that these replies can be sent even
3048 under memory pressure.
3052 Frequency : rare (liblustre)
3053 Description: fsx running with liblustre complained that using truncate() to
3054 extend the file doesn't work. This patch corrects that issue.
3055 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3056 ATTR_SIZE and lsm use in llu_setattr_raw.
3060 Frequency : rare, only 2.6 kernels
3061 Description: Unusual file access patterns on the MDS may result in inode
3062 data being lost in very rare circumstances.
3063 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3064 code in the 2.6 kernel caused the inode and/or EA data not to
3065 be read from disk, causing single-file corruption.
3069 Frequency : rare, only 2.6 filesystems using extents
3070 Description: Heavy concurrent write and delete load may cause data corruption.
3071 Details : It was possible under high-load situations to have an extent
3072 metadata block in the block device cache from a just-unlinked
3073 file overwrite a newly-allocated data block. We now unmap any
3074 metadata buffers that alias just-allocated data blocks.
3078 Frequency : filesystems with default stripe_count larger than 77
3079 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3080 Details : lconf specifies an inode size of 4096 bytes when the default
3081 stripe_count is larger than 77. This conflicts with the default
3082 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3083 this case to avoid pinning too much memory for large EAs.
3085 ------------------------------------------------------------------------------
3087 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3092 Frequency : rare (extremely heavy IO load with hundreds of clients)
3094 Description: Client is evicted, gets IO error writing to file
3095 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3096 caused clients to be evicted for AST timeouts. The fixes in
3097 bug 5192 mean we no longer need to have such short AST timeouts
3098 so ldlm_timeout has been increased.
3101 Frequency : occasional during --force or --failover shutdown under load
3102 Bugzilla : 5949, 4834
3103 Description: Server oops/LBUG if stopped with --force or --failover under load
3104 Details : a collection of import/export refcount and cleanup ordering
3105 issues fixed for safer force cleanup
3108 Frequency : only filesystems larger than 120 OSTs
3109 Bugzilla : 5990, 6223
3110 Description: lfs getstripe would oops on a very large filesystem
3111 Details : lov_getconfig used kfree on vmalloc'd memory
3114 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3115 Bugzilla : 6242, 6243
3116 Description: reading from files that had been truncated to a non-zero size
3117 but never opened returned no data
3118 Details : ll_file_read() reads zeros from no-object files to EOF
3123 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3124 Details : There's little harm in aborting MDS/OSS recovery and letting it
3125 try again, so I removed the LASSERT and return an error instead.
3127 Severity : enhancement
3129 Description: New debugging infrastructure for tracking down data corruption
3130 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3131 (b) cover more of the client-side code path, and (c) try to narrow
3132 down where problems occurred
3136 Bugzilla : 3819, 4364, 4397, 6313
3137 Description: Racing close and eviction MDS could cause assertion in mds_close
3138 Details : It was possible to get multiple mfd references during close and
3139 client eviction, leading to one thread referencing a freed mfd.
3141 Severity: : enhancement
3142 Bugzilla : 3262, 6359
3143 Description: Attempts to reconnect to servers are now more aggressive.
3144 Details : This builds on the enhanced upcall-less recovery that was added
3145 in 1.4.2. When trying to reconnect to servers, clients will
3146 now try each server in the failover group every 10 seconds. By
3147 default, clients would previously try one server every 25 seconds.
3152 Description: After recovery, certain operations trigger a failed
3153 assertion on a client.
3154 Details : Failing over an mds, using lconf -d --failover, while a
3155 client was doing a readdir() call would cause the client to
3156 LBUG after recovery completed and the readdir() was resent.
3158 Severity : enhancement
3160 Description: Default groups are now added by lconf
3161 Details : You can now run lconf --group <servicename> without having to
3162 manually add groups with lmc.
3165 Frequency : occasional
3167 Description: Nodes with an elan id of 0 trigger a failed assertion
3170 Frequency : always when accessing e.g. tty/console device nodes
3172 Description: tty and some other devices nodes cannot be used on lustre
3173 Details : file's private_data field is used by device data and lustre
3174 values in there got lost. New field was added to struct file to
3175 store fs-specific private data.
3178 Frequency : when exporting Lustre via NFS
3180 Description: NFSD failed occasionally when looking up a path component
3181 Details : NFSD is looking up ".." which was broken in ext3 directories
3182 that had grown large enough to become hashed.
3185 Frequency : Clusters with multiple interfaces not on the same subnet
3187 Description: Nodes will repeatedly try to reconnect to an interface which it
3188 cannot reach and report an error to the log.
3189 Details : Extra peer list entries will be created by lconf with some peers
3190 unreachable. lconf now validates the peer before adding it.
3193 Frequency : Only if a default stripe is set on the filesystem root.
3195 Description: Setting a default stripe on the filesystem root prevented the
3196 filesystem from being remounted.
3197 Details : The client was sending extra request flags in the root getattr
3198 request and did not allocate a reply buffer for the dir EA.
3201 Frequency : occasional, higher if lots of files are accessed by one client
3202 Bugzilla : 6159, 6097
3203 Description: Client trips assertion regarding lsm mismatch/magic
3204 Details : While revalidating inodes the VFS looks up inodes with ifind()
3205 and in rare cases can find an inode that is being freed.
3206 The ll_test_inode() code will free the lsm during ifind()
3207 when it finds an existing inode and then the VFS later attaches
3208 this free lsm to a new inode.
3212 Bugzilla : 6422, 7030
3213 Description: MDS deadlock between mkdir and client eviction
3214 Details : Creating a new file via mkdir or mknod (starting a transaction
3215 and getting the ns lock) can deadlock with client eviction
3216 (gets ns lock and trying to finish a synchronous transaction).
3219 Frequency : occasional
3220 Description: While starting a server, the fsfilt_ext3 module could not be
3222 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3223 kernels. Previously, lconf would still use the ext3 name
3224 when trying to load modules. Now, it will correctly use
3225 ext3 on 2.4 and ldiskfs on 2.6.
3227 Severity : enhancement
3228 Description: The default stripe count has been changed to 1
3229 Details : The interpretation of the default stripe count (0, to lfs
3230 or lmc) has been changed to mean striping across a single
3231 OST, rather than all available. For general usage we have
3232 found a stripe count of 1 or 2 works best.
