1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.1.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: allow userland application know is lost one of stripes.
18 Details : fill lvb_blocks with error code on ost and return it to
19 application if error flag found.
23 Description: NULL lov_tgts causing MDS oops
24 Details : more safe checks for NULL lov_tgts for avoid oops.
26 Severity : enhancement
28 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
30 Severity : enhancement
32 Description: Add rhel5 support to HEAD.
36 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
37 Details : If an OST is down the MDS will hang indefinitely in
38 obd_statfs_rqset() waiting for the statfs data. While for
39 MDS QOS usage of statfs, it should not stuck in waiting.
41 Severity : enhancement
43 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
45 Severity : enhancement
47 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
51 Description: make mgs_setparam() handle fsname containing dash
52 Details : fsname containing a dash does not work with lctl conf_param
54 Severity : enhancement
56 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
58 Severity : enhancement
60 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
64 Frequency : rare, at shutdown
65 Description: access already free / zero obd_namespace.
66 Details : if client_disconnect_export was called without force flag set,
67 and exist connect request in flight, this can produce access to
68 NULL pointer (or already free pointer) when connect_interpret
69 store ocd flags in obd_namespace.
73 Frequency : only at startup
74 Description: not alloc memory with spinlock held.
75 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
78 Severity : enhancement
80 Description: make lustre randomly fail allocating memory
81 Details : Make lustre randomly failed allocating memory for testing purpose.
83 Severity : enhancement
85 Description: lost problems with lov objid file
86 Details : Fixes some scability and access to not inited memory problems
87 in work with lov objdid file.
92 Description: lfs find does not continue on file error
93 Details : Continue other files processing when a file/dir is absent.
97 Description: Inconsistent usage of lustre_pack_reply()
98 Details : Standardize the usage of lustre_pack_reply() such that it
99 always generate a CERROR on failure.
102 Frequency : occasional
104 Description: MDS hang or stay in waiting lock
105 Details : If client receive lock with CBPENDING flag ldlm need send lock
106 cancel as separate rpc, to avoid situation when cancel request
107 can't processed due all i/o threads stay in wait lock.
111 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
115 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
119 Description: Update to SLES9 kernel-2.6.5-7.287.3.
123 Description: Update to RHEL4 latest kernel.
125 Severity : enhancement
127 Description: Build SLES10 patchless client fails
128 Details : The configure was broken by run ./configure with
129 --with-linux-obj=.... argument for patchless client. When the
130 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
131 can't search header adequately. Use absolute path such as
132 -I($LINUX)/include instead.
136 Description: interrupt oig_wait produce painc on resend.
137 Details : brw_redo_request can be used for resend requests from ptlrpcd and
138 private set, and this produce situation when rq_ptlrpcd_data not
139 copyed to new allocated request and triggered LBUG on assert
140 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
141 ptlrpcd set if request is changed and can be safety changed to use
146 Description: LASSERT_{REQ,REP}SWAB macros are buggy
147 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
148 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
149 these macros and replace them with their difinitions instead.
153 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
154 Details : Update which_patch & target file for SLES10 latest kernel.
157 Description: Remove client patches from SLES 10 kernel.
158 Details : This causes SLES 10 clients to behave as patchless clients
159 even on a Lustre-patched (server) kernel.
161 Severity : enhancement
163 Description: self-adjustable client's lru lists
164 Details : use adaptive algorithm for managing client cached locks lru
165 lists according to current server load, other client's work
166 pattern, memory activities, etc. Both, server and client
167 side namespaces provide number of proc tunables for controlling
170 Severity : enhancement
172 Description: light-weight GSS support
173 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
174 and reduce performance overhead.
176 Severity : enhancement
178 Description: Linux keyring support
179 Details : Support using service of Linux keyring for Lustre GSS internal
180 context refresh/cache mechanism.
184 Description: Fix errors in lfs documentation
185 Details : Fixes man pages
189 Description: don't use GFP_* in generic Lustre code.
190 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
194 Description: obdclass is limited by single OBD_ALLOC(idarray)
195 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
199 Description: warnings with build patchless client with vanila 2.6.19 and up
200 Details : change old ctl_table style and replace ctl_table/ctl_table_header
201 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
205 Frequency : Only for SLES
206 Description: sanity_quota fail test_1
207 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
208 file, which makes TSTID[2] inunique.
212 Frequency : Only for SLES9
213 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
214 Details : Update target/ChangeLog/which_patch .
218 Frequency : Only for test_10 in sanity-quota.sh
219 Description: a bug in quota test code
220 Details : checking a test flag in a wrong place
224 Frequency : Only for test_13 in sanity-quota.sh
225 Description: a bug in quota test script
226 Details : shouldn't check overfull allocated quota
230 Description: jbd statistics
231 Details : Port older jbd statistics patch for sles10
233 Severity : enhancement
235 Description: Add printing inode info into message about error in writepage.
240 Description: reply_lock_interpret crash due to race with it and lock cancel.
241 Details : Do not replay locks that are being cancelled. Do not reference
242 locks by their address during replay, just by their handle.
246 Frequency : When flocks are used.
247 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
248 Details : Flock locks might destroy just granted lock if it could be merged
249 with another existing flock, this is done in completion handler,
250 so teach ldlm_cli_enquque_fini that this is a valid case for
256 Description: Oops in read and write path when failing to allocate lock.
257 Details : Check if lock allocation failed and return error back.
261 Description: lstripe command fails for valid OST index
262 Details : The stripe offset is compared to lov->desc.ld_tgt_count
263 instead of lov->desc.ld_active_tgt_count.
267 Description: sanity.sh failed test 103
268 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
273 Description: df doesn't work properly if diskfs blocksize != 4K
274 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
278 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
279 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
283 Description: obd_health_check_timeout too short
284 Details : set obd_health_check_timeout as 1.5x of obd_timeout
288 Description: llapi_file_create() does not allow some changes
289 Details : add llapi_file_open() that allows specifying the mode and
290 open flags, and also returns an open file handle.
294 Description: merge and cleanup kernel patches.
295 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
299 Description: Add journal checksum support.(Kernel part)
300 Details : The journal checksum feature adds two new flags i.e
301 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
302 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
303 indicates that the commit block contains the checksum for
304 the blocks described by the descriptor blocks. Now commit
305 record can be sent to disk without waiting for descriptor
306 blocks to be written to disk. This behavior is controlled
307 using JBD2_FEATURE_ASYNC_COMMIT flag.
311 Description: OSS needs mutliple precreate threads
312 Details : Add ability to start more than one create thread per OSS.
316 Description: Sanity.sh test_65k failed: setstripe should have succeeded
317 Details : execute lfs setstripe on client
321 Description: mds_obd_create error creating tmp object
322 Details : When the user sets quota on root, llog will be affected and can't
323 create files and write files.
326 Frequency : Always on ia64 patchless client, and possibly others.
328 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
329 Details : This allows the patchless client to be loaded on architectures
334 Description: RedHat Update kernel for RHEL5
335 Details : Add the kernel config file for RHEL5.
339 Description: RedHat Update kernel for RHEL5
340 Details : Modify the kernel config file more closer RHEL5.
344 Description: Build failure against Centos5 (RHEL5)
345 Details : Define PAGE_SIZE when it isn't present.
349 Description: test_7 on sanity-quota.sh can't be used on separate machines
350 Details : add facet to handle it
354 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
355 Details : with NFS, the anon dentry's parent was set to itself in
356 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
357 corresponding dentry other than use rec->ur_name.
359 Severity : enhancement
361 Description: lfs setstripe enhancement
362 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
366 Description: enable data checksumming by default
367 Details : enable checksum by default, allow --disable-checksum
368 configure option and "-o nochecksum" mount option. Checksums
369 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
373 Description: lustre support for RHEL5
374 Details : Add support for RHEL5.
378 Description: Client eviction due to failover config
379 Details : after a connection loss, the lustre client should attempt to
380 reconnect to the last active server first before trying the
381 other potential connections.
385 Description: when mds and osts use different quota unit(32bit and 64bit),
386 quota will be released repeatly.
387 Details : void sending multiple quota reqs to mds, which will keep the
388 status between the reqs.
392 Description: osts not allocated evenly to files
393 Details : change the condition to increase offset_idx
396 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
398 Description: Data corruption for OSTs that are formatted larger than 2TB
400 Details : When generating the bio request for lustre file writes the
401 sector number would overflow a temporary variable before being
402 used for the IO. The data reads correctly from Lustre (which
403 will overflow in a similar manner) but other file data or
404 filesystem metadata may be corrupted in some cases.
408 Description: Tune the kernel for good SCSI performance.
409 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
410 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
415 Description: rewrite ext2-derived code in obdclass/uuid.c
416 Details : rewrite inherited code (uuid parsing code from ext2 utils)
417 from scratch preserving functionality.
421 Description: "lfs find -obd UUID" prints directories
422 Details : "lfs find -obd UUID" will return all directory names instead
423 of just file names. It is incorrect because the directories
424 do not reside on the OSTs.
428 Description: use i_size_read and i_size_write in 2.6 port
429 Details : replace inode->i_size access with i_size_read/write()
432 Frequency : when removing large files
434 Description: scheduling issue during removal of large Lustre files
435 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
436 It causes scheduling issues when removing large files (17TB in the
440 Frequency : only with liblustre clients on XT3
442 Description: evictions taking too long
443 Details : allow llrd to evict clients directly on OSTs
446 Frequency : only on ppc
448 Description: /proc/fs/lustre/devices broken on ppc
449 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
450 We should make sure the type of 'index' is loff_t and then cast
451 explicitly as needed below. Do not assign an explicitly cast
455 Frequency : only for rhel5
457 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
458 Details : Modify the target file & which_kernel.
460 Severity : enhancement
462 Description: omit set fsid for export NFS
463 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
464 as FS_REQUIRES_DEV for avoid problems with generate fsid.
468 Frequency : Always, for kernels after 2.6.16
469 Description: Fix warning idr_remove called for id=.. which is not allocated.
470 Details : Last kernels save old s_dev before kill super and not allow
471 to restore from callback - restore it before call kill_anon_super.
475 Description: buffer overruns could theoretically occur
476 Details : llapi_semantic_traverse() modifies the "path" argument by
477 appending values to the end of the origin string, and a
478 overrun may occur. Adding buffer overrun check in liblustreapi.
482 Description: Fix error on 'ls .' at the top of the Lustre mount.
483 Details : Don't revalidate dentry if it is a root dentry.
487 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
488 Details : Modify vm-tunables-rhel4.patch.
492 Description: Kernel config for 2.6.18-vanilla.
493 Details : Modify targets/2.6-vanilla.target.in.
494 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
495 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
496 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
497 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
501 Description: improve handling recoverable errors
502 Details : if request processig with error which can be recoverable on server
503 request should be resend, otherwise page released from cache and
508 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
509 Details : Modify target file & which_patch.
510 A flaw was found in the IA32 system call emulation provided
511 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
512 value could be stored in the %RAX register, which could trigger an
513 out-of-bounds system call table access. An untrusted local user
514 could exploit this flaw to run code in the kernel
515 (ie a root privilege escalation). (CVE-2007-4573).
519 Description: O_DIRECT bypasses client statistics.
520 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
521 read_ahead_stats were updated. copied the stats section from
522 osc_send_oap_rpc() into async_internal().
526 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
530 Description: change order of libsysio includes
531 Details : '#include sysio.h' should always come before '#include xtio.h'
535 Description: handle "serious error: objid * already exists" more gracefully
536 Details : If LAST_ID value on disk is smaller than the objects existing in
537 the O/0/d* directories, it indicates disk corruption and causes an
538 LBUG(). If the object is 0-length, then we should use the existing
539 object. This will help to avoid a full fsck in most cases.
541 Severity : enhancement
543 Description: adapt the lustre_config script to support the upgrade case
544 Details : Add "-u" option for lustre_config script to support upgrading 1.4
545 server targets to 1.6 in parallel.
549 Description: To avoid grant space > avaible space when the disk is almost
550 full. Without this patch you might see the error "grant XXXX >
551 available" or some LBUG about grant, when the disk is almost
553 Details : In filter_check_grant, for non_grant cache write, we should
554 check the left space by if (*left > ungranted + bytes), instead
555 of (*left > ungranted), because only we are sure the left space
556 is enough for another "bytes", then the ungrant space should be
557 increase. In client, we should update cl_avail_grant only there
558 is OBD_MD_FLGRANT in the reply.
562 Description: Update RHEL 4 kernel to fix local root privilege escalation.
563 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
564 described in CVE-2007-4573. This problem could allow untrusted
565 local users to gain root access.
568 Frequency : when using O_DIRECT and quotas
570 Description: Incorrect file ownership on O_DIRECT output files
571 Details : block usage reported by 'lfs quota' does not take into account
572 files that have been written with O_DIRECT.
577 Description: touch file failed when fs is not full
578 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
579 otherwise we can get ENOSP while fs is not full.
583 Description: parallel lock callbacks
584 Details : Instead of sending blocking and completion callbacks as separated
585 requests, adding them to a set and sending in parallel.
588 Frequency : only for Cray XT3
589 Bugzilla : 12829/13455
590 Description: Changing primary group doesn't change the group lustre assigns to
592 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
593 overridden with the primary group provided by the group upcall,
594 whereas the supplied fsgid can be trusted if it is in the list of
595 supplementary groups returned by the group upcall.
597 Severity : enhancement
599 Description: Allow masking D_WARNING, D_ERROR messages from console
600 Details : Console messages can now be disabled via lnet.printk.
604 Description: User code with malformed file open parameter crashes client node
605 Details : Before packing join_file req, all the related reference should be
606 checked carefully in case some malformed flags cause fake
607 join_file req on client.
609 --------------------------------------------------------------------------------
611 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
613 * Support for kernels:
614 2.6.5-7.283 (SLES 9),
615 2.6.9-55.EL (RHEL 4),
616 2.6.16.46-0.14 (SLES 10),
617 2.6.18.8 vanilla (kernel.org)
618 * Client support for unpatched kernels:
619 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
620 2.6.16 - 2.6.22 vanilla (kernel.org)
621 * Due to recently discovered recovery problems, we do not recommend
622 using patchless RHEL 4 clients with this or any earlier release.