3234 Severity : enhancement
3235 Description: Add support for compiling against Cray portals.
3236 Details : Conditional compiling for some areas that are different
3240 Frequency : occasional
3241 Bugzilla : 6409, 6834
3242 Description: Creating files with an explicit stripe count may lead to
3243 a failed assertion on the MDS
3244 Details : If some OSTs are full or unavailable, creating files may
3245 trigger a failed assertion on the MDS. Now, Lustre will
3246 try to use other servers or return an error to the
3250 Frequency : occasional
3252 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3253 caused lock timeout during readahead (since 1.4.2).
3254 Details : Processes doing readahead might match a lock that hasn't been
3255 granted yet if there are overlapping and conflicting lock
3256 requests. The readahead process waits on ungranted lock
3257 (original lock is CBPENDING), while OST waits for that process
3258 to cancel CBPENDING read lock and eventually evicts client.
3260 Severity : enhancement
3262 Description: Initial enabling of flock support for clients
3263 Details : Implements fcntl advisory locking and file status functions.
3264 This feature is provided as an optional mount flag (default
3265 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3266 locking are implemented yet, and those that are are not guaranteed
3267 to be completely correct in production environments.
3268 mount -t lustre -o [flock|noflock] ...
3271 Frequency : occasional
3273 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3274 assertion in the ext3 JBD (journaling) layer.
3275 Details : The b_committed_data struct is protected by the big kernel lock
3276 in 2.4 kernels, serializing journal_commit_transaction() and
3277 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3278 access to this struct. In 2.6 kernels there is finer grained
3279 locking to improve SMP performance of the JBD layer.
3283 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3288 Description: merge and cleanup kernel patches.
3291 -----------------------------------------------------------------------------
3293 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3295 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3296 versions of Lustre. Please update all servers and clients to
3297 version 1.4.2 or later at the same time. You must also run
3298 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3299 to update the configuration logs.
3301 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3302 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3303 - swab llog records of type '0' so we get proper header size/idx (5861)
3304 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3305 - fix rename of one directory over another leaking an inode (5953)
3306 - avoid SetPageDirty on 2.6 (5981)
3307 - don't re-add just-being-destroyed locks to the waiting list (5653)
3308 - when creating new directories, inherit the parent's custom
3309 striping settings if present parent (3048)
3310 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3311 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3312 - don't hold client locks on temporary worklist from l_lru (5666)
3313 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3314 - automatically evict dead clients (5921)
3315 - Update file size properly in create+truncate+fstat case (6196)
3316 - Do not unhash mountpoint dentries, do not allow removal of
3318 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3319 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3320 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3321 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3322 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3323 - Correctly handle reads of files with no objects (6243)
3324 - lctl recover will also mark a device active if deactivate used (5933)
3326 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3327 - allow --write-conf on an MDS with different nettype than client (5619)
3328 - don't write config llogs to MDS for mounts not from that MDS (5617)
3329 - lconf should create multiple TCP connections from a client (5201)
3330 - init scripts are now turned off by default; run chkconfig --on
3331 lustre and chkconfig --on lustrefs to use them
3332 - upcalls are no longer needed for clients to recover to failover
3334 - add --abort-recovery option to lconf to abort recovery on device
3336 - add support for an arbitrary number of OSTs (3026)
3337 - Quota support protocol changes.
3338 - forward compatibility changes to wire structs (6007)
3339 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3340 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3341 - improved "lustre status" script
3342 - initialize blocksize for non-regular files (6062)
3343 - added --disable-server and --disable-client configure options (5782)
3344 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3345 - Vanilla 2.4.29 support
3346 - increase maximum number of obd devices to 520 (6242)
3347 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3348 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3349 SLES 9 2.6.5 kernels
3350 - limit stripes per file to 160 (the maximum EA size) (6093)
3352 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3355 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3356 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3357 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3358 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3359 - small ext3 extents cleanups and fixes (5733)
3360 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3361 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3362 - avoid needless client->OST connect, fix handle mismatch (5317)
3363 - fix DLM error path that led to out-of-sync client, long delays (5779)
3364 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3365 - fix several locking issues related to i_size (5492,5624,5654,5672)
3366 - don't move pending lock onto export if it is already evicted (5683)
3367 - fix kernel oops when creating .foo in unlinked directory (5548)
3368 - fix deadlock in obdfilter statistics vs. object create (5811)
3369 - use time_{before,after} to avoid timer jiffies wrap (5882)
3370 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3371 - Do not leak request if server was not able to process it (5154)
3372 - If mds_open unable to find parent dir, make that negative lookup(5154)
3373 - don't create new directories with extent-mapping (5909, 5936)
3375 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3376 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3377 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3378 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3379 - support for RHEL4 kernel on i686 (5773)
3380 - provide error messages when incompatible logs are encountered (5898)
3382 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3383 * version 1.4.0.10 (1.4.1 release candidate 1)
3385 - don't keep a lock reference when lock is not granted (4238)
3386 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3387 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3388 - drop import inflight refcount on signal_completed_replay error (5255)
3389 - unlock page after async write error during send (3677)
3390 - handle missing objects in filter_preprw_read properly (5265)
3391 - no transno return for symlink open, don't save no-trasno open (3440)
3392 - don't try to complete elan receive that already failed (4012)
3393 - free RPC server reply state on error (5406)
3394 - clean up thread from ptlrpc_start_thread() on error (5160)
3395 - readahead could read extra page into cache that wasn't ejected (5388)
3396 - prevent races in class_attach/setup/cleanup/detach (5260)
3397 - don't dereference de->d_inode after l_dput of de (5458)
3398 - use "int" for stripe value returned from lock_to_stripe (5544)
3399 - mballoc allocation and error-checking fixes in 2.6 (5504)
3400 - block device patches to fix I/O request sizes in 2.6 (5482)
3401 - look up hostnames for IB nals (5602)
3402 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3403 - don't start multiple acceptors for the same port (5277)
3404 - fix incorrect LASSERT in mds_getattr_name (5635)
3405 - export a proc file for general "ping" checking (5628)
3406 - fix "lfs check" to not block when the MDS is down (5628)
3408 - service request history (4965)
3409 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3410 - create an "evict_client" /proc entry on OSTs, like the MDS has
3411 - fix mount usage message, return errors per mount(8) (5168)
3412 - change grep [] to grep "[]" in tests so they work in more UMLs
3413 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3414 - remove ancient LOV_MAGIC_V0 EA support (5047)
3415 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3416 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3417 - allow readahead allocations to fail when low on memory (5383)
3418 - mmap locking landed again, after considerable improvement (2828)
3419 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3421 2004-11-23 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 - flock/lockf fixes (but it's still disabled, pending 5135)
3429 - don't use EXT3 constants in llite code (5094)
3430 - memory shortage at startup could cause assertion (5176)
3432 - reorganization of lov code
3433 - single portals codebase
3435 - add extents/mballoc support (5025)
3436 - direct I/O reads in the obdfilter (4048)
3437 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3439 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3442 - send OST transaction number in read/write reply to free req (4966)
3443 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3444 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3445 - the watchdog thread now runs as interruptible (5246)
3446 - handle missing objects in filter_preprw_read properly (5265)
3447 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3448 - drop import inflight refcount on signal_completed_replay error (5255)
3449 - unlock page after async write error during send (3677)
3450 - return original error code on reconstructed replies (3761)
3451 - no transno return for symlink open, don't save no-trasno open (3440)
3453 - add pid to ldlm debugging output (4922)
3454 - bump the watchdog timeouts -- we can't handle 30sec yet
3455 - extra debugging for orphan dentry/inode bug (5259)
3457 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3460 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3461 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3462 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3463 - fix NULL dereference in /proc/sys/portals/routes (4827)
3464 - allow failed mdc_close() operations to be interrupted (4561)
3465 - stop precreate on OST before MDS would time out on it (4778)
3466 - don't send partial-page writes before EOF from client (4410)
3467 - discard client grant for sub-page writes on large-page clients (4520)
3468 - don't free dentries not owned by NFS code, check generation (4806)
3469 - fix lsm leak if mds_create_objects() fails (4801)
3470 - limit debug_daemon file size, always print CERROR messages (4789)
3471 - use transno after validating reply (3892)
3472 - process timed out requests if import state changes (3754)
3473 - update mtime on OST during writes, return in glimpse (4829)
3474 - add mkfsoptions to LDAP (4679)
3475 - use ->max_readahead method instead of zapping global ra (5039)
3476 - don't interrupt __l_wait_event() during strace
3478 - add software watchdogs to catch hung threads quickly (4941)
3479 - make lustrefs init script start after nfs is mounted
3480 - fix CWARN/ERROR duplication (4930)
3481 - return async write errors to application if possible (2248)
3482 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3483 - print NAL number in %x format (4645)
3484 - update barely-supported suse-2.4.21-171 series (4842)
3485 - support for sles 9 %post scripts
3486 - support for building 2.6 kernel-source packages
3487 - support for sles km_* packages
3489 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3492 - ignore -ENOENT errors in osc_destroy (3639)
3493 - notify osc create thread that OSC is being cleaned up (4600)
3494 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3495 - reconstruct ost_handle() like mds_handle() (4657)
3496 - create a new thread to do import eviction to avoid deadlock (3969)
3497 - let lconf resolve symlinked-to devices (4629)
3498 - don't unlink "objects" from directory with default EA (4554)
3499 - hold socknal file ref over connect in case target is down (4394)
3500 - allow more than 32000 subdirectories in a single directory (3244)
3501 - fix blocks count for O_DIRECT writes (3751)
3502 - OST returns ENOSPC from object create when no space left (4539)
3503 - don't send truncate RPC if file size isn't changing (4410)
3504 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3505 - bind to privileged port in socknal and tcpnal (3689)
3507 - rate limit CERROR/CWARN console message to avoid overload (4519)
3508 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3509 - basic mmap support (3918)
3510 - kernel patch series update from b1_4 (4711)
3512 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3515 - avoid crash during MDS cleanup with OST shut down (2775)
3516 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3517 - don't use bad inodes on the MDS (3744)
3518 - dynamic object preallocation to improve recovery speed (4236)
3519 - don't hold spinlock over lock dumping or change debug flags (4401)
3520 - don't zero obd_dev when it is force cleaned (3651)
3521 - print grants to console if they go negative (4431)
3522 - "lctl deactivate" will stop automatic recovery attempts (3406)
3523 - look for existing locks in ldlm_handle_enqueue() (3764)
3524 - don't resolve lock handle twice in recovery avoiding race (4401)
3525 - revalidate should check working dir is a directory (4134)
3527 - don't always mark "slow" obdfilter messages as errors (4418)
3529 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3532 - don't close LustreDB during write_conf until it is done (3860)
3533 - fix typo in lconf for_each_profile (3821)
3534 - allow dumping logs from multiple threads at one time (3820)
3535 - don't allow multiple threads in OSC recovery (3812)
3536 - fix debug_size parameters (3864)
3537 - fix mds_postrecov to initialize import for llog ctxt (3121)
3538 - replace config semaphore with spinlock (3306)
3539 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3540 - don't allow enqueue to complete on a destroyed export (3822)
3541 - down write_lock before checking llog header bitmap (3825)
3542 - recover from lock replay timeout (3764)
3543 - up llog sem before sending rpc (3652)
3544 - reduce ns lock hold times when setting kms (3267)
3545 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3546 - fix NULL deref and obd_dev leak on setup error (3312)
3547 - replace some LBUG about llog ops with error handling (3841)
3548 - don't match INVALID dentries from d_lookup and spin (3784)
3549 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3550 - fix invalid assertion in ptlrpc_set_wait (3880)
3552 - add libwrap support for the TCP acceptor (3996)
3553 - add /proc/sys/portals/routes for non-root route listing (3994)
3554 - allow setting MDS UUID in .xml (2580)
3555 - print the stack of a process that LBUGs (4228)
3557 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3560 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3561 - make sure to unset replay flag from failed open requests (3440)
3562 - if default stripe count is 0, use OST count for inode size (3636)
3563 - update parent mtime/ctime on client for create/unlink (2611)
3564 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3565 - free recovery state on server during a forced cleanup (3571)