623 * Recommended e2fsprogs version: 1.39.cfs8
624 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
625 * Starting with this release, the ldiskfs backing filesystem required
626 by Lustre is now in its own package, lustre-ldiskfs. This package
627 should be installed. It is versioned separately from Lustre and
628 may be released separately in future.
632 Description: block reactivating mgc import until all deactivates complete
633 Details : Fix race when failing back MDT/MGS to itself (testing)
636 Frequency : at statup only
638 Description: mds_lov_synchronize race leads to various problems
639 Details : simultaneous MDT->OST connections at startup can cause the
640 sync to abort, leaving the OSC in a bad state.
642 Severity : enhancement
644 Description: add optional extra BUILD_VERSION info
645 Details : add a new environment variable (namely LUSTRE_VERS) which allows
646 to override the lustre version.
649 Frequency : 2.6.18 servers only
651 Description: ll_kern_mount() doesn't release the module reference
652 Details : The ldiskfs module reference count never drops down to 0
653 because ll_kern_mount() doesn't release the module reference.
658 Description: server LBUG when using old ost_num_threads parameter
659 Details : Accept the old ost_num_threads parameter but warn that it
660 is deprecated, and fix an off-by-one error that caused an LBUG.
665 Description: Transient SCSI error results in persistent IO issue
666 Details : iobuf->dr_error is not reinitialized to 0 between two
670 Frequency : sometimes when underlying device returns I/O errors
672 Description: OSTs not going read-only during write failures
673 Details : OSTs are not remounted read-only when the journal commit threads
674 get I/O errors because fsfilt_ext3 calls journal_start/stop()
675 instead of the ext3 wrappers.
679 Description: poor connect scaling with increasing client count
680 Details : Don't run filter_grant_sanity_check for more than 100 exports
681 to improve scaling for large numbers of clients.
684 Frequency : SLES10 only
686 Description: sanity-quota.sh quotacheck failed: rc = -22
687 Details : Quotas cannot be enabled on SLES10.
690 Frequency : liblustre clients only
692 Description: getdirentries does not give error when run on compute nodes
693 Details : getdirentries does not fail when the size specified as an argument
694 is too small to contain at least one entry
696 Severity : enhancement
698 Description: Add LNET router traceability for debug purposes
699 Details : If a checksum failure occurs with a router as part of the
700 IO path, the NID of the last router that forwarded the bulk data
701 is printed so it can be identified.
706 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
707 Details : Due to a race condition, liblustre clients were occasionally
710 Severity : enhancement
712 Description: lfs setstripe use optional parameters instead of postional
715 Severity : enhancement
717 Description: Nanosecond timestamp support for ldiskfs
718 Details : The on-disk ldiskfs filesystem has added support for nanosecond
719 resolution timestamps. There is not yet support for this at
720 the Lustre filesystem level.
723 Frequency : during server recovery
725 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
726 Details : request with rq_no_resend flag not awake l_wait_event if they get a
730 Frequency : nfs export on patchless client
732 Description: connectathon hang when test nfs export over patchless client
733 Details : Disconnected dentry cannot be found with lookup, so we do not need
734 to unhash it or make it invalid
737 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
738 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
739 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
744 Description: open req refcounting wrong on reconnect
745 Details : If reconnect happened between getting open reply from server and
746 call to mdc_set_replay_data in ll_file_open, we will schedule
747 replay for unreferenced request that we are about to free.
748 Subsequent close will crash in variety of ways.
749 Check that request is still eligible for replay in
750 mdc_set_replay_data().
755 Description: disable writes to filesystem when reading health_check file
756 Details : the default for reading the health_check proc file has changed
757 to NOT do a journal transaction and write to disk, because this
758 can cause reads of the /proc file to hang and block HA state
759 checking on a healthy but otherwise heavily loaded system. It
760 is possible to return to the previous behaviour during configure
761 with --enable-health-write.
763 Severity : enhancement
765 Description: 64-bit inode version
766 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
767 to the inode. This will be required for version-based recovery.
772 Description: MDS fails to start if a duplicate client export is detected
773 Details : in some rare cases it was possible for a client to connect to
774 an MDS multiple times. Upon recovery the MDS would detect this
775 and fail during startup. Handle this more gracefully.
777 Severity : enhancement
779 Description: Add -o localflock option to simulate old noflock
781 Details : This will achieve local-only flock/fcntl locks
787 Description: log_commit_thread vs filter_destroy race leads to crash
788 Details : Take import reference before releasing llog record semaphore
793 Description: Wrong request locking in request set processing
794 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
795 imp_delayed_list, in this place should be used imp_lock.
798 Frequency : when reconnection
800 Description: Grant Leak when osc reconnect to OST
801 Details : When osc reconnect ost, OST(filter) should check whether it
802 should grant more space to client by comparing fed_grant and
803 cl_avail_grant, and return the granted space to client instead
804 of "new granted" space, because client will call osc_init_grant
805 to update the client grant space info.
808 Frequency : when client reconnect to OST
810 Description: Grant Leak when osc do resend and replay bulk write
811 Details : When osc reconnect to OST, OST(filter)should clear grant info of
812 bulk write request, because the grant info while be sync between
813 OSC and OST when reconnect, and we should ignore the grant info
814 these of resend/replay write req.
819 Description: Grant space more than avaiable left space sometimes.
820 Details : When then OST is about to be full, if two bulk writing from
821 different clients came to OST. Accord the avaliable space of the
822 OST, the first req should be permitted, and the second one
823 should be denied by ENOSPC. But if the seconde arrived before
824 the first one is commited. The OST might wrongly permit second
825 writing, which will cause grant space > avaiable space.
828 Frequency : when client is evicted
830 Description: Grant might be wrongly erased when osc is evicted by OST
831 Details : when the import is evicted by server, it will fork another
832 thread ptlrpc_invalidate_import_thread to invalidate the
833 import, where the grant will be set to 0. While the original
834 thread will update the grant it got when connecting. So if
835 the former happened latter, the grant will be wrongly errased
836 because of this race.
841 Description: Checking Stale with correct fid
842 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
843 to check whether it is stale, because sometimes, we want the
844 enqueue happened anyway, and op_data.fid2 will not be initialized.
846 Severity : enhancement
848 Description: update patchless client
849 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
852 Frequency : only with 2.4 kernel
854 Description: random memory corruption
855 Details : size of struct ll_inode_info is to big for union inode.u and this
856 can be cause of random memory corruption.
861 Description: Memory leak in recovery
862 Details : Lov_mds_md was not free in an error handler in mds_create_object.
863 It should also check obd_fail before fsfilt_start, otherwise if
864 fsfilt_start return -EROFS,(failover mds during mds recovery).
865 then the req will return with repmsg->transno = 0 and rc = EROFS.
866 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
867 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
868 be freed no matter whether fsfilt_commit success or not.
871 Frequency : only with huge count clients
873 Description: Prevents from taking the superblock lock in llap_from_page for
875 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
876 for llap_from_page call in ll_removepage prevents from taking the
877 superblock lock for a soon died page.
882 Description: Not check open intent error before release open handle
883 Details : in some rare cases, the open intent error is not checked before
884 release open handle, which may cause
885 ASSERTION(open_req->rq_transno != 0), because it tries to release
886 the failed open handle.
891 Description: Set cat log bitmap only after create log success.
892 Details : in some rare cases, the cat log bitmap is set too early. and it
893 should be set only after create log success.
897 Description: Accessing a block bevice can re-enable I/O when Lustre is
898 tearing down a device.
899 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
904 Description: service threads may hog cpus when there are a lot of requests
905 Details : Insert cond_resched to give other threads a chance to use some CPU
910 Description: the cat log was not initialized in recovery
911 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
912 unlink log on mds will not be initialized until mds post
913 recovery. And also in mds post recovery, the unlink log will
914 initialization will be done asynchronausly, so there will be race
915 between add unlink log and unlink log initialization.
919 Description: brw_stats were being printed incorrectly
920 Details : brw_stats were being printed as log2 but all of them were not
921 recorded as log2. Also remove some code duplication arising from
922 filter_tally_{read,write}.
926 Frequency : rare, only in recovery.
927 Description: ASSERTION(req->rq_type != LI_POISON) failed
928 Details : imp_lock should be held while iterating over imp_sending_list for
929 prevent destroy request after get timeout in ptlrpc_queue_wait.
933 Description: replay-single.sh test 52 fails
934 Details : A lock's skiplist need to be cleanup when it being unlinked
935 from its resource list.
939 Description: Short directio read returns full requested size rather than
941 Details : Direct I/O operations should return actual amount of bytes
942 transferred rather than requested size.
944 Severity : enhancement
946 Description: metadata RPC reduction (e.g. for rm performance)
947 Details : decrease the amount of synchronous RPC between clients and servers
948 by canceling conflicing lock before the operation on the client
949 and packing thier handles into the main operation RPC to server.
951 Severity : enhancement
953 Description: Async OSC create to avoid the blocking unnecessarily.
954 Details : If a OST has no remain object, system will block on the creating
955 when need to create a new object on this OST. Now, ways use
956 pre-created objects when available, instead of blocking on an
957 empty osc while others are not empty. If we must block, we block
958 for the shortest possible period of time.
960 Severity : enhancement
962 Description: refine locking for avoid write wrong info into lov_objid file
963 Details : fix possible races with add new target and write/update data in
966 --------------------------------------------------------------------------------
968 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
973 Frequency : on some architectures
975 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
976 Details : Uninitialized flags sometimes cause configuration commands to
980 Frequency : patchless clients only
982 Description: missing __iget() symbol export
983 Details : The __iget() symbol export is missing. To avoid the need for
984 this on patchless clients the deathrow inode reaper is turned
985 off, and we depend on the VM to clean up old inodes. This
986 dependency was during via the fix for bug 12181.
988 --------------------------------------------------------------------------------
990 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
992 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
993 INTEROPERATE with older versions automatically. In many cases a
994 special upgrade step is needed. Please read the
995 user documentation before upgrading any part of a live system.
996 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
997 version will not interoperate with 1.6 betas before beta5 (1.5.95).
998 * WARNING: Lustre configuration and startup changes are required with
999 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1005 Severity : enhancement
1007 Description: MountConf
1008 Details : Lustre configuration is now managed via mkfs and mount
1009 commands instead of lmc and lconf. New obd types (MGS, MGC)
1010 are added for dynamic configuration management. See
1011 https://mail.clusterfs.com/wikis/lustre/MountConf for
1014 Severity : enhancement
1016 Description: dynamic OST addition
1017 Details : OSTs can now be added to a live filesystem
1019 Severity : enhancement
1021 Description: startup order invariance
1022 Details : MDTs and OSTs can be started in any order. Clients only
1023 require the MDT to complete startup.
1025 Severity : enhancement
1027 Description: parallel, asynchronous orphan cleanup
1028 Details : orphan cleanup is now performed in separate threads for each
1029 OST, allowing parallel non-blocking operation.
1031 Severity : enhancement
1033 Description: optimized stripe assignment
1034 Details : stripe assignments are now made based on ost space available,
1035 ost previous usage, and OSS previous usage, in order to try
1036 to optimize storage space and networking resources.
1038 Severity : enhancement
1040 Description: Permanently set tunables
1041 Details : All writable /proc/fs/lustre tunables can now be permanently
1042 set on a per-server basis, at mkfs time or on a live system.
1044 Severity : enhancement
1046 Description: Lustre message v2
1047 Details : Add lustre message format v2.
1049 Severity : enhancement
1051 Description: client OST exclusion list
1052 Details : Clients can be started with a list of OSTs that should be
1053 declared "inactive" for known non-responsive OSTs.
1057 Description: SPEC SFS validation failure on NFS v2 over lustre.
1058 Details : Changes the blocksize for regular files to be 2x RPC size,
1059 and not depend on stripe size.
1061 Severity : enhancement
1063 Description: Multiple MD RPCs in flight.
1064 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1065 To avoid overly-overloading MDS, introduce a limit on number of
1066 MDS RPCs in flight for a single client and add /proc controls
1067 to adjust this limit.
1069 Severity : enhancement
1071 Description: client read/write statistics
1072 Details : Add client read/write call usage stats for performance
1073 analysis of user processes.
1074 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1075 file access. extents_stats shows chunk size distribution.
1076 extents_stats_per_process show chunk size distribution per
1079 Severity : enhancement
1081 Description: mds statistics
1082 Details : Add detailed mds operations statistics in
1083 /proc/fs/lustre/mds/*/stats.
1087 Description: Failure of copying files with lustre special EAs.
1088 Details : Client side always return success for setxattr call for lustre
1089 special xattr (currently only "trusted.lov").
1094 Description: Refcount LNET uuids
1095 Details : The global LNET uuid list grew linearly with every startup;
1096 refcount repeated list entries instead of always adding to
1099 Severity : enhancement
1101 Description: Dynamic service threads
1102 Details : Within a small range, start extra service threads
1103 automatically when the request queue builds up.
1106 Frequency : mixed-endian client/server environments
1108 Description: mixed-endian crashes
1109 Details : The new msg_v2 system had some failures in mixed-endian
1112 Severity : enhancement
1114 Description: Easy OST removal
1115 Details : OSTs can be permanently deactivated with e.g. 'lctl
1116 conf_param lustre-OST0001.osc.active=0'
1118 Severity : enhancement
1120 Description: MGS proc entries
1121 Details : Added basic proc entries for the MGS showing what filesystems
1124 Severity : enhancement
1126 Description: provide MGS failover
1127 Details : Added config lock reacquisition after MGS server failover.
1129 Severity : enhancement
1131 Description: add Linux 2.4 support
1132 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1136 Description: a large application tries to do I/O to the same resource and dies
1137 in the middle of it.