3566 - unregister_reply for resent reqs (3063)
3567 - loop back devices mounting and status check on 2.6 (3563)
3568 - fix resource-creation race that can provoke i_size == 0 (3513)
3569 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3570 - more debugging for page-accounting assertion (3746)
3571 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3572 - avoid deadlock after precreation failure (3758)
3573 - fix race and lock order deadlock in orphan handling (3450, 3750)
3574 - add validity checks when grabbing inodes from l_ast_data (3599)
3576 - add /proc/.../recovery_status to obdfilter (3428)
3577 - lightweight CDEBUG infrastructure, debug daemon (3668)
3578 - change default OSC RPC parameters to be better on small clusters
3579 - turn off OST read cache for files smaller than 32MB
3580 - install man pages and include them in rpms (3100)
3581 - add new init script for (un)mounting lustre filesystems (2593)
3582 - run chkconfig in %post for init scripts (3701)
3583 - drop scimac NAL (unmaintained)
3585 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3588 - clean kiobufs before and after use (3485)
3589 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3590 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3591 - change init script to fail more clearly if not run as root (1528)
3592 - allow clients to reconnect during replay (1742)
3593 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3594 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3595 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3596 - fix cond_resched (3554)
3597 - only evict unfinished clients after recovery (3515)
3598 - allow bulk resend, prevent data loss (3570)
3599 - dynamic ptlrpc request buffer allocation (2102)
3600 - don't allow unlinking open directory if it isn't empty (2904)
3601 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3602 - remove extraneous obd dereference causing LASSERT failure (3334)
3603 - don't use get_cycles() when creating temp. files on the mds (3156)
3604 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3605 - handle EEXIST for set-stripe, set proper directory name (3336)
3607 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3608 - fix ksocknal_fmb_callback() error messages (2918)
3610 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3613 - don't copy lvb into (possibly NULL) reply on error (2983)
3614 - don't deref dentry after dput, don't free lvb on error (2922)
3615 - use the kms to determine writeback rpc length (2947)
3616 - increment oti_logcookies when osc is inactive (2948)
3617 - update client's i_blocks count via lvb messages (2543)
3618 - handle intent open/close of special files properly (1557)
3619 - mount MDS with errors=remount-ro, like obdfilter (2009)
3620 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3621 - don't use cancelling-locks' kms values (2947)
3622 - use highest lock extent for kms, not last one (2925)
3623 - don't dereference ERR_PTR() dentry in error handling path (3107)
3624 - fix thread race in portals_debug_dumplog() (3122)
3625 - create lprocfs device entries at setup instead of at attach (1519)
3626 - common AST error handler, don't evict client on completion race (3145)
3627 - zero nameidata in detach_mnt in 2.6 (3118)
3628 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3629 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3630 - initialize RPC timeout timer earlier for 2.6 (3219)
3631 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3632 - print nal/nid for unknown nid (3258)
3633 - additional checks for oscc recovery before doing precreate (3284)
3634 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3635 - don't crash in mdc_close for bad permissions on open (3285)
3636 - zero i_rdev for non-device files (3147)
3637 - clear page->private before handing to FS, better assertion (3119)
3638 - tune the read pipeline (3236)
3639 - fix incorrect decref of invalidated dentry (2350)
3640 - provide read-ahead stats and refine rpc in flight stats (3328)
3641 - don't hold journal transaction open across create RPC (3313)
3642 - update atime on MDS at close time (3265)
3643 - close LDAP connection when recovering to avoid server load (3315)
3644 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3645 - don't leak open file on MDS after open resend (3325)
3646 - serialize filter_precreate and filter_destroy_precreated (3329)
3647 - loop device shouldn't call sync_dev() for nul device (3092)
3648 - clear page cache after eviction (2766)
3649 - resynchronize MDS->OST in background (2824)
3650 - refuse to mount the same filesystem twice on same mountpoint (3394)
3651 - allow llmount to create routes for mounting behind routers (3320)
3652 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3653 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3654 - fix rare problem with rename on htree directories (3417)
3656 - allow default OST striping configuration per directory (1414)
3657 - fix compilation for qswnal for 2.6 kernels (3125)
3658 - increase maximum number of MDS request buffers for large systems
3659 - change liblustreapi to be useful for external progs like lfsck (3098)
3660 - increase local configuration timeout for slow disks (3353)
3661 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3663 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3666 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3667 - don't overwrite extent policy data in reply if lock was blocked (2901)
3668 - drop filter export grants atomically with removal from device (2663)
3669 - del obd_self_export from work_list in class_disconnect_exports (2908)
3670 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3671 - swab reply message in mdc_close, other PPC fixes (2464)
3672 - fix destroying of named logs (2325)
3673 - overwrite old logs when running lconf --write_conf (2264)
3674 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3675 - fix race in target_handle_connect (2898)
3676 - mds_reint_create() should take same inode create lock (2926)
3677 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3678 - don't close files for self_export to avoid uninitialized obd (2936)
3679 - allow MDS with the same name as client node (2939)
3680 - hold dentry reference for closed log files for unlink (2325)
3681 - reserve space for all logs during transactions (2059)
3682 - don't evict page beyond end of stripe extent (2925)
3683 - don't oops on a deleted current working directory (2399)
3684 - handle hard links to targets without a parent properly (2517)
3685 - don't dereference NULL lock when racing during eviction (2867)
3686 - don't grow lock extents when lots of conflicting locks (2919)
3688 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3691 - account for cache space usage on clients to avoid data loss (974)
3692 - lfsck support in lustre kernel code (2349)
3693 - reduce journal credits needed for BRW writes (2370)
3694 - orphan handling to avoid losing space on client/server crashes
3695 - ptlrpcd can be blocked, stopping ALL progress (2477)
3696 - use lock value blocks to assist in proper KMS, faster stat (1021)
3697 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3698 - recovery for initial connections (2355)
3699 - fixes for mds_cleanup_orphans (1934)
3700 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3701 - block all file creations until orphan recovery completes (1901)
3702 - client remove rq_connection from request struct (2423)
3703 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3704 - recovery timer race (2670)
3705 - mdc_close recovey bug (2532)
3706 - ptlrpc cleanup bug (2710)
3707 - mds timeout on local locks (2588)
3708 - namespace lock held during RPCs (2431)
3709 - handle interrupted sync write properly (2503)
3710 - don't try to handle a message that hasn't been replied to (2699)
3711 - client assert failure during cleanup after abort recovery (2701)
3712 - leak mdc device after failed mount (2712)
3713 - ptlrpc_check_set allows timedout requests to complete (2714)
3714 - wait for inflight reqs when ptlrpcd finishes (2710)
3715 - make sure unregistered services are removed from the srv_list
3716 - reset bulk XID's when resending them (caught by 1138 test)
3717 - unregister_bulk after timeout
3718 - fix lconf error (2694)
3719 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3720 - readahead locks pages, leaves pending causing memory pressure (2673)
3721 - increase OST request buffers to 4096 on large machines (2729)
3722 - fix up permission of existing directories in simple_mkdir (2661)
3723 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3724 - don't assign transno to errored transactions (2742)
3725 - don't delete objects on OST if given a bogus objid from MDS (2751)
3726 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3727 - if rq_no_resend, then timeout request after recovery (2432)
3728 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3729 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3730 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3731 - MDS recovery shouldn't delete objects that it has given out (2730)
3732 - if enqueue arrives after completion, don't clobber LVB (2819)
3733 - don't unlock pages twice when trigger_group_io returns error (2814)
3734 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3735 - don't write pages to disk if there was an error (1450)
3736 - don't ping imports that have recovery disabled (2676)
3737 - take buffered bytes into account when balancing socknal conn (2817)
3738 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3739 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3740 - remove little-endian swabbing of llog records (1987)
3741 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3742 - retry reposting request buffers if they fail (1191)
3743 - grow extent at grant time to avoid granting a revoked lock (2809)
3744 - lock revoke doesn't evict page if covered by a second lock (2765)
3745 - disable VM readahead to avoid reading outside lock extents (2805)
3747 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3748 - updated kernel patches for hp-2.4.20 kernel (2681)
3750 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3753 - fix truncated write corruption (2366)
3754 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3756 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3757 - ensure len(uuid) < 37 in lmc (1171)
3758 - fix ia64 OOPS in llog_test (2255)
3759 - zero end of page at obdfilter for partial page writes (2648)
3760 - don't leave stale dentries around after renames (bug 2428)
3761 - fix timeouts when evicting a client with a single lock held (2642)
3762 - set deadline for the initial HELLO message to drain (2634)
3763 - print out dotted-quad IP addresses in the socknal (2302)
3765 - additional debugging for MDS client eviction problem (2443)
3766 - fix mkfsoptions support for osts (2603, 2604)
3768 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3771 - add series for the vanilla 2.6.0 kernel
3772 - add series for the vanilla 2.4.24 kernel
3773 - add series for a cray x86/64 UL kernel drop
3774 - fix xattr patches for the vanilla 2.4.19 series
3776 - generate true UUIDs in lmc (1171)
3777 - have portals stack dumping break in UML (2466)
3778 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3779 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3780 - wake callback waiting threads on client eviction (2460)
3781 - Add --ptldebug and --subsystem to lmc (1719)
3782 - update assertion to allow safe interrupt allocation
3783 - set rq_no_resend for cancel requests (2432)
3784 - recalculate ptlrpcd timeout after resend (2494)
3785 - call vfs_rmdir when removing pending directories (2368)
3786 - fix renaming a file to itself (2429)
3787 - lmc creates a default one-stripe lov (2454)
3788 - expand procfs space to handle large clusters (2326)
3789 - increase UML stack to avoid overflow
3790 - update lconf's list of debug and subsystem masks
3791 - fix lfs find --obd (2510)
3792 - /proc tunable for disabling filter read caching (2591)
3793 - stop rpm packages from altering slapd.conf (2301)
3794 - disable nagle in the socknal under 0conf (2578)
3795 - choose mds inode size based on stripe count (2572)
3796 - fix kernel-source rpm problems (2516)
3798 - add --disable-doc to avoid pdf generation (2421)
3799 - update documentation, tests, type-os, comments
3800 - avoid format warnings on ia64
3801 - remove the TOE NAL
3802 - tiny code cleanups by removing unused fields
3804 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3807 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3808 - remove the most blatant lies from BUILDING file (2371)
3809 - change default debug level to reasonable production setting
3810 - reduce client side cache size to reduce cache flush time
3811 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3812 - make TCP zerocopy and pinger support enabled by default (2476)
3813 - sync writes completed after process exits caused crashes (2319)
3814 - maintain correct mount count on the MDS (2356)
3815 - backout 1557, because 2316 wasn't really fixed
3816 - better file I/O statistics gathering in /proc
3817 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3818 - another kernel patch to fix zero-copy TCP function export
3819 - don't take duplicate lock when processing re-sent getattr (2420)
3820 - lctl uses obd_self_export instead of creating new conn (2353)
3821 - MDS/OST recovery case which requires object creation asserted (2425)
3822 - move lfs from /usr/sbin to /usr/bin in packages
3823 - fix race between mds_client_add and mds_client_free (2417)
3824 - use kmalloc instead of slabs in portals (2430)
3825 - don't create duplicate records when a failover MDS is present (2442)
3826 - remove unnecessary mount age check (2332)
3827 - don't remove directory inodes from locks prematurely (2451)
3828 - don't break if MDS service name is the same as hostname (2103)
3829 - fix races in client write RPC generation when cache full (2482)
3831 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3834 - remove now-unused request->rq_obd (278)
3835 - if an allocation fails, print out how much memory we've used (1933)
3836 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3837 - add the "configurable stack size" patch to most series files (1256)
3838 - ability to write large log records, for 100+ OST configs (2306)
3839 - fix NULL deref when filter_prep fails (2314)
3840 - fix operator precedence error in filter_sync
3841 - dynamic allocation of socknal TX descriptors (2315)
3842 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3843 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3844 - ensure that CWARN messages in Portals always get to the syslog
3845 - __init/__exit are not for prototype decls (ldlm_init/exit)
3846 - x86-64 compile warning fixes
3847 - fix gateway LMC keyword conflict (2318)
3848 - fix MDS lock inversions in getattr/reint paths (1844)
3849 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3850 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3851 - DLM race condition prevented some lock evictions (2328)
3852 - ENOMEM detection and retry on socknal sends (2230)
3853 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3854 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3855 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3856 - better error messages when a client is rejected during recovery (1505)
3857 - avoid cancelling locks which were never granted, after failure (2330)
3858 - fix i_sem/journal inversion in mds_client_add (2333)
3859 - fix truncate/getattr lock cycle deadlock (2334)
3860 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3861 - fix two rare exit paths which could leak an l_lock() ref (2321)
3862 - fencepost error in MDS/OST orphan recovery (2226)
3863 - make log record alignment 8 bytes (1988)
3864 - lstripe now fails when requested offset > ost_count (2237)