1138 Details : Check the req->rq_arrival time after the call to
1139 ost_brw_lock_get(), but before we do anything about
1140 processing it & sending the BULK transfer request. This
1141 should help move old stale pending locks off the queue as
1142 quickly as obd_timeout.
1145 Frequency : when an incorrect nid is specified during startup
1147 Description: ptlrpc connect to non-existant node causes kernel crash
1148 Details : LNET can't be re-entered from an event callback, which
1149 happened when we expire a message after the export has been
1150 cleaned up. Instead, hand the zombie cleanup off to another
1153 Severity : enhancement
1155 Description: plain/inodebits lock performance improvement
1156 Details : Grouping plain/inodebits in granted list by their request modes
1157 and bits policy, thus improving the performance of search through
1161 Frequency : only if OST filesystem is corrupted
1163 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1164 Details : for a short time RPCs with bulk IO are in the replay list,
1165 but replay of bulk IOs is unimplemented. If the OST filesystem
1166 is corrupted due to disk cache incoherency and then replay is
1167 started it is possible to trip an assertion. Avoid putting
1168 committed RPCs into the replay list at all to avoid this issue.
1171 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1173 Description: System hang on startup
1174 Details : This bug allowed the liblustre (e.g. catamount) client to
1175 return to the app before handling all startup RPCs. This
1176 could leave the node unresponsive to lustre network traffic
1177 and manifested as a server ptllnd timeout.
1179 Severity : enhancement
1181 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1182 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1183 boolean to control whether to print peer debug info when a
1184 client's RPC times out.
1187 Frequency : only for kernels with patches from Lustre below 1.4.3
1189 Description: Remove old rdonly API
1190 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1193 Frequency : only for devices with external journals
1195 Description: Set external device read-only also
1196 Details : During a commanded failover stop, we set the disk device
1197 read-only while the server shuts down. We now also set any
1198 external journal device read-only at the same time.
1201 Frequency : when upgrading from 1.4 while trying to change parameters
1203 Description: The wrong (new) MDC name was used when setting parameters for
1204 upgraded MDT's. Also allows changing of OSC (and MDC)
1205 parameters if --writeconf is specified at tunefs upgrade time.
1208 Frequency : when setting specific ost indicies
1210 Description: QOS code breaks on skipped indicies
1211 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1212 created with --index need not be sequential.
1214 Severity : enhancement
1216 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1217 Details : The uninit_groups feature works in conjunction with the kernel
1218 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1219 up the pass1 processing of e2fsck. This is a read-only feature
1220 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1221 mount filesystems that have had this feature enabled.
1223 Severity : enhancement
1225 Description: Improve multi-block allocation algorithm to avoid fragmentation
1226 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1227 allocation locality and avoid filesystem fragmentation.
1229 ------------------------------------------------------------------------------
1231 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1233 * Support for kernels:
1234 2.4.21-47.0.1.EL (RHEL 3)
1235 2.6.5-7.283 (SLES 9)
1236 2.6.9-42.0.10.EL (RHEL 4)
1237 2.6.12.6 vanilla (kernel.org)
1238 2.6.16.27-0.9 (SLES 10)
1239 * Recommended e2fsprogs version: 1.39.cfs5
1241 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1245 Frequency : occasional, depends on client load and configuration
1246 Bugzilla : 12181, 12203
1247 Description: data loss for recently-modified files
1249 Details : In some cases it is possible that recently written or created
1250 files may not be written to disk in a timely manner (this should
1251 normally be within 30s unless client IO load is very high).
1252 The problem appears as zero-length files or files that are a
1253 multiple of 1MB in size after a client crash or client eviction
1254 that are missing data at the end of the file.
1256 This problem is more likely to be hit on clients where files are
1257 repeatedly created and unlinked in the same directory, clients
1258 have a large amount of RAM, have many CPUs, the filesystem has
1259 many OSTs, the clients are rebooted frequently, and/or the files
1260 are not accessed by other nodes after being written.
1262 The presence of the problem can be detected by looking at
1263 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1264 smaller than the second (nr_unused) then dirty files will not
1265 be flushed automatically to disk. "sync; sleep 10" should be
1266 run several times on the node before unmounting it to update
1267 Lustre (this is also safe to run on nodes without this problem).
1269 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1270 that can cause this same problem, so customers using that kernel
1271 also need to update the kernel in addition to Lustre. In order
1272 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1274 It is normal that files written just before a client crash (less
1275 than 30s) may not yet have been flushed to disk, even for local
1279 Frequency : frequent on thin XT3 nodes
1281 Description: UUID collision on thin XT3 Linux nodes
1282 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1283 randomly, since we relied on an insufficiently-seeded PRNG.
1288 Description: OSS hangs after "All ost request buffers busy"
1289 Details : A deadlock between quota and journal operations caused OSS
1290 hangs after printing "All ost request buffers busy."
1293 Frequency : always on liblustre builds
1295 Description: Cleanup compiler warnings on liblustre
1298 Frequency : always on liblustre builds on XT3
1300 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1304 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1305 > 32000 subdirectories
1306 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1307 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1308 subdirectory count crosses 32000. This will aid e2fsck to
1309 correctly handle more than 32000 subdirectories.
1312 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1314 Description: System hang on startup
1315 Details : This bug allowed the liblustre (e.g. catamount) client to
1316 return to the app before handling all startup RPCs. This
1317 could leave the node unresponsive to lustre network traffic
1318 and manifested as a server ptllnd timeout.
1320 Severity : enhancement
1322 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1323 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1324 boolean to control whether to print peer debug info when a
1325 client's RPC times out.
1330 Description: make O_SYNC working on 2.6 kernels
1331 Details : 2.6 kernels use different method for mark pages for write,
1332 so need add a code to lustre for O_SYNC work.
1337 Description: Failure to close file and release space on NFS
1338 Details : Put inode details into lock acquired in ll_intent_file_open.
1339 Use mdc_intent_lock in ll_intent_open to properly
1340 detect all kind of errors unhandled by mdc_enqueue.
1345 Description: proc file read during shutdown sometimes raced obd removal,
1347 Details : Add lock to prevent obd access after proc file removal.
1350 Frequency : Only for files larger than 4GB on 32-bit clients.
1352 Description: improperly doing page alignment of locks
1353 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1354 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1359 Description: RPCs being resent when they shouldn't be
1360 Details : Some RPCs that should not be resent are being resent. This
1361 can cause inconsistencies in the RPC state machine. Do not
1362 resend such requests.
1365 Frequency : rare, only with NFS export
1367 Description: Crash on NFS re-export node
1368 Details : under very unusual load conditions an assertion is hit in
1369 ll_intent_file_open()
1372 Frequency : only if OST filesystem is corrupted
1374 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1375 Details : for a short time RPCs with bulk IO are in the replay list,
1376 but replay of bulk IOs is unimplemented. If the OST filesystem
1377 is corrupted due to disk cache incoherency and then replay is
1378 started it is possible to trip an assertion. Avoid putting
1379 committed RPCs into the replay list at all to avoid this issue.
1384 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1385 Details : Large single O_DIRECT read and write calls can fail to allocate
1386 a sufficiently large buffer to process the request. In case of
1387 allocation failure the allocation is retried with a smaller
1388 buffer and broken into smaller requests.
1390 Severity : enhancement
1392 Description: Add -o localflock option to simulate old noflock behaviour.
1393 Details : This will achieve local-only flock/fcntl locks coherentness.
1398 Description: versioning check is incomplete
1399 Details : Checking the version difference of client vs. server, report
1400 error if the gap is too big.
1405 Description: add support PG_writeback bit
1406 Details : add support for PG_writeback bit for Lustre, for more carefull
1407 work with page cache in 2.6 kernel. This also fix some deadlocks
1408 and remove hack for work O_SYNC with 2.6 kernel.
1410 Severity : enhancement
1412 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1413 Details : The uninit_groups feature works in conjunction with the kernel
1414 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1415 up the pass1 processing of e2fsck. This is a read-only feature
1416 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1417 mount filesystems that have had this feature enabled.
1419 Severity : enhancement
1421 Description: Improve multi-block allocation algorithm to avoid fragmentation
1422 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1423 allocation locality and avoid filesystem fragmentation.
1425 ------------------------------------------------------------------------------
1427 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1429 * Support for kernels:
1430 2.6.9-42.0.3.EL (RHEL 4)
1431 2.6.5-7.276 (SLES 9)
1432 2.4.21-47.0.1.EL (RHEL 3)
1433 2.6.12.6 vanilla (kernel.org)
1434 2.6.16.21-0.8 (SLES10)
1435 * Recommended e2fsprogs version: 1.39.cfs2-0
1437 * The backwards-compatible /proc/sys/portals symlink has been removed
1438 in this release. Before upgrading, please ensure that you change
1439 any configuration scripts or /etc/sysctl.conf files that access
1440 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1441 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1442 in advance of the upgrade on any system running Lustre 1.4.6 or
1443 newer, since /proc/sys/lnet was added in that version.
1444 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1448 Frequency : only when quota is used
1450 Description: avoid scanning export list for quota master
1451 Details : Change the algorithms to avoid scanning export list in order
1452 to improve the efficiency.
1455 Frequency : MDS failover only, very rarely
1457 Description: "went back in time" messages on mds failover
1458 Details : The greatest transno may be lost when the current operation
1459 finishes with an error (transno==0) and the client's last_rcvd
1460 record is over-written. Save the greatest transno in the
1461 mds_last_transno for this case.
1464 Frequency : always for specific kernels and striping counts
1466 Description: client may get "Matching packet too big" without ACL support
1467 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1468 when trying to access files in certain configurations. The
1469 clients should in fact be denied when mounting because they do
1470 not understand ACLs.
1473 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1475 Description: many clients connecting with IO in progress causes connect timeouts
1476 Details : Avoid synchronous journal commits to avoid delays caused by many
1477 clients connecting/disconnecting when bulk IO is in progress.
1478 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1479 OST_IO_PORTAL to avoid delays behind potentially many pending
1483 Frequency : occasionally with multiple writers to a single file
1485 Description: shared writes to file may result in wrong size reported by stat()
1486 Details : Allow growing of kms when extent lock is cancelled
1489 Frequency : always with random mmap IO to multi-striped file
1491 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1492 Details : Only if the hole is at the end of OST object so that kms is too
1493 small. Fix is to increase kms accordingly in ll_nopage.
1496 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1498 Description: writes to a missing object would leak memory on the OST
1499 Details : If there is an inconsistency between the MDS and OST filesystems,
1500 such that the MDS references an object that doesn't exist, writes
1501 to that object will leak memory due to incorrect cleanup in the
1502 error handling path, eventually running out of memory on the OST.
1507 Description: Creating too long symlink causes lustre errors
1508 Details : Check symlink and name lengths before sending requests to MDS.
1511 Frequency : only if flock is enabled (not on by default)
1513 Description: posix locks not released on fd closure on 2.6.9+
1514 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1515 kernels, this caused such locks not to be released on fd close and
1516 then assertions on fs unmount about still used locks.
1519 Frequency : MDS failover only, very rarely
1521 Description: clients may get ASSERTION(granted_lock != NULL)
1522 Details : When request was taking a long time, and a client was resending
1523 a getattr by name lock request. The were multiple lock requests
1524 with the same client lock handle and
1525 mds_getattr_name->fixup_handle_for_resent_request found one of the
1526 lock handles but later failed with ASSERTION(granted_lock != NULL).
1531 Description: handle->h_buffer_credits > 0, assertion failure
1532 Details : h_buffer_credits is zero after truncate, causing assertion
1533 failure. This patch extends the transaction or creates a new
1537 Frequency : NFS re-export or patchless client
1538 Bugzilla : 11179, 10796
1539 Description: Crash on NFS re-export node (__d_move)
1540 Details : We do not want to hash the dentry if we don't have a lock.
1541 But if this dentry is later used in d_move, we'd hit uninitialised
1542 list head d_hash, so we just do this to init d_hash field but
1543 leave dentry unhashed.
1546 Frequency : NFS re-export or patchless client
1548 Description: NFS exports has problem with symbolic link
1549 Details : lustre client didn't properly install dentry when re-exported
1550 to NFS or running patchless client.
1553 Frequency : NFS re-export or patchless client
1555 Description: Various nfs/patchless fixes.
1556 Details : fixes reuse disconected alias for lookup process - this fixes
1557 warning "find_exported_dentry: npd != pd",
1558 fix permission error with open files at nfs.
1559 fix apply umask when do revalidate.
1562 Frequency : occasional
1564 Description: Crash on NFS re-export node
1565 Details : calling clear_page() on the wrong pointer triggered oops in
1566 generic_mapping_read().
1569 Frequency : rarely, using O_DIRECT IO
1571 Description: unaligned directio crashes client with LASSERT
1572 Details : check for unaligned buffers before trying any requests.
1575 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1577 Description: stale data returned from RAID cache
1578 Details : If only a small amount of IO is done to the RAID device before
1579 reading it again it is possible to get stale data from the RAID
1580 cache instead of reading it from disk.
1583 Frequency : always for sles10 kernel
1585 Description: sles10 support
1586 Details : ll_follow_link: compile fixes and using of nd_set_link
1587 under newer kernels.
1590 Frequency : depends on arch, kernel and compiler version, always on sles10
1593 Description: recursive or deep enough symlinks cause stack overflow
1594 Details : getting rid of large stack-allocated variable in
1598 Frequency : depends on hardware
1600 Description: lustre write performance loss in the SLES10 kernel
1601 Details : the performance loss is caused by using of write barriers in the
1602 ext3 code. The SLES10 kernel turns barrier support on by
1603 default. The fix is to undo that change for ldiskfs.