3865 - ensure that all kernel series have a complete list.h (1607)
3866 - fix crashes in special-file operations (2316)
3867 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3869 - allow configurable automake binary, for testing new versions
3870 - small update to the lfs documentation
3872 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3874 * fix negative export reference count in fsfilt_sync (2312)
3876 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3877 * release candidate 0.9.1
3879 - orphans are moved into the PENDING directory for possible recovery
3880 - replayed opens now open by fid for orphan/rename safety (1042)
3881 - last close of an orphan inode generates a transno (683)
3882 - chdir() and mount() now pin the directory entry (1020)
3883 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3884 - discard very old requests without processing them (1502)
3885 - remove some common, well-understood CERRORs (1505)
3886 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3887 - clear "grant" flags in OST replies until OST grant code lands (1644)
3888 - fix read performance by not clobbering i_blksize on client (1598)
3889 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3890 - make LRU size a /proc tunable, clears locks when reduced (707)
3891 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3892 - server locks take references on exports now (1558)
3893 - build fixes for 2.4.20-rh trees (1663)
3894 - return an error from lov_create if all OSCs are inactive (1751)
3895 - fix import levels when a reconnect happens without a timeout (1597)
3896 - exit early from mds_open if we get a lookup error (1749)
3897 - partial page read at EOF wouldn't wait for disk before sending (1642)
3898 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3899 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3900 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3901 - write/truncate lock inversion (1639)
3902 - Don't auto-load obdclass, portals modules during cleanup (1495)
3903 - fix timestamps from jumping to "now" (1763)
3904 - extra journal assertions (1648)
3905 - add an extra multiunlink test (1771)
3906 - fix read_record/write_record API (1776)
3907 - fix leak of offset_extent, possible incorrect i_size later (1772)
3908 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3909 - Debugging for the kqswnal_get_idle_tx problems (1820)
3910 - Allow recovery to be attempted multiple times (1536)
3911 - Write out MDS last_rcvd file after it is first created (1600)
3912 - Fix tx_descriptor leak in failed transmit situations (1827)
3913 - ext3 journaling fixes for assertion failure after IO error (1871)
3914 - class_export_put() on freed export after completion AST error (1896)
3915 - Fix revalidate looping in VFS (1322)
3916 - Don't access a freed export during MDS_REINT timeout (1521)
3917 - Add open-unlink recovery support on the MDS (1673,1764)
3918 - Return an error if no MDS data was read from last_rcvd (1946)
3919 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3920 - Don't setup a disk device that doesn't match exported UUID (317)
3921 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3922 - avoid committing NULL handle in force close
3923 - local.sh is now a one-stripe LOV configuration
3924 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3925 - fix ext3 htree duplicate directory entry corruption (1516)
3926 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3927 - update to new LOV EA format (2097)
3928 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3930 2003-06-15 Phil Schwan <phil@clusterfs.com>
3933 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3934 - per-import/export recovery handling (958, 931, 959)
3935 - multiple last-rcvd slots, for serving multiple FSes (949)
3936 - connections are again shared between multiple imp/exports (963, 964)
3937 - "umount -f" would hang if any requests needed to be sent (393, 978)
3938 - avoid pinning large req buffer by copying for queued messages (989)
3939 - add "uuid" to "lctl device" command to help upcalls (991)
3940 - "open" RPCs with transnos would confuse recovery counters (1037)
3941 - do proper endian conversion of all wire messages (288, 340, 891)
3942 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3943 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3944 - fix ptlrpc_connection leak in target_handle_connect (1174)
3945 - fix import refcounting bug in OST and MDS cleanup (1134)
3946 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3947 - fix very unlikely obd_types race condition (501)
3948 - remove osc_open hack for echo_client (1187)
3949 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3950 - a failure in read_inode2 leads to deadlock (1139)
3951 - cancel ack-locks as soon as transaction is committed (1072)
3952 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3953 - make sure to commitrw after any preprw to avoid deadlock (1162)
3954 - failing to execute a file in a lustre FS would lock inode (1203)
3955 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3956 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3957 - obd_uuid2tgt would walk off the end of the list (1255)
3958 - on IA64 the osc would give portals incorrect bulk size (1258)
3959 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3960 - fix lock inversion caused by new llite matching code (1282)
3961 - limit the number of dirty pages on a client to 10MB (1286)
3962 - timed out locks were not being corrected cancelled (1289)
3963 - fix O_DIRECT above 4GB on IA-32 (1292)
3964 * major user-visible changes
3965 - fail out/fail over policy now controlled by the upcall (993)
3967 - add OBD_PING to check server availability and failure (954)
3968 - lustre messages are now sent in sending host order (288, 340, 891)
3969 - add eadatalen to MDS getattr reply (340)
3970 - OST read replies may contain second buffer, with per-page status (593)
3972 2003-03-11 Phil Schwan <phil@clusterfs.com>
3975 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3976 - failed open()s could cause deadlock; fixed (867, 869)
3977 - stop cancelling OST locks when files are closed (481)
3978 - overlapping XID spaces caused network corruption (851, 853)
3979 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3980 - setattr_raw added, to do single-RPC, server-side setattrs
3981 - lmc/lconf syntax change for OST UUIDs
3982 - fix crashy race condition between ptlrpc_free_req and osc_close
3983 - don't use request in mdc_enqueue if we hit a timeout (889)
3984 - don't set the inode i_size for regular files from the MDS (896)
3985 - handle out of order completion AST (842)
3986 - don't LBUG if a lock request times out after receiving AST (913)
3987 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3988 - if a bad lock AST arrives, send an error instead of dropping entirely
3989 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3990 - fix leak in bulk IO when only partially completed (899, 900, 926)
3991 - fix O_DIRECT for ia64 (55)
3992 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3993 - C-z after timeout could hang a process forever; fixed (977)
3995 - client-side I/O cache (678, 924, 929, 941, 970)
3997 - READPAGE and SETATTRs which don't take server-side locks get
4000 2003-02-11 Phil Schwan <phil@clusterfs.com>
4003 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4004 - Fix fsfilt_extN_readpage() to read a full page of directory
4005 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4006 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4007 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4008 - Most intent operations were removed, in favour of a new RPC mode
4009 that does a single RPC to the server and bypasses most of the VFS
4010 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4011 - Aggressively cancel local locks on DLM servers
4012 - mds_reint_unlink sends EA to the client if it's the last nlink.
4013 client uses that EA to unlink OST objects.
4014 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4015 - recursive symlinks were fixed (439)
4016 - fixed NULL deref in DEBUG_REQ
4017 - filter_update_lastobjid no longer calls sync, which annoyed extN
4018 - fixed multi-client small-writes to a single file problem (445)
4019 - fixed mtime updates during file writes (607)
4020 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4021 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4022 - fixed Linux OST size reporting problem (444, 656)
4023 - OST now updates object mtime with writes or setattr (607, 619)
4024 - client verifies file size before zeroing page past EOF (445)
4025 - OST now writes last allocated objid to disk with allocation (108)
4026 - LOV on echo now works (409)
4028 - mds_reint_unlink sends a new buffer, with the EA included. this
4029 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4030 set if a regular file was being unlinked, and it was the last link
4031 - use PtlGet from the target for bulk writes (315)
4032 - OST now updates object mtime with writes or setattr (607, 619)
4033 - LDLM now has a grant-time callback to revalidate locked items, if
4035 - Many MDS operations were reorganized to combat race conditions
4037 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4038 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4040 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4043 - Fully reactivate OST imports after reconnection (512, others)
4044 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4045 - More graceful error handling for truncating on dead OST (515)
4046 - Don't error out unless we're actually accessing dead stripes (474)
4047 - Fix garbage sizes when stripes are missing (410)
4048 - LRU counters were broken, causing constant lock purge (433, 432)
4049 - garbage on read from stripes with failed OSTs (441)
4050 - mark OSCs as active before reconnecting during recovery (438)
4051 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4052 - lfind did not preserve OST order in output (443)
4053 - symlinks cause hung clients, incorrect data (439)
4054 - stop dereferencing request after dropping refcount (457)
4055 - don't LASSERT(spin_is_locked) on non-SMP (455)
4056 - fixes for many rename() bugs
4057 - fstat didn't correctly synchronize attributes (399)
4058 - server must handle lock cancellation during blocking AST prep (487)
4059 - bulk descriptors were free()d too soon (511)
4060 - fix paths in lconf, which would load incorrect modules (451, 507)
4061 - fix confusing lconf 'host not found' error message (386)
4062 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4063 - fix race condition in mdc_blocking_ast() for inode access (526)
4064 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4065 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4066 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4067 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4068 - fix lstripe in several ways (kernel side) (527)
4069 - fix request leak in ldlm_cli_enqueue (262)
4070 - incorrect OSC was marked inactive after OST failure
4071 - call mds_fs_cleanup before unmounting filesystem (524)
4072 - fix races between taking ns_lock and ldlm_lock_change_resource
4073 - fix races updating LOV export open file list
4074 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4075 - fix recovery NULL deref in ldlm_cli_cancel_unused
4076 - fix some DLM races by using new hash table for lock handles (419)
4077 - permit the client to specify desired inodes, at replay
4078 - duplicate requests when we queue them for replay reintegration
4079 - fix last_rcvd offset calculation
4080 - sync after each recovered transaction, so we always make progress
4081 - never, not always, ERESTART requests without transnos
4082 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4084 - skip replay if the MDS says that the client is already connected
4085 - don't check for a recovery-enabled export to match lctl's UUID
4086 - don't INC_USE_COUNT for phantom exports
4087 - don't crash when cleaning up phantom exports (567)
4088 - don't double-finish or set replay data for errored mdc_open requests
4089 - abort requests when they time out, so we don't get old replies
4090 - send/receive replies for AST messages again
4091 - if the client says that it doesn't have the lock, cancel it on the
4093 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4094 mark it as destroyed, it will all work out when decref is called
4095 - fix module use counts (22, 581)
4097 - ASTs now expect a reply (server cancels lock on error reply)
4099 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4102 - fix many simultaneous client startup (392)
4103 - fix dentry->d_it clobbering
4104 - credentials weren't being shipped for readdir/getattr operations
4105 - remove invalid assertions triggered during some concurrent MD
4107 - proper Lustre versions added (336, 389)
4108 - fix memory leak for create error case (398)
4109 - fix LOV locking bug that would get cli/srv out of sync
4110 - fix echo client over LOV (409)
4111 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4112 - fix double-O_EXCL intent crash (424)
4113 - avoid sending multiple lock CANCELs (352)
4115 - MDS can do multi-client recovery (modulo bugs in new code)
4117 - many updates, edits, cleanups
4119 2002-11-18 Phil Schwan <phil@clusterfs.com>
4122 - fix null d_it dereference (346)
4123 - fix full OST/dbench hang (333)
4124 - fix permission problem with file removal (286)
4125 - fix removal of OSCs from LOV when they fail
4126 - fix NULL deref during bulk timeout (214)
4127 - fix problems related to multiple filesystems on one MDS (241)
4128 - fixed serious subtle metadata locking bugs
4129 - free locks on clients when inodes are removed due to memory
4131 - fix inode pointer in lock data (285)
4132 - partial support for multiple MDS on a single host (241)
4133 - data locks weren't cancelled at clear_inode time (290, 311)
4134 - intent locks could lead to unbounded lock growth (205)
4135 - added a maximum lock count, an LRU list, and a flusher
4136 - fix multiple rename (365)
4137 - properly abstracted the echo client
4138 - OSC locked 1 byte too many; fixed
4139 - rewrote brw callback code:
4140 - fixed recovery bugs related to LOVs (306)
4141 - fixed too-many-pages-in-one-write crash (191)
4142 - fixed (again) crash in sync_io_timeout (214)
4143 - probably fixed callback-related race (385)
4145 - Add capability to MDS protocol
4146 - LDLM cancellations and callbacks on different portals
4148 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4151 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4152 - defer open object destruction to close time (601981, 138)