1605 ------------------------------------------------------------------------------
1607 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1609 * Support for kernels:
1610 2.6.9-42.0.3EL (RHEL 4)
1611 2.6.5-7.276 (SLES 9)
1612 2.4.21-47.0.1.EL (RHEL 3)
1613 2.6.12.6 vanilla (kernel.org)
1617 Frequency : quota enabled and large files being deleted
1619 Description: releasing more than 4GB of quota at once hangs OST
1620 Details : If a user deletes more than 4GB of files on a single OST it
1621 will cause the OST to spin in an infinite loop. Release
1622 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1627 Description: statfs data retrieved from /proc may be stale or zero
1628 Details : When reading per-device statfs data from /proc, in the
1629 {kbytes,files}_{total,free,avail} files, it may appear
1630 as zero or be out of date.
1633 Frequency : systems with MD RAID1 external journal devices
1635 Description: lconf's call to blkid is confused by RAID1 journal devices
1636 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1637 of returning all block devices that match the journal UUID.
1640 Frequency : always, for aggregate stripe size over 4GB
1642 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1643 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1644 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1647 Frequency : always if "lfs find" used on a local file/directory
1649 Description: "lfs find" segfaults if used on a local file/directory
1650 Details : The case where a directory component was not specified wasn't
1651 handled correctly. Handle this properly.
1654 Frequency : always on ppc64
1656 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1657 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1658 perform endianness conversion.
1661 Frequency : rarely (truncate to non-zero file size after write under load)
1662 Bugzilla : 10730, 10687
1663 Description: Files padded with zeros to next 4K multiple
1664 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1665 it is possible that files that are truncated to a non-zero size
1666 immediately after being written are filled with zero bytes beyond
1667 the truncated size. No file data is lost.
1669 Severity : enhancement
1671 Description: Allow recovery/failover for liblustre clients.
1672 Details : liblustre clients were unaware of failover configurations until
1675 Severity : enhancement
1677 Description: user file locks should fail when not mounting with flock option
1678 Details : Set up an error-returning stub in ll_file_operations.lock field
1679 to prevent incorrect behaviour when client is mounted without
1680 flock option. Also, set up properly f_op->flock field for
1684 Frequency : always on ia64
1686 Description: "lfs df" loops on printing out MDS statfs information
1687 Details : The obd_ioctl_data was not initialized and in some systems
1688 this caused a failure during the ioctl that did not return
1689 an error. Initialize the struct and return an error on failure.
1692 Frequency : SLES 9 only
1694 Description: Error of copying files with lustre special EAs as root
1695 Details : Client side always return success for setxattr call for lustre
1696 special xattr (currently only "trusted.lov").
1699 Frequency : rarely on clusters with both ia64+i386 clients
1701 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1702 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1703 clients are doing concurrent writes to the same file it is
1704 possible that the ia64 clients may LASSERT because the OST
1705 extent locks are not PAGE_SIZE aligned. Ensure that grown
1706 locks are always aligned on the request boundary.
1709 Frequency : specific use, occasional
1711 Description: Overwriting in use executable truncates on-disk binary image
1712 Details : If one node attempts to overwrite an executable in use by
1713 another node, we now correctly return ETXTBSY instead of
1714 truncating the file.
1716 Severity : enhancement
1718 Description: Async OSC create to avoid the blocking unnecessarily.
1719 Details : If a OST has no remain object, system will block on the creating
1720 when need to create a new object on this OST. Now, ways use
1721 pre-created objects when available, instead of blocking on an
1722 empty osc while others are not empty. If we must block, we block
1723 for the shortest possible period of time.
1728 Description: chmod on Lustre root is propagated to other clients
1729 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1730 invalid by the follow_mount time.
1733 Frequency : liblustre clients only
1735 Description: Race in 'instant cancel' lock handling could lead to such locks
1736 never to be granted in case of SMP MDS
1737 Details : Do not destroy not yet granted but cbpending locks in
1741 Frequency : replay/resend of open
1743 Description: non null lock assetion failure in mds_intent_policy
1744 Details : Trying to replay/resend lockless open requests resulted in
1745 mds_open() returning 0 with no lock. Now it sets a flag if
1746 a lock is going to be returned.
1748 Severity : enhancement
1750 Description: Checksum enhancements
1751 Details : New checksum enhancements allow for resending RPCs that failed
1754 Severity : enhancement
1756 Description: Tunables on number of dirty pages in cacche
1757 Details : Allow to set limit on number of dirty pages cached.
1762 Description: client crash on unmount - lock still has references
1763 Details : In some error handling cases it was possible to leak a lock
1764 reference on a client while accessing a file. This was not
1765 harmful to the client during operation, but would cause the
1766 client to crash when the filesystem is unmounted.
1769 Frequency : specific case, rare
1771 Description: ETXTBSY on mds though file not in use by client
1772 Details : ETXTBSY is no longer incorrectly returned when attempting to
1773 chmod or chown a directory that the user previously tried to
1774 execute or a currently-executing binary.
1777 Frequency : extremely rare except on liblustre-based clients
1779 Description: Lustre space not freed when files are deleted
1780 Details : Clean up open-unlinked files after client eviction. Previously
1781 the unlink was skipped and the files remained as orphans.
1786 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1787 Details : In some cases it was possible to send a blocking callback to
1788 a client doing a glimpse, even though that client didn't get
1789 a lock granted. When the glimpse lock is cancelled on the OST
1790 the freed lock is left on the waiting list and corrupted the list.
1793 Frequency : all core dumps
1795 Description: Broke core dumps to lustre
1796 Details : Negative dentry may be unhashed if parent does not have UPDATE
1797 lock, but some callers, e.g. do_coredump, expect dentry to be
1798 hashed after successful create, hash it in ll_create_it.
1800 ------------------------------------------------------------------------------
1802 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1804 * Support for kernels:
1805 2.6.9-42.0.2.EL (RHEL 4)
1806 2.6.5-7.276 (SLES 9)
1807 2.4.21-40.EL (RHEL 3)
1808 2.6.12.6 vanilla (kernel.org)
1812 Frequency : always on RHEL 3
1814 Description: Number of open files grows over time
1815 Details : The number of open files grows over time, whether or not
1816 Lustre is started. This was due to a filp leak introduced
1817 by one of our kernel patches.
1819 ------------------------------------------------------------------------------
1821 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1823 * Support for kernels:
1824 2.6.9-42.EL (RHEL 4)
1825 2.6.5-7.276 (SLES 9)
1826 2.4.21-40.EL (RHEL 3)
1827 2.6.12.6 vanilla (kernel.org)
1832 Bugzilla : 5719, 9635, 9792, 9684
1833 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1834 Details : If a server is under heavy load and cannot reply to new
1835 connection requests before the client resends the (re)connect,
1836 the connection handling code can behave badly if two service
1837 threads are concurrently handing separate (re)connections from
1838 the same client. Add better locking to the connection handling
1839 code, and ensure that only a single connection will be processed
1840 for a given client UUID, even if the lock is dropped.
1842 Severity : enhancement
1844 Description: add TCP zero-copy support to kernel
1845 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1846 sends if the hardware supports scatter-gather and checksumming.
1847 This allows socklnd to do client-write and server-read more
1848 efficiently and reduce CPU utilization from skbuf copying.
1851 Frequency : only if NFS exporting from client
1853 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1854 Details : If setting attributes on a file created under NFS that had
1855 never been opened it would be possible to oops the client
1856 if the file had no objects.
1859 Frequency : always for liblustre
1861 Description: liblustre client does MDS+OSTs setattr RPC for each write
1862 Details : When doing a write from a liblustre client, the client
1863 incorrectly issued an RPC to the MDS and each OST the file was
1864 striped over in order to update the timestamps. When writing
1865 with small chunks and many clients this could overwhelm the MDS
1866 with RPCs. In all cases it would slow down the write because
1867 these RPCs are unnecessary.
1869 Severity : enhancement
1871 Description: allow number of MDS service threads to be changed at module load
1872 Details : It is now possible to change the number of MDS service threads
1873 running. Adding "options mds mds_num_threads={N}" to the MDS's
1874 /etc/modprobe.conf will set the number of threads for the next
1875 time Lustre is restarted (assuming the "mds" module is also
1876 reloaded at that time). The default number of threads will
1877 stay the same, 32 for most systems.
1882 Description: OST crash if filesystem is unformatted or corrupt
1883 Details : If an OST is started on a device that has never been formatted
1884 or if the filesystem is corrupt and cannot even mount then the
1885 error handling cleanup routines would dereference a NULL pointer.
1890 Description: NULL pointer deref in llap_from_page.
1891 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1892 incorrect) mapping if the page was truncated/reclaimed while it was
1893 searched for. Check for this condition and skip such pages when
1894 doing readahead. Introduce extra check to llap_from_page() to
1895 verify page->mapping->host is non-NULL (so page is not anonymous).
1898 Frequency : Sometimes when using sys_sendfile
1900 Description: "page not covered by a lock" warnings from ll_readpage
1901 Details : sendfile called ll_readpage without right page locks present.
1902 Now we introduced ll_file_sendfile that does necessary locking
1903 around call to generic_file_sendfile() much like we do in
1907 Frequency : with certain MDS communication failures at client mount time
1909 Description: NULL pointer deref after failed client mount
1910 Details : a client connection request may delayed by the network layer
1911 and not be sent until after the PTLRPC layer has timed out the
1912 request. If the client fails the mount immediately it will try
1913 to clean up before the network times out the request. Add a
1914 reference from the request import to the obd device and delay
1915 the cleanup until the network drops the request.
1918 Frequency : occasionally during client (re)connect
1920 Description: assertion failure during client (re)connect
1921 Details : processing a client connection request may be delayed by the
1922 client or server longer than the client connect timeout. This
1923 causes the client to resend the connection request. If the
1924 original connection request is replied in this interval, the
1925 client may trip an assertion failure in ptlrpc_connect_interpret()
1926 which thought it would be the only running connect process.
1929 Frequency : only with obd_echo servers and clients that are rebooted
1931 Description: kernel BUG accessing uninitialized data structure
1932 Details : When running an obd_echo server it did not start the ping_evictor
1933 thread, and when a client was evicted an uninitialized data
1934 structure was accessed. Start the ping_evictor in the RPC
1935 service startup instead of the OBD startup.
1937 Severity : enhancement
1938 Bugzilla : 10193 (patchless)
1939 Description: Remove dependency on various unexported kernel interfaces.
1940 Details : No longer need reparent_to_init, exit_mm, exit_files,
1941 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1944 Frequency : rare (only users of deprecated and unsupported LDAP config)
1946 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1947 Details : LDAP apparently contains 'lustreName' attributes instead of
1948 'name'. A simple remapping of the name is sufficient.
1951 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1953 Description: waiting_locks_callback trips kernel BUG if client is evicted
1954 Details : Running with the dump_on_timeout debug flag turned on makes
1955 it possible that the waiting_locks_callback() can try to dump
1956 the Lustre kernel debug logs from an interrupt handler. Defer
1957 this log dumping to the expired_lock_main() thread.
1959 Severity : enhancement
1961 Description: Support NFS exporting on 2.6 kernels.
1962 Details : Implement non-rawops metadata methods for NFS server to use without
1963 changing NFS server code.
1966 Frequency : very rare (synthetic metadata workload only)
1968 Description: two racing renames might cause an MDS thread to deadlock
1969 Details : Running the "racer" program may cause one MDS thread to rename
1970 a file from being the source of a rename to being the target of
1971 a rename at exactly the same time that another thread is doing
1972 so, and the second thread has already enqueued these locks after
1973 doing a lookup of the target and is trying to relock them in
1974 order. Ensure that we don't try to re-lock the same resource.
1977 Frequency : only very large systems with liblustre clients
1979 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1980 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1981 all OSTs in parallel. This allows the eviction of stale liblustre
1982 clients to proceed much faster than if they were done in series,
1983 and also offers similar improvements for other set_info RPCs.
1988 Description: excessive CPU usage during initial read phase on client
1989 Details : During the initial read phase on a client, it would agressively
1990 retry readahead on the file, consuming too much CPU and impacting
1991 performance (since 1.4.5.8). Improve the readahead algorithm
1992 to avoid this, and also improve some other common cases (read
1993 of small files in particular, where "small" is files smaller than
1994 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1999 Description: MDS crash when receiving packet with unknown intent.
2000 Details : Do not LBUG in unknown intent case, just return -EFAULT
2002 Severity : enhancement
2003 Bugzilla : 9293, 9385
2004 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2005 Details : Do not serialize getattr (non-intent version) and statfs.
2008 Frequency : occasional, when OST network is overloaded/intermittent
2010 Description: client evicted by OST after bulk IO timeout
2011 Details : If a client sends a bulk IO request (read or write) the OST
2012 may evict the client if it is unresposive to its data GET/PUT
2013 request. This is incorrect if the network is overloaded (takes
2014 too long to transfer the RPC data) or dropped the OST GET/PUT
2015 request. There is no need to evict the client at all, since
2016 the pinger and/or lock callbacks will handle this, and the
2017 client can restart the bulk request.
2020 Frequency : Always when mmapping file with no objects
2022 Description: client crashes when mmapping file with no objects
2023 Details : Check that we actually have objects in a file before doing any
2024 operations on objects in ll_vm_open, ll_vm_close and
2030 Description: Request leak when working with deleted CWD
2031 Details : Introduce advanced request refcount tracking for requests
2032 referenced from lustre intent.
2034 Severity : Enhancement
2036 Description: Cache open file handles on client.
2037 Details : MDS now will return special lock along with openhandle, if
2038 requested and client is allowed to hold openhandle, even if unused,
2039 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2040 closing files for every read/write openration.
2042 Severity : Enhancement
2044 Description: Cache open negative dentries on client when possible.
2045 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2046 negative dentries on lock revocation.
2051 Description: Remounting a client read-only wasn't possible with a zconf mount
2052 Details : It wasn't possible to remount a client read-only with llmount.
2054 Severity : enhancement
2055 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2056 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2057 MPICH is included to add Lustre-specific ADIO interfaces.
2058 This is based closely on the UFS ADIO layer and only differs
2059 in file creation, in order to allow the OST striping to be set.