4153 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4154 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4155 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4156 - don't call obd_setattr in ll_file_release for destroyed objects
4157 * protocol change to lustre_msg: move |version| and add |flags|
4158 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4159 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4160 * added replay of create, unlink, link and rename operations during
4161 MDS failover; recovery should be much more robust now
4162 * remove failed OSCs from LOVs (only lov_create uses this so far)
4163 * the lustre-HOWTO was brought (more) up to date (582544)
4165 2002-10-23 Phil Schwan <phil@clusterfs.com>
4168 - in-use dentries weren't being reused properly (617851)
4169 - prevent multiple LDLM setup (599178)
4170 - fix LOV size calculations for truncate (617853)
4171 - fix client handling of MDS intent errors (POSIX)
4172 - fix permission bug in lovstripe.c test (624321)
4173 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4174 - truncate past end of file could corrupt data
4175 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4176 - a race in recovery could return ETIMEDOUT to apps (623947)
4177 - building outside the source directory was fixed
4178 * the lustre-HOWTO was brought (more) up to date (582544)
4179 * major progress was made on recovery functionality
4181 2002-10-10 Phil Schwan <phil@clusterfs.com>
4184 - recovery deadlock fix
4185 - rm -rf causes LBUG fix (617817)
4186 - file open by multiple tasks fix (618962)
4187 - directory permissions bugs (602707 and 620007)
4188 - journal_stop fixed with locking (611313)
4189 - O_APPEND failures resolved (618273, perhaps 614459)
4190 - lconf PATH fix (619770)
4191 - IA64 build fix (621450)
4192 - RPC buffer sizes scale with amount of memory
4194 2002-10-01 Phil Schwan <phil@clusterfs.com>
4197 - locks would be cancelled without throwing away data pages,
4198 resulting in inconsistent data (605627)
4199 - inode attributes were not always being refreshed (605627, 612449)
4200 - lconf now continues to cleanup after lctl reports an error
4201 - MDS now enforces user permissions (602707)
4202 - lprocfs cleanup fixed, but not yet enabled (614157)
4203 - fixed infinite server hang, should a client not respond to an AST
4204 - avoid going into recovery if user calls readlink() with a buffer
4205 that's too small (613941)
4206 - AST RPCs no longer require replies (614867) -- this may be changed
4207 - don't crash server if client sends an IOV that's too big (611336)
4208 - fixed lock conversion deadlock (611892)
4209 - fixed the following of symlinks (614622)
4210 * recovery: the server can remove locks from a client that dies, other
4211 clients can make progress
4212 * more extN patch fixes
4213 * compile-time configurable ptlrpc buffer allocations
4215 - collaborative read cache document
4216 - Lustre Lite Performance CDR document-in-progress
4218 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4221 - fix typo in patch-2.4.18
4223 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4226 - clear ptlrpc request each time in handle_incoming_request()
4227 - unlink of files now destroys the object on the OST
4229 2002-09-19 Peter Braam <braam@clusterfs.com>
4231 * add hard link support
4232 * change obdfile creation method
4233 * kernel patch changed
4235 2002-09-19 Peter Braam <braam@clusterfs.com>
4238 - stack overflow bug in extN fixed
4240 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4242 * documentation updates
4243 - add man pages for config tools
4244 - update tests/README to describe testing with new config tools
4245 - finish metadata API descriptions
4246 * bug fixes and cleanups
4247 - statfs workaround for 16TB limit
4248 - LOV stripe allocation improved, can stripe on subset of OSTs
4249 - LOV file size/IO offset was wrong for files > 4GB in size
4250 - object EA data was being dropped, caused files to be unreadable
4251 - memory overflow with non-LOV OST caused memory corruption
4252 - fixed regression tests to work with new config tools, obdfilter
4253 - fixed bug when directory size became larger than 1 block
4254 - fixed bug (for single client case) when PWD was deleted
4255 - invalidate local directory pages when doing intent-based ops
4256 - avoid LDLM oops when lock callback contained bad data
4258 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4260 * documentation updates
4261 * bug fixes and cleanups
4262 - configuration tools
4265 - 64-bit compile warnings
4266 - 64-bit internal statfs data
4268 * test_brw on persistent OST devices
4270 * lprocfs (disabled)
4272 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4274 * documentation updates
4275 * bug fixes and cleanups
4276 * configuration tools
4278 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4281 * many small fixes to 0_5_4
4282 * io/network handling
4283 * thinkos in MDS operations
4285 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4288 * crucial basic fixes to 0.5.3
4289 * IOR, Iozone work over Elan
4292 2002-08-07 Phil Schwan <phil@clusterfs.com>
4293 * version 0_5_3, our first alpha
4294 * we use the new Portals iovs
4295 * documentation updates
4296 * bug fixes and cleanups
4297 * small changes in the DLM wire protocol
4299 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4300 * version 0_5_1 with some initial stability,
4301 * locking on MD and file I/O.
4302 * documentation updates
4303 * several bug fixes since 0.5.0
4304 * small changes in wire protocol
4306 2002-07-18 Phil Schwan <phil@clusterfs.com>
4308 * delivered as Lustre Light Alpha
4309 * fixed a crash after handling invalid MDS requests
4310 * fixed directory pages for architectures with non-4k pages sizes
4312 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4313 * release version v0_4_4
4314 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4316 * Fixes a number of interruption problems with OST operations.
4317 * Update documentation for portals header changes
4318 * Move all wire protocol structs/defines to lustre_idl.h
4319 * Fixes symlink length bug.
4320 * Add tcpdump to repository.
4322 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4323 * release version v0_4_3
4324 * Fixes statfs for inodes on extN.
4325 * Fixes bug in runtests which would delete /etc/hosts.
4326 * Use 64-bit object IDs wherever possible (not into VFS though)
4327 Remove ost_get_info, which is unused by lustre, and out of date.
4329 2002-07-03 Peter Braam <braam@clusterfs.com>
4330 * release version v0_4_2 Fixes a lookup error (type not passed)
4331 * move forward to head of Portals
4332 * move forward to latest Lustre kernel
4334 2002-06-25 Peter Braam <braam@clusterfs.com>
4335 * release version v0_4_1. Hopefully stable on single node use.