2060 This is user-contributed code and not supported by CFS.
2065 Description: extended inode attributes (immutable, append-only) work improperly
2066 when 2.4 and 2.6 kernels are used on client/server or vice versa
2067 Details : Introduce kernel-independent values for these flags.
2069 Severity : enhancement
2072 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2073 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2074 granularity. Now, Fractional MB granularity is be supported,
2075 this is very useful for low memory system.
2077 Severity : enhancement
2079 Description: Getattr by fid
2080 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2081 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2082 have locks and avoids getattr rpc requests for it on every stat.
2085 Frequency : Always, for filesystems larger than 2TB
2087 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2088 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2089 filesystems larger than 2TB to be mounted with mballoc enabled.
2092 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2094 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2095 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2096 filesystems larger than 2TB is mounted then the kernel will
2097 silently corrupt the start of the filesystem. CONFIG_LBD is
2098 enabled for all CFS-supported kernels, but the possibility of
2099 this happening with a modified kernel config exists.
2101 Severity : enhancement
2103 Description: add client O_DIRECT support for 2.6 kernels
2104 Details : It is now possible to do O_DIRECT reads and writes to files
2105 in the Lustre client mountpoint on 2.6 kernel clients.
2107 Severity : enhancement
2109 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2110 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2111 parallel, not waiting for response from every OST before sending
2112 a rpc to the next OST.
2117 Description: setattr vs write race when updating file timestamps
2118 Details : Client processes that update a file timestamp into the past
2119 right after writing to the file (e.g. tar) it is possible that
2120 the updated file modification time can be reset to the current
2121 time due to a race between processing the setattr and write RPC.
2123 Severity : enhancement
2125 Description: Bring 'lfs find' closer in line with regular Linux find.
2126 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2127 -print0 options and obtains all the needed info through the lustre
2130 Severity : enhancement
2132 Description: support up to 1024 configured devices on one node
2133 Details : change obd_dev array from statically allocated to dynamically
2134 allocated structs as they are first used to reduce memory usage
2139 Description: Flush dirty partially truncated pages during truncate
2140 Details : Immediatelly flush partially truncated pages in filter_setattr,
2141 this way we completely avoid having any pages in page cache on OST
2142 and can retire ugly workarounds during writes to flush such pages.
2147 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2148 Details : protect inode from truncation within vfs_unlink() context
2149 just take a reference before calling vfs_unlink() and release it
2150 when parent's i_sem is free.
2155 Description: last_id value checked outside lock on OST caused LASSERT failure
2156 Details : If there were multiple MDS->OST object precreate requests in
2157 flight, it was possible that the OST's last object id was checked
2158 outside a lock and incorrectly tripped an assertion. Move checks
2159 inside locks, and discard old precreate requests.
2162 Frequency : always, if extents are used on OSTs
2164 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2165 Details : OSTs using the extents format would not zero the high 16 bits of
2166 the index physical block number. This is not a problem for any
2167 OST filesystems smaller than 16TB, and no kernels support ext3
2168 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2169 new/modified files) and can be fixed for existing filesystems
2170 with e2fsprogs-1.39-cfs1.
2175 Description: import connection selection may be incorrect if timer wraps
2176 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2177 import connections to be ignored if the 32-bit jiffies counter
2178 wraps. Use a 64-bit jiffies counter.
2181 Frequency : very large clusters immediately after boot
2183 Description: LNET request buffers exhausted under heavy short-term load
2184 Details : If a large number of client requests are generated on a service
2185 that has previously never seen so many requests it is possible
2186 that the request buffer growth cannot keep up with the spike in
2187 demand. Instead of dropping incoming requests, they are held in
2188 the LND until the RPC service can accept more requests.
2191 Frequency : Sometimes during replay
2193 Description: Assertion failure in ll_local_open after replay.
2194 Details : If replay happened on an open request reply before we were able
2195 to set replay handler, reply will become not swabbed tripping the
2196 assertion in ll_local_open. Now we set the handler right after
2197 recognising of open request
2200 Frequency : very rare
2202 Description: kernel reports "badness in vsnprintf"
2203 Details : Reading from the "recovery_status" /proc file in small chunks
2204 may cause a negative length in lprocfs_obd_rd_recovery_status()
2205 call to vsnprintf() (which is otherwise harmless). Exit early
2206 if there is no more space in the output buffer.
2208 Severity : enhancement
2210 Description: clear OBD RPC statistics by writing to them
2211 Details : It is now possible to clear the OBD RPC statistics by writing
2212 to the "stats" file.
2217 Description: Client mtime is not the same on different clients after utimes
2218 Details : In some cases, the client was using the utimes() syscall on
2219 a file cached on another node. The clients now validate the
2220 ctime from the MDS + OSTs to determine which one is right.
2225 Description: Inability to activate failout mode
2226 Details : lconf script incorrectly assumed that in pythong string's numeric
2227 value is used in comparisons.
2230 Frequency : always with multiple stripes per file
2232 Description: Inefficient object allocation for mutli-stripe files
2233 Details : When selecting which OSTs to stripe files over, for files with
2234 a stripe count that divides evenly into the number of OSTs,
2235 the MDS is always picking the same starting OST for each file.
2236 Return the OST selection heuristic to the original design.
2241 Description: mount failures may take full timeout to return an error
2242 Details : Under some heavy load conditions it is possible that a
2243 failed mount can wait for the full obd_timeout interval,
2244 possibly several minutes, before reporting an error.
2245 Instead return an error as soon as the status is known.
2247 Frequency : quota enabled and large files being deleted
2249 Description: releasing more than 4GB of quota at once hangs OST
2250 Details : If a user deletes more than 4GB of files on a single OST it
2251 will cause the OST to spin in an infinite loop. Release
2252 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2257 Description: statfs data retrieved from /proc may be stale or zero
2258 Details : When reading per-device statfs data from /proc, in the
2259 {kbytes,files}_{total,free,avail} files, it may appear
2260 as zero or be out of date.
2263 Frequency : systems with MD RAID1 external journal devices
2265 Description: lconf's call to blkid is confused by RAID1 journal devices
2266 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2267 of returning all block devices that match the journal UUID.
2270 Frequency : always, for aggregate stripe size over 4GB
2272 Description: assertion fails when trying to use 4GB stripe size
2273 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2274 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2277 Frequency : always on ppc64
2279 Description: the first write on an ext3 filesystem with mballoc got stuck
2280 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2281 perform endianness conversion.
2283 ------------------------------------------------------------------------------
2285 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2287 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2288 INTEROPERATE with older versions automatically. Please read the
2289 user documentation before upgrading any part of a live system.
2290 * WARNING: Lustre networking configuration changes are required with
2291 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2294 * Support for kernels:
2295 2.6.9-22.0.2.EL (RHEL 4)
2296 2.6.5-7.244 (SLES 9)
2297 2.6.12.6 vanilla (kernel.org)
2300 Severity : enhancement
2301 Bugzilla : 7981/8208
2302 Description: Introduced Lustre Networking (LNET)
2303 Details : LNET is new networking infrastructure for Lustre, it includes
2304 a reorganized network configuration mode (see the user
2305 documentation for full details) as well as support for routing
2306 between different network fabrics. Lustre Networking Devices
2307 (LNDs) for the supported network fabrics have also been
2308 created for this new infrastructure.
2310 Severity : enhancement
2311 Description: Introduced Access control lists
2312 Details : clients can set ACLs on files and directories in order to have
2313 more fine-grained permissions than the standard Unix UGO+RWX.
2314 The MDS must be started with the "-o acl" mount option.
2316 Severity : enhancement
2317 Description: Introduced filesystem quotas
2318 Details : Administrators may now establish per-user quotas on the
2321 Severity : enhancement
2323 Description: Configuration change for the XT3
2324 The PTLLND is now used to run Lustre over Portals on the XT3
2325 The configure option(s) --with-cray-portals are no longer used.
2326 Rather --with-portals=<path-to-portals-includes> is used to
2327 enable building on the XT3. In addition to enable XT3 specific
2328 features the option --enable-cray-xt3 must be used.
2333 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2334 Details : A race between allocating a new llcd and re-getting the llcd_lock
2335 allowed another thread to grab newly-allocated llcd.
2337 Severity : enhancement
2339 Description: 2.6 OST async journal commit and locking fix to improve performance
2340 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2341 2.6 kernels are now async as they already were in 2.4 kernels so
2342 that they can commit concurrently with the network bulk transfer.
2343 For block-allocated files the filter allocation semaphore is held
2344 to avoid filesystem fragmentation during allocation. BKL lock
2345 removed for 2.6 xattr operations where it is no longer needed.
2350 Description: lconf incorrectly determined whether two IP networks could talk
2351 Details : In some more complicated routing and multiple-network
2352 configurations, lconf will avoid trying to make a network
2353 connection to a disjoint part of the IP space. It was doing the
2354 math incorrectly for one set of cases.
2359 Description: Fix for potential infinite loop processing records in an llog.
2360 Details : If an llog record is corrupted/zeroed, it is possible to loop
2361 forever in llog_process(). Validate the llog record length
2362 and skip the remainder of the block on error.
2365 Frequency : occasional (liblustre only)
2367 Description: liblustre could not open files whose last component is a symlink
2368 Details : sysio_path_walk() would incorrectly pass the open intent to
2369 intermediate path components.
2372 Frequency : rare (liblustre only with non-standard tuning)
2373 Bugzilla : 7201 (7350)
2374 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2375 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2376 after it has been released, only do this for referenced locks
2378 Severity : enhancement
2380 Description: specifying an (invalid) directory default stripe_size of -1
2381 would reset the directory default striping
2382 Details : stripe_size -1 was used internally to signal directory stripe
2383 removal, now use "all default" to signal dir stripe removal
2384 as a directory striping of "all default" is not useful
2387 Frequency : common for large clusters running liblustre clients
2389 Description: doing an ls when liblustre clients are running is slow
2390 Details : sending a glimpse AST to a liblustre client waits for every AST
2391 to time out, as liblustre clients will not respond. Since they
2392 cannot cache data we refresh the OST lock LVB from disk instead.
2394 Severity : enhancement
2396 Description: doing an ls at the same time as file IO can be slow
2397 Details : enqueue and other "small" requests can be blocked behind many
2398 large IO requests. Create a new OST IO portal for non-IO
2399 requests so they can be processed faster.
2402 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2404 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2405 created file with mode 000
2406 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2407 allow an ftruncate() as a non-root user to a file with mode 000.
2408 Lustre now respects this flag to disable mode checks when
2409 truncating a file owned by the user
2412 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2415 Description: Revoking locks from clients that went dead or catatonic might take
2417 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2418 cancellation of such locks instant on servers without waiting for
2419 any reply from clients. Clients drops these locks when cancel
2420 notification from server is received without replying.
2423 Frequency : liblustre-only, when liblustre client dies or becomes busy
2425 Description: Doing ls on Linux clients can take a long time with active
2427 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2428 granting such locks to it in the first place if possible. Locks
2429 are taken by proxy on the OST during the read or write and
2430 dropped immediately afterward. Add connect flags handling, do
2431 not grant locks to liblustre clients for glimpse ASTs.
2433 Severity : enhancement
2435 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2436 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2437 sequential reads and behaves badly with applications doing random
2438 reads. Improve it by reading ahead at least read region, and
2439 avoiding excessive large RPC for small reads.
2441 Severity : enhancement
2443 Description: Creating more than 1000 files for a single job may cause a load
2444 imbalance on the OSTs if there are also a large number of OSTs.
2445 Details : qos_prep_create() uses an OST index reseed value that is an
2446 even multiple of the number of available OSTs so that if the
2447 reseed happens in the middle of the object allocation it will
2448 still utilize the OSTs as uniformly as possible.
2453 Description: OST or MDS may oops in ping_evictor_main()
2454 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2455 but doesn't restart at beginning of obd_exports_timed list
2458 Severity : enhancement
2460 Description: improve by-nid export eviction on the MDS and OST
2461 Details : allow multiple exports with the same NID to be evicted at one
2462 time without re-searching the exports list.
2465 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2467 Description: MDS may oops in groups_free()
2468 Details : in rare race conditions a newly allocated group_info struct is
2469 freed again, and this can be NULL. The 2.4 compatibility code
2470 for groups_free() checked for a NULL pointer, but 2.6 did not.
2473 Frequency : common for liblustre clients doing little filesystem IO
2474 Bugzilla : 9352, 7313
2475 Description: server may evict liblustre clients accessing contended locks
2476 Details : if a client is granted a lock or receives a completion AST
2477 with a blocking AST already set it would not reply to the AST
2478 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2481 Frequency : lfs setstripe, only systems with more than 160 OSTs
2483 Description: unable to set striping with a starting offset beyond OST 160
2484 Details : llapi_create_file() incorrectly limited the starting stripe
2485 index to the maximum single-file stripe count.
2488 Frequency : LDAP users only
2490 Description: lconf did not handle in-kernel recovery with LDAP properly
2491 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2493 Severity : enhancement
2495 Description: bind OST threads to NUMA nodes to improve performance
2496 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2497 node and do their allocations there to localize memory access
2499 Severity : enhancement
2501 Description: llmount can determine client NID directly from Myrinet (GM)
2502 Details : the client NID code from gmnalnid was moved directly into
2503 llmount, removing the need to use this or specifying the
2504 client NID explicitly when mounting GM clients with zeroconf
2507 Frequency : if client is started with down MDS
2509 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2510 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2511 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2512 ptlrpc_import_delayed_req() because we didn't check if the
2513 request was interrupted, and we also didn't break out of the
2514 event loop if there was no timeout
2519 Description: data loss during non-page-aligned writes to a single file from
2520 both multiple nodes and multiple threads on one node at same time
2521 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2522 inconsistency led to false short-reads, that were cached and later
2523 used by ->prepare_write() to fill in partially written page,
2524 leading to data loss.
2527 Frequency : always, if lconf --abort_recovery used
2529 Description: lconf --abort_recovery fails with 'Operation not supported'
2530 Details : lconf was attempting to abort recovery on the MDT device and not
2533 Severity : enhancement
2535 Description: remove cleanup logs
2536 Details : replace lconf-generated cleanup logs with lustre internal
2537 cleanup routines. Eliminates the need for client-cleanup and
2540 Severity : enhancement
2542 Description: add support for EAs (user and system) on lustre filesystems
2543 Details : it is now possible to store extended attributes in the Lustre
2544 client filesystem, and with the user_xattr mount option it
2545 is possible to allow users to store EAs on their files also
2547 Severity : enhancement
2549 Description: Add possibility (config option) to show minimal available OST free
2551 Details : When compiled with --enable-mindf configure option, statfs(2)
2552 (and so, df) will return least minimal free space available from
2553 all OSTs as amount of free space on FS, instead of summary of
2554 free spaces of all OSTs.
2556 Severity : enhancement
2558 Description: do not expand extent locks acquired on OST-side
2559 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2560 by server: they are not cached anyway.
2563 Frequency : when mmap is used/binaries executed from Lustre
2565 Description: Unmmap pages before throwing them away from read cache.
2566 Details : llap_shrink cache now attempts to unmap pages before discarding
2567 them (if unmapping failed - do not discard). SLES9 kernel has
2568 extra checks that trigger if this unmapping is not done first.
2573 Description: lconf didn't resolve symlinks before checking to see whether a
2574 given mountpoint was already in use
2577 Frequency : when migrating failover services
2578 Bugzilla : 6395, 9514
2579 Description: When migrating a subset of services from a node (e.g. failback
2580 from a failover service node) the remaining services would
2581 time out and evict clients.
2582 Details : lconf --force (implied by --failover) sets the global obd_timeout
2583 to 5 seconds in order to quickly disconnect, but this caused
2584 other RPCs to time out too quickly. Do not change the global
2585 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2587 Severity : enhancement
2588 Frequency : if MDS is started with down OST
2589 Bugzilla : 9439,5706
2590 Description: Allow startup/shutdown of an MDS without depending on the
2591 availability of the OSTs.
2592 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2593 Add appropriate locking and lov-osc refcounts for safe
2594 cleaning. Add osc abort_inflight calls in case the
2595 synchronize never started.
2598 Frequency : occasional (Cray XT3 only)
2600 Description: root not authorized to access files in CRAY_PORTALS environment
2601 Details : The client process capabilities were not honoured on the MDS in
2602 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2603 been accessed by an authorized user then root was able to access
2604 the file on the local client also. The root user capabilities
2605 are now allowed on the MDS, as this environment has secure UID.
2608 Frequency : occasional
2610 Description: ldiskfs "too long searching" message happens too often
2611 Details : A debugging message (otherwise harmless) prints too often on
2612 the OST console. This has been reduced to only happen when
2613 there are fragmentation problems on the filesystem.
2618 Description: Division by zero in statfs when all OSCs are inactive
2619 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2620 lov_statfs divides by value returned by lov_get_stripecnt().
2624 Bugzilla : 9489, 3273
2625 Description: First write from each client to each OST was only 4kB in size,
2626 to initialize client writeback cache, which caused sub-optimal
2627 RPCs and poor layout on disk for the first writen file.
2628 Details : Clients now request an initial cache grant at (re)connect time
2629 and so that they can start streaming writes to the cache right
2630 away and always do full-sized RPCs if there is enough data.
2631 If the OST is rebooted the client also re-establishes its grant
2632 so that client cached writes will be honoured under the grant.
2637 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2638 Details : Now I/O RPCs go to different portal number and (presumably) fast
2639 lock requests (and glimses) and other RPCs get their own service
2640 threads pool that should be able to service those RPCs
2643 Severity : enhancement
2645 Description: Ability to exchange lustre version between client and servers and
2646 issue warnings at client side if client is too old. Also for
2647 liblustre clients there is ability to refuse connection of too old
2649 Details : New 'version' field is added to connect data structure that is
2650 filled with version info. That info is later checked by server and
2654 Frequency : rare, liblustre only.
2655 Bugzilla : 9296, 9581
2656 Description: Two simultaneous writes from liblustre at offset within same page
2657 might proceed at the same time overwriting eachother with stale
2659 Details : I/O lock withing llu_file_prwv was released too early, before data
2660 actually was hitting the wire. Extended lock-holding time until
2661 server acknowledges receiving data.
2664 Frequency : extremely rare. Never observed in practice.
2666 Description: avoid generating lustre_handle cookie of 0.
2667 Details : class_handle_hash() generates handle cookies by incrementing
2668 global counter, and can hit 0 occasionaly (this is unlikely, but
2669 not impossible, because initial value of cookie counter is
2670 selected randonly). Value of 0 is used as a sentinel meaning
2671 "unassigned handle" --- avoid it. Also coalesce two critical
2672 sections in this function into one.
2674 Severity : enhancement
2676 Description: allow liblustre clients to delegate truncate locking to OST
2677 Details : To avoid overhead of locking, liblustre client instructs OST to
2678 take extent lock in ost_punch() on client's behalf. New connection
2679 flag is added to handle backward compatibility.
2681 Severity : enhancement
2682 Bugzilla : 4928, 7341, 9758
2683 Description: allow number of OST service threads to be specified
2684 Details : a module parameter allows the number of OST service threads
2685 to be specified via "options ost ost_num_threads={N}" in the
2686 OSS's /etc/modules.conf or /etc/modprobe.conf.
2690 Bugzilla : 6146, 9635, 9895
2691 Description: servers crash with bad pointer in target_handle_connect()
2692 Details : In rare cases when a client is reconnecting it was possible that
2693 the connection request was the last reference for that export.
2694 We would temporarily drop the export reference and get a new
2695 one, but this may have been the last reference and the export
2696 was just destroyed. Get new reference before dropping old one.
2698 Severity : enhancement
2699 Frequency : if client is started with failover MDS
2701 Description: Allow multiple MDS hostnames in the mount command
2702 Details : Try to read the configuration from all specified MDS
2703 hostnames during a client mount in case the "primary"
2706 Severity : enhancement
2708 Description: Stop sending data to evicted clients as soon as possible.
2709 Details : Check if the client we are about to send or are sending data to
2710 was evicted already. (Check is done every second of waiting,
2711 for which l_wait_event interface was extended to allow checking
2712 of exit condition at specified intervals).
2715 Frequency : rare, normally only when NFS exporting is done from client
2717 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2719 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2722 Frequency : always, if srand() is called before liblustre initialization
2724 Description: Liblustre uses system PRNG disturbing its usage by user application
2725 Details : Introduce internal to lustre fast and high-quality PRNG for
2726 lustre usage and make liblustre and some other places in generic
2727 lustre code to use it.
2729 Severity : enhancement
2730 Bugzilla : 9477, 9557, 9870
2731 Description: Verify that the MDS configuration logs are updated when xml is
2732 Details : Check if the .xml configuration logs are newer than the config
2733 logs stored on the MDS and report an error if this is the case.
2734 Request --write-conf, or allow starting with --old_conf.
2736 Severity : enhancement
2738 Description: Handle symlinks in the path when checking if Lustre is mounted.
2739 Details : Resolve intermediate symlinks when checking if a client has
2740 mounted a filesystem to avoid duplicate client mounts.
2745 Description: lconf can hit an error exception but still return success.
2746 Details : The lconf command catches the Command error exception at the top
2747 level script context and will exit with the associated exit
2748 status, but doesn't ensure that this exit status is non-zero.
2753 Description: failure of ptlrpc thread startup can cause oops
2754 Details : Starting a ptlrpc service thread can fail if there are a large
2755 number of threads or the server memory is very fragmented.
2756 Handle this without oopsing.
2759 Frequency : always, only if liblustre and non-default acceptor port was used
2761 Description: liblustre cannot connect to servers with non-default acceptor port
2762 Details : tcpnal_set_default_params() was not called and was therefore
2763 ignoring the environment varaible TCPNAL_PORT, as well as other
2764 TCPNAL_ environment variables
2769 Description: two objects could be created on the same OST for a single file
2770 Details : If an OST is down, in some cases it was possible to create two
2771 objects on a single OST for a single file. No problems other
2772 than potential performance impact and spurious error messages.
2776 Bugzilla : 5681, 9562
2777 Description: Client may oops in ll_unhash_aliases
2778 Details : Client dcache may become inconsistent in race condition.
2779 In some cases "getcwd" can fail if the current directory is
2785 Description: Inode refcounting problems in NFS export code
2786 Details : link_raw functions used to call d_instantiate without obtaining
2787 extra inode reference first.
2791 Bugzilla : 9942, 9903
2792 Description: Referencing freed requests leading to crash, memleaks with NFS.
2793 Details : We used to require that call to ll_revalidate_it was always
2794 followed by ll_lookup_it. Also with revalidate_special() it is
2795 possible to call ll_revalidate_it() twice for the same dentry
2796 even if first occurence returned success. This fix changes semantic
2797 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2798 reference on a request referred from the intent.
2799 ll_intent_release() then releases such a request.
2802 Frequency : rare, normally benchmark loads only
2804 Description: unlinked inodes were kept in memory on the client
2805 Details : If a client is repeatedly creating and unlinking files it
2806 can accumulate a lot of stale inodes in the inode slab cache.
2807 If there is no other client load running this can cause the
2808 client node to run out of memory. Instead flush old inodes
2809 from client cache that have the same inode number as a new inode.
2812 Frequency : SLES9 2.6.5 kernel and long filenames only
2813 Bugzilla : 9969, 10379
2814 Description: utime reports stale NFS file handle
2815 Details : SLES9 uses out-of-dentry names in some cases, which confused
2816 the lustre dentry revalidation. Change it to always use the
2820 Frequency : rare, unless heavy write-truncate concurrency is continuous
2821 Bugzilla : 4180, 6984, 7171, 9963, 9331
2822 Description: OST becomes very slow and/or deadlocked during object unlink
2823 Details : filter_destroy() was holding onto the parent directory lock
2824 while truncating+unlinking objects. For very large objects this
2825 may block other threads for a long time and slow overall OST
2826 responsiveness. It may also be possible to get a lock ordering
2827 deadlock in this case, or run out of journal credits because of
2828 the combined truncate+unlink. Solution is to do object truncate
2829 first in one transaction without parent lock, and then do the
2830 final unlink in a new transaction with the parent lock. This
2831 reduces the lock hold time dramatically.
2834 Frequency : rare, 2.4 kernels only
2836 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2837 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2838 before unmounting their filesystem in order to avoid deadlock.
2839 The kernel_locked() function in 2.4 kernels only checks whether
2840 the kernel lock is held, not whether it is this process that is
2841 holding it as 2.6 kernels do.
2846 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2847 Details : The client ptlrpc code may be trying to reconnect to a down
2848 server before a previous connection attempt has timed out.
2849 Increase the reconnect interval to be longer than the connection
2850 timeout interval to avoid sending duplicate connections to
2854 Frequency : echo_client brw_test command
2856 Description: fix echo_client to work with OST preallocated code
2857 Details : OST preallocation code (5137) didn't take echo_client IO path
2858 into account: echo_client calls filter methods outside of any
2859 OST thread and, hence, there is no per-thread preallocated
2860 pages and buffers to use. Solution: hijack pga pages for IO. As
2861 a byproduct, this avoids unnecessary data copying.
2865 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2866 Description: Client can oops in mdc_commit_close() after open replay
2867 Details : It was possible for the MDS to return an open request with no
2868 transaction number in mds_finish_transno() if the client was
2869 evicted, but without actually returning an error. Clients
2870 would later try to replay that open and may trip an assertion
2871 Simplify the client close codepath, and always return an error
2872 from the MDS in case the open is not successful.
2875 Frequency : rare, 2.6 OSTs only
2877 Description: OST may deadlock under high load on fragmented files
2878 Details : If there was a heavy load and highly-fragmented OST filesystems
2879 it was possible to have all the OST threads deadlock waiting on
2880 allocation of biovecs, because the biovecs were not released
2881 until the entire RPC IO was completed. Instead, release biovecs
2882 as soon as they are complete to ensure forward IO progress.
2884 Severity : enhancement
2886 Description: Support for specifying external journal device at mount
2887 Details : If an OST or MDS device is formatted with an external journal
2888 device, this device major/minor is stored in the ext3 superblock
2889 and may not be valid for failover. Allow detecting and
2890 specifying the external journal at mount time.
2895 Description: Mounting an MDS with pending unlinked files may cause oops
2896 Details : target_finish_recovery() calls mds_postrecov() which returned
2897 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2898 considers this an error and immediately begins cleaning up the
2899 lov, just after starting the mds_lov process
2901 Severity : enhancement
2903 Description: Implement 'lfs df' to report actual free space on per-OST basis
2904 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2905 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2906 usage of objects; '-h' to report in human readable format.
2908 ------------------------------------------------------------------------------
2910 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2917 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2918 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2919 at startup during orphan unlinking. Instead, skip buddy update
2920 before setup, as it will be regenerated after recovery is complete.
2925 Description: If an OST is inactive, its locks might reference stale inodes.
2926 Details : lov_change_cbdata() must iterate over all namespaces, even if
2927 they are inactive to clear inode references from the lock.
2929 Severity : enhancement
2930 Frequency : occasional, if non-standard max_dirty_mb used
2932 Description: Client will block write RPCs if not enough grant
2933 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2934 then the client will block writes while waiting for another RPC
2935 to complete instead of consuming its dirty limit. With change
2936 we get improved performance when max_dirty_mb is small.
2938 Severity : enhancement
2939 Bugzilla : 3389, 6253
2940 Description: Add support for supplementary groups on the MDS.
2941 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2942 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2943 lookups for user supplementary groups into a cache.
2947 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2948 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2949 for liblustre O_EXCL is enforced by MDS.
2953 Description: readdir never returns NULL in liblustre.
2954 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2955 dentry in struct dirent.
2959 Frequency : liblustre only, depends on application IO pattern
2960 Description: liblustre clients evicted if not contacting servers
2961 Details : Don't put liblustre clients into the ping_evictor list, so
2962 they will not be evicted by the pinger ever.
2964 Severity : enhancement
2966 Description: Add ability to evict clients by NID from MDS.
2967 Details : By echoing "nid:$NID" string into
2968 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2969 $NID would be instantly evicted from this MDS and from all active
2970 OSTs connected to it.
2974 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2975 Details : lookup_it_finish() used to query file size from OSTs that was not
2980 Description: service threads change working directory to that of init
2981 Details : Starting lustre service threads may pin the working directory
2982 of the parent thread, making that filesystem busy. Threads
2983 now change to the working directory of init to avoid this.
2987 Frequency : during shutdown only
2988 Description: shutdown with a failed MDS or OST can cause unmount to hang
2989 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2994 Frequency : 2.6 only
2995 Description: chmod/chown may include an extra supplementary group
2996 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2997 supplementary group and if none is specified this is used.
3000 Bugzilla : 5479 (6816)
3002 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3003 Details : If lookup is in progress on a file that is unlinked we might try
3004 to revalidate the inode and fail in revalidate after lookup is
3005 complete and ll_file_open() enqueues the open again but
3006 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3009 Frequency : always, if lconf --abort_recovery used
3011 Description: lconf --abort_recovery fails with 'Operation not supported'
3012 Details : lconf was attempting to abort recovery on the MDT device and not
3015 ------------------------------------------------------------------------------
3017 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3022 Frequency : rare (only unsupported configurations with a node running as an
3024 Bugzilla : 6514, 5137
3025 Description: Mounting a Lustre file system on a node running as an OST could
3027 Details : OSTs now preallocates memory needed to write out data at
3028 startup, instead of when needed, to avoid having to
3029 allocate memory in possibly low memory situations.
3030 Specifically, if the file system is mounted on on OST,
3031 memory pressure could force it to try to write out data,
3032 which it needed to allocate memory to do. Due to the low
3033 memory, it would be unable to do so and the node would
3034 become unresponsive.
3036 Severity : enhancement
3038 Description: Addition of lconf --service command line option
3039 Details : lconf now accepts a '--service <arg>' option, which is
3040 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3042 Severity : enhancement
3044 Description: Failover mode is now the default for OSTs.
3045 Details : By default, OSTs will now run in failover mode. To return to
3046 the old behaviour, add '--failout' to the lmc line for OSTs.
3048 Severity : enhancement
3050 Description: Health checks are now provided for MDS and OSTs
3051 Details : Additional detailed health check information on MSD and OSTs
3052 is now provided through the procfs health_check value.
3055 Frequency : occasional, depends on IO load
3057 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3058 after numerous create/delete cycles
3059 Details : The ext3 inode allocation policy would not allocate new inodes
3060 very well on the OSTs because there are no new directories
3061 being created. Instead we look for groups with free space if
3062 the parent directories are nearly full.
3067 Description: Network or server problems during mount may cause partially
3068 mounted clients instead of returning an error.
3069 Details : The config llog parsing code may overwrite the error return
3070 code during mount error handling, returning success instead
3076 Description: MDS can fail to allocate large reply buffers
3077 Details : After long uptimes the MDS can fail to allocate large reply
3078 buffers (e.g. zconf client mount config records) due to memory
3079 fragmentation or consumption by the buffer cache. Preallocate
3080 some large reply buffers so that these replies can be sent even
3081 under memory pressure.
3085 Frequency : rare (liblustre)
3086 Description: fsx running with liblustre complained that using truncate() to
3087 extend the file doesn't work. This patch corrects that issue.
3088 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3089 ATTR_SIZE and lsm use in llu_setattr_raw.
3093 Frequency : rare, only 2.6 kernels
3094 Description: Unusual file access patterns on the MDS may result in inode
3095 data being lost in very rare circumstances.
3096 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3097 code in the 2.6 kernel caused the inode and/or EA data not to
3098 be read from disk, causing single-file corruption.
3102 Frequency : rare, only 2.6 filesystems using extents
3103 Description: Heavy concurrent write and delete load may cause data corruption.
3104 Details : It was possible under high-load situations to have an extent
3105 metadata block in the block device cache from a just-unlinked
3106 file overwrite a newly-allocated data block. We now unmap any
3107 metadata buffers that alias just-allocated data blocks.
3111 Frequency : filesystems with default stripe_count larger than 77
3112 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3113 Details : lconf specifies an inode size of 4096 bytes when the default
3114 stripe_count is larger than 77. This conflicts with the default
3115 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3116 this case to avoid pinning too much memory for large EAs.
3118 ------------------------------------------------------------------------------
3120 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3125 Frequency : rare (extremely heavy IO load with hundreds of clients)
3127 Description: Client is evicted, gets IO error writing to file
3128 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3129 caused clients to be evicted for AST timeouts. The fixes in
3130 bug 5192 mean we no longer need to have such short AST timeouts
3131 so ldlm_timeout has been increased.
3134 Frequency : occasional during --force or --failover shutdown under load
3135 Bugzilla : 5949, 4834
3136 Description: Server oops/LBUG if stopped with --force or --failover under load
3137 Details : a collection of import/export refcount and cleanup ordering
3138 issues fixed for safer force cleanup
3141 Frequency : only filesystems larger than 120 OSTs
3142 Bugzilla : 5990, 6223
3143 Description: lfs getstripe would oops on a very large filesystem
3144 Details : lov_getconfig used kfree on vmalloc'd memory
3147 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3148 Bugzilla : 6242, 6243
3149 Description: reading from files that had been truncated to a non-zero size
3150 but never opened returned no data
3151 Details : ll_file_read() reads zeros from no-object files to EOF
3156 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3157 Details : There's little harm in aborting MDS/OSS recovery and letting it
3158 try again, so I removed the LASSERT and return an error instead.
3160 Severity : enhancement
3162 Description: New debugging infrastructure for tracking down data corruption
3163 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3164 (b) cover more of the client-side code path, and (c) try to narrow
3165 down where problems occurred
3169 Bugzilla : 3819, 4364, 4397, 6313
3170 Description: Racing close and eviction MDS could cause assertion in mds_close
3171 Details : It was possible to get multiple mfd references during close and
3172 client eviction, leading to one thread referencing a freed mfd.
3174 Severity: : enhancement
3175 Bugzilla : 3262, 6359
3176 Description: Attempts to reconnect to servers are now more aggressive.
3177 Details : This builds on the enhanced upcall-less recovery that was added
3178 in 1.4.2. When trying to reconnect to servers, clients will
3179 now try each server in the failover group every 10 seconds. By
3180 default, clients would previously try one server every 25 seconds.
3185 Description: After recovery, certain operations trigger a failed
3186 assertion on a client.
3187 Details : Failing over an mds, using lconf -d --failover, while a
3188 client was doing a readdir() call would cause the client to
3189 LBUG after recovery completed and the readdir() was resent.
3191 Severity : enhancement
3193 Description: Default groups are now added by lconf
3194 Details : You can now run lconf --group <servicename> without having to
3195 manually add groups with lmc.
3198 Frequency : occasional
3200 Description: Nodes with an elan id of 0 trigger a failed assertion
3203 Frequency : always when accessing e.g. tty/console device nodes
3205 Description: tty and some other devices nodes cannot be used on lustre
3206 Details : file's private_data field is used by device data and lustre
3207 values in there got lost. New field was added to struct file to
3208 store fs-specific private data.
3211 Frequency : when exporting Lustre via NFS
3213 Description: NFSD failed occasionally when looking up a path component
3214 Details : NFSD is looking up ".." which was broken in ext3 directories
3215 that had grown large enough to become hashed.
3218 Frequency : Clusters with multiple interfaces not on the same subnet
3220 Description: Nodes will repeatedly try to reconnect to an interface which it
3221 cannot reach and report an error to the log.
3222 Details : Extra peer list entries will be created by lconf with some peers
3223 unreachable. lconf now validates the peer before adding it.
3226 Frequency : Only if a default stripe is set on the filesystem root.
3228 Description: Setting a default stripe on the filesystem root prevented the
3229 filesystem from being remounted.
3230 Details : The client was sending extra request flags in the root getattr
3231 request and did not allocate a reply buffer for the dir EA.
3234 Frequency : occasional, higher if lots of files are accessed by one client
3235 Bugzilla : 6159, 6097
3236 Description: Client trips assertion regarding lsm mismatch/magic
3237 Details : While revalidating inodes the VFS looks up inodes with ifind()
3238 and in rare cases can find an inode that is being freed.
3239 The ll_test_inode() code will free the lsm during ifind()
3240 when it finds an existing inode and then the VFS later attaches
3241 this free lsm to a new inode.
3245 Bugzilla : 6422, 7030
3246 Description: MDS deadlock between mkdir and client eviction
3247 Details : Creating a new file via mkdir or mknod (starting a transaction
3248 and getting the ns lock) can deadlock with client eviction
3249 (gets ns lock and trying to finish a synchronous transaction).
3252 Frequency : occasional
3253 Description: While starting a server, the fsfilt_ext3 module could not be
3255 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3256 kernels. Previously, lconf would still use the ext3 name
3257 when trying to load modules. Now, it will correctly use
3258 ext3 on 2.4 and ldiskfs on 2.6.
3260 Severity : enhancement
3261 Description: The default stripe count has been changed to 1
3262 Details : The interpretation of the default stripe count (0, to lfs
3263 or lmc) has been changed to mean striping across a single
3264 OST, rather than all available. For general usage we have
3265 found a stripe count of 1 or 2 works best.
3267 Severity : enhancement
3268 Description: Add support for compiling against Cray portals.
3269 Details : Conditional compiling for some areas that are different
3273 Frequency : occasional
3274 Bugzilla : 6409, 6834
3275 Description: Creating files with an explicit stripe count may lead to
3276 a failed assertion on the MDS
3277 Details : If some OSTs are full or unavailable, creating files may
3278 trigger a failed assertion on the MDS. Now, Lustre will
3279 try to use other servers or return an error to the
3283 Frequency : occasional
3285 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3286 caused lock timeout during readahead (since 1.4.2).
3287 Details : Processes doing readahead might match a lock that hasn't been
3288 granted yet if there are overlapping and conflicting lock
3289 requests. The readahead process waits on ungranted lock
3290 (original lock is CBPENDING), while OST waits for that process
3291 to cancel CBPENDING read lock and eventually evicts client.
3293 Severity : enhancement
3295 Description: Initial enabling of flock support for clients
3296 Details : Implements fcntl advisory locking and file status functions.
3297 This feature is provided as an optional mount flag (default
3298 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3299 locking are implemented yet, and those that are are not guaranteed
3300 to be completely correct in production environments.
3301 mount -t lustre -o [flock|noflock] ...
3304 Frequency : occasional
3306 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3307 assertion in the ext3 JBD (journaling) layer.
3308 Details : The b_committed_data struct is protected by the big kernel lock
3309 in 2.4 kernels, serializing journal_commit_transaction() and
3310 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3311 access to this struct. In 2.6 kernels there is finer grained
3312 locking to improve SMP performance of the JBD layer.
3316 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3321 Description: merge and cleanup kernel patches.
3324 -----------------------------------------------------------------------------
3326 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3328 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3329 versions of Lustre. Please update all servers and clients to
3330 version 1.4.2 or later at the same time. You must also run
3331 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3332 to update the configuration logs.
3334 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3335 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3336 - swab llog records of type '0' so we get proper header size/idx (5861)
3337 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3338 - fix rename of one directory over another leaking an inode (5953)
3339 - avoid SetPageDirty on 2.6 (5981)
3340 - don't re-add just-being-destroyed locks to the waiting list (5653)
3341 - when creating new directories, inherit the parent's custom
3342 striping settings if present parent (3048)
3343 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3344 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3345 - don't hold client locks on temporary worklist from l_lru (5666)
3346 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3347 - automatically evict dead clients (5921)
3348 - Update file size properly in create+truncate+fstat case (6196)
3349 - Do not unhash mountpoint dentries, do not allow removal of
3351 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3352 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3353 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3354 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3355 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3356 - Correctly handle reads of files with no objects (6243)
3357 - lctl recover will also mark a device active if deactivate used (5933)
3359 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3360 - allow --write-conf on an MDS with different nettype than client (5619)
3361 - don't write config llogs to MDS for mounts not from that MDS (5617)
3362 - lconf should create multiple TCP connections from a client (5201)
3363 - init scripts are now turned off by default; run chkconfig --on
3364 lustre and chkconfig --on lustrefs to use them
3365 - upcalls are no longer needed for clients to recover to failover
3367 - add --abort-recovery option to lconf to abort recovery on device
3369 - add support for an arbitrary number of OSTs (3026)
3370 - Quota support protocol changes.
3371 - forward compatibility changes to wire structs (6007)
3372 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3373 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3374 - improved "lustre status" script
3375 - initialize blocksize for non-regular files (6062)
3376 - added --disable-server and --disable-client configure options (5782)
3377 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3378 - Vanilla 2.4.29 support
3379 - increase maximum number of obd devices to 520 (6242)
3380 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3381 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3382 SLES 9 2.6.5 kernels
3383 - limit stripes per file to 160 (the maximum EA size) (6093)
3385 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3388 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3389 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3390 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3391 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3392 - small ext3 extents cleanups and fixes (5733)
3393 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3394 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3395 - avoid needless client->OST connect, fix handle mismatch (5317)
3396 - fix DLM error path that led to out-of-sync client, long delays (5779)
3397 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3398 - fix several locking issues related to i_size (5492,5624,5654,5672)
3399 - don't move pending lock onto export if it is already evicted (5683)
3400 - fix kernel oops when creating .foo in unlinked directory (5548)
3401 - fix deadlock in obdfilter statistics vs. object create (5811)
3402 - use time_{before,after} to avoid timer jiffies wrap (5882)
3403 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3404 - Do not leak request if server was not able to process it (5154)
3405 - If mds_open unable to find parent dir, make that negative lookup(5154)
3406 - don't create new directories with extent-mapping (5909, 5936)
3408 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3409 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3410 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3411 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3412 - support for RHEL4 kernel on i686 (5773)
3413 - provide error messages when incompatible logs are encountered (5898)
3415 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3416 * version 1.4.0.10 (1.4.1 release candidate 1)
3418 - don't keep a lock reference when lock is not granted (4238)
3419 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3420 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3421 - drop import inflight refcount on signal_completed_replay error (5255)
3422 - unlock page after async write error during send (3677)
3423 - handle missing objects in filter_preprw_read properly (5265)
3424 - no transno return for symlink open, don't save no-trasno open (3440)
3425 - don't try to complete elan receive that already failed (4012)
3426 - free RPC server reply state on error (5406)
3427 - clean up thread from ptlrpc_start_thread() on error (5160)
3428 - readahead could read extra page into cache that wasn't ejected (5388)
3429 - prevent races in class_attach/setup/cleanup/detach (5260)
3430 - don't dereference de->d_inode after l_dput of de (5458)
3431 - use "int" for stripe value returned from lock_to_stripe (5544)
3432 - mballoc allocation and error-checking fixes in 2.6 (5504)
3433 - block device patches to fix I/O request sizes in 2.6 (5482)
3434 - look up hostnames for IB nals (5602)
3435 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3436 - don't start multiple acceptors for the same port (5277)
3437 - fix incorrect LASSERT in mds_getattr_name (5635)
3438 - export a proc file for general "ping" checking (5628)
3439 - fix "lfs check" to not block when the MDS is down (5628)
3441 - service request history (4965)
3442 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3443 - create an "evict_client" /proc entry on OSTs, like the MDS has
3444 - fix mount usage message, return errors per mount(8) (5168)
3445 - change grep [] to grep "[]" in tests so they work in more UMLs
3446 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3447 - remove ancient LOV_MAGIC_V0 EA support (5047)
3448 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3449 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3450 - allow readahead allocations to fail when low on memory (5383)
3451 - mmap locking landed again, after considerable improvement (2828)
3452 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3454 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3457 - send OST transaction number in read/write reply to free req (4966)
3458 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3459 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3460 - the watchdog thread now runs as interruptible (5246)
3461 - flock/lockf fixes (but it's still disabled, pending 5135)
3462 - don't use EXT3 constants in llite code (5094)
3463 - memory shortage at startup could cause assertion (5176)
3465 - reorganization of lov code
3466 - single portals codebase
3468 - add extents/mballoc support (5025)
3469 - direct I/O reads in the obdfilter (4048)
3470 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3472 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3475 - send OST transaction number in read/write reply to free req (4966)
3476 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3477 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3478 - the watchdog thread now runs as interruptible (5246)
3479 - handle missing objects in filter_preprw_read properly (5265)
3480 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3481 - drop import inflight refcount on signal_completed_replay error (5255)
3482 - unlock page after async write error during send (3677)
3483 - return original error code on reconstructed replies (3761)
3484 - no transno return for symlink open, don't save no-trasno open (3440)
3486 - add pid to ldlm debugging output (4922)
3487 - bump the watchdog timeouts -- we can't handle 30sec yet
3488 - extra debugging for orphan dentry/inode bug (5259)
3490 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3493 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3494 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3495 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3496 - fix NULL dereference in /proc/sys/portals/routes (4827)
3497 - allow failed mdc_close() operations to be interrupted (4561)
3498 - stop precreate on OST before MDS would time out on it (4778)
3499 - don't send partial-page writes before EOF from client (4410)
3500 - discard client grant for sub-page writes on large-page clients (4520)
3501 - don't free dentries not owned by NFS code, check generation (4806)
3502 - fix lsm leak if mds_create_objects() fails (4801)
3503 - limit debug_daemon file size, always print CERROR messages (4789)
3504 - use transno after validating reply (3892)
3505 - process timed out requests if import state changes (3754)
3506 - update mtime on OST during writes, return in glimpse (4829)
3507 - add mkfsoptions to LDAP (4679)
3508 - use ->max_readahead method instead of zapping global ra (5039)
3509 - don't interrupt __l_wait_event() during strace
3511 - add software watchdogs to catch hung threads quickly (4941)
3512 - make lustrefs init script start after nfs is mounted
3513 - fix CWARN/ERROR duplication (4930)
3514 - return async write errors to application if possible (2248)
3515 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3516 - print NAL number in %x format (4645)
3517 - update barely-supported suse-2.4.21-171 series (4842)
3518 - support for sles 9 %post scripts
3519 - support for building 2.6 kernel-source packages
3520 - support for sles km_* packages
3522 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3525 - ignore -ENOENT errors in osc_destroy (3639)
3526 - notify osc create thread that OSC is being cleaned up (4600)
3527 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3528 - reconstruct ost_handle() like mds_handle() (4657)
3529 - create a new thread to do import eviction to avoid deadlock (3969)
3530 - let lconf resolve symlinked-to devices (4629)
3531 - don't unlink "objects" from directory with default EA (4554)
3532 - hold socknal file ref over connect in case target is down (4394)
3533 - allow more than 32000 subdirectories in a single directory (3244)
3534 - fix blocks count for O_DIRECT writes (3751)
3535 - OST returns ENOSPC from object create when no space left (4539)
3536 - don't send truncate RPC if file size isn't changing (4410)
3537 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3538 - bind to privileged port in socknal and tcpnal (3689)
3540 - rate limit CERROR/CWARN console message to avoid overload (4519)
3541 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3542 - basic mmap support (3918)
3543 - kernel patch series update from b1_4 (4711)
3545 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3548 - avoid crash during MDS cleanup with OST shut down (2775)
3549 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3550 - don't use bad inodes on the MDS (3744)
3551 - dynamic object preallocation to improve recovery speed (4236)
3552 - don't hold spinlock over lock dumping or change debug flags (4401)
3553 - don't zero obd_dev when it is force cleaned (3651)
3554 - print grants to console if they go negative (4431)
3555 - "lctl deactivate" will stop automatic recovery attempts (3406)
3556 - look for existing locks in ldlm_handle_enqueue() (3764)
3557 - don't resolve lock handle twice in recovery avoiding race (4401)
3558 - revalidate should check working dir is a directory (4134)
3560 - don't always mark "slow" obdfilter messages as errors (4418)
3562 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3565 - don't close LustreDB during write_conf until it is done (3860)
3566 - fix typo in lconf for_each_profile (3821)
3567 - allow dumping logs from multiple threads at one time (3820)
3568 - don't allow multiple threads in OSC recovery (3812)
3569 - fix debug_size parameters (3864)
3570 - fix mds_postrecov to initialize import for llog ctxt (3121)
3571 - replace config semaphore with spinlock (3306)
3572 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3573 - don't allow enqueue to complete on a destroyed export (3822)
3574 - down write_lock before checking llog header bitmap (3825)
3575 - recover from lock replay timeout (3764)
3576 - up llog sem before sending rpc (3652)
3577 - reduce ns lock hold times when setting kms (3267)
3578 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3579 - fix NULL deref and obd_dev leak on setup error (3312)
3580 - replace some LBUG about llog ops with error handling (3841)
3581 - don't match INVALID dentries from d_lookup and spin (3784)
3582 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3583 - fix invalid assertion in ptlrpc_set_wait (3880)
3585 - add libwrap support for the TCP acceptor (3996)
3586 - add /proc/sys/portals/routes for non-root route listing (3994)
3587 - allow setting MDS UUID in .xml (2580)
3588 - print the stack of a process that LBUGs (4228)
3590 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3593 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3594 - make sure to unset replay flag from failed open requests (3440)
3595 - if default stripe count is 0, use OST count for inode size (3636)
3596 - update parent mtime/ctime on client for create/unlink (2611)
3597 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3598 - free recovery state on server during a forced cleanup (3571)
3599 - unregister_reply for resent reqs (3063)
3600 - loop back devices mounting and status check on 2.6 (3563)
3601 - fix resource-creation race that can provoke i_size == 0 (3513)
3602 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3603 - more debugging for page-accounting assertion (3746)
3604 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3605 - avoid deadlock after precreation failure (3758)
3606 - fix race and lock order deadlock in orphan handling (3450, 3750)
3607 - add validity checks when grabbing inodes from l_ast_data (3599)
3609 - add /proc/.../recovery_status to obdfilter (3428)
3610 - lightweight CDEBUG infrastructure, debug daemon (3668)
3611 - change default OSC RPC parameters to be better on small clusters
3612 - turn off OST read cache for files smaller than 32MB
3613 - install man pages and include them in rpms (3100)
3614 - add new init script for (un)mounting lustre filesystems (2593)
3615 - run chkconfig in %post for init scripts (3701)
3616 - drop scimac NAL (unmaintained)
3618 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3621 - clean kiobufs before and after use (3485)
3622 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3623 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3624 - change init script to fail more clearly if not run as root (1528)
3625 - allow clients to reconnect during replay (1742)
3626 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3627 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3628 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3629 - fix cond_resched (3554)
3630 - only evict unfinished clients after recovery (3515)
3631 - allow bulk resend, prevent data loss (3570)
3632 - dynamic ptlrpc request buffer allocation (2102)
3633 - don't allow unlinking open directory if it isn't empty (2904)
3634 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3635 - remove extraneous obd dereference causing LASSERT failure (3334)
3636 - don't use get_cycles() when creating temp. files on the mds (3156)
3637 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3638 - handle EEXIST for set-stripe, set proper directory name (3336)
3640 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3641 - fix ksocknal_fmb_callback() error messages (2918)
3643 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3646 - don't copy lvb into (possibly NULL) reply on error (2983)
3647 - don't deref dentry after dput, don't free lvb on error (2922)
3648 - use the kms to determine writeback rpc length (2947)
3649 - increment oti_logcookies when osc is inactive (2948)
3650 - update client's i_blocks count via lvb messages (2543)
3651 - handle intent open/close of special files properly (1557)
3652 - mount MDS with errors=remount-ro, like obdfilter (2009)
3653 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3654 - don't use cancelling-locks' kms values (2947)
3655 - use highest lock extent for kms, not last one (2925)
3656 - don't dereference ERR_PTR() dentry in error handling path (3107)
3657 - fix thread race in portals_debug_dumplog() (3122)
3658 - create lprocfs device entries at setup instead of at attach (1519)
3659 - common AST error handler, don't evict client on completion race (3145)
3660 - zero nameidata in detach_mnt in 2.6 (3118)
3661 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3662 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3663 - initialize RPC timeout timer earlier for 2.6 (3219)
3664 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3665 - print nal/nid for unknown nid (3258)
3666 - additional checks for oscc recovery before doing precreate (3284)
3667 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3668 - don't crash in mdc_close for bad permissions on open (3285)
3669 - zero i_rdev for non-device files (3147)
3670 - clear page->private before handing to FS, better assertion (3119)
3671 - tune the read pipeline (3236)
3672 - fix incorrect decref of invalidated dentry (2350)
3673 - provide read-ahead stats and refine rpc in flight stats (3328)
3674 - don't hold journal transaction open across create RPC (3313)
3675 - update atime on MDS at close time (3265)
3676 - close LDAP connection when recovering to avoid server load (3315)
3677 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3678 - don't leak open file on MDS after open resend (3325)
3679 - serialize filter_precreate and filter_destroy_precreated (3329)
3680 - loop device shouldn't call sync_dev() for nul device (3092)
3681 - clear page cache after eviction (2766)
3682 - resynchronize MDS->OST in background (2824)
3683 - refuse to mount the same filesystem twice on same mountpoint (3394)
3684 - allow llmount to create routes for mounting behind routers (3320)
3685 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3686 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3687 - fix rare problem with rename on htree directories (3417)
3689 - allow default OST striping configuration per directory (1414)
3690 - fix compilation for qswnal for 2.6 kernels (3125)
3691 - increase maximum number of MDS request buffers for large systems
3692 - change liblustreapi to be useful for external progs like lfsck (3098)
3693 - increase local configuration timeout for slow disks (3353)
3694 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3696 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3699 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3700 - don't overwrite extent policy data in reply if lock was blocked (2901)
3701 - drop filter export grants atomically with removal from device (2663)
3702 - del obd_self_export from work_list in class_disconnect_exports (2908)
3703 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3704 - swab reply message in mdc_close, other PPC fixes (2464)
3705 - fix destroying of named logs (2325)
3706 - overwrite old logs when running lconf --write_conf (2264)
3707 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3708 - fix race in target_handle_connect (2898)
3709 - mds_reint_create() should take same inode create lock (2926)
3710 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3711 - don't close files for self_export to avoid uninitialized obd (2936)
3712 - allow MDS with the same name as client node (2939)
3713 - hold dentry reference for closed log files for unlink (2325)
3714 - reserve space for all logs during transactions (2059)
3715 - don't evict page beyond end of stripe extent (2925)
3716 - don't oops on a deleted current working directory (2399)
3717 - handle hard links to targets without a parent properly (2517)
3718 - don't dereference NULL lock when racing during eviction (2867)
3719 - don't grow lock extents when lots of conflicting locks (2919)
3721 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3724 - account for cache space usage on clients to avoid data loss (974)
3725 - lfsck support in lustre kernel code (2349)
3726 - reduce journal credits needed for BRW writes (2370)
3727 - orphan handling to avoid losing space on client/server crashes
3728 - ptlrpcd can be blocked, stopping ALL progress (2477)
3729 - use lock value blocks to assist in proper KMS, faster stat (1021)
3730 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3731 - recovery for initial connections (2355)
3732 - fixes for mds_cleanup_orphans (1934)
3733 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3734 - block all file creations until orphan recovery completes (1901)
3735 - client remove rq_connection from request struct (2423)
3736 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3737 - recovery timer race (2670)
3738 - mdc_close recovey bug (2532)
3739 - ptlrpc cleanup bug (2710)
3740 - mds timeout on local locks (2588)
3741 - namespace lock held during RPCs (2431)
3742 - handle interrupted sync write properly (2503)
3743 - don't try to handle a message that hasn't been replied to (2699)
3744 - client assert failure during cleanup after abort recovery (2701)
3745 - leak mdc device after failed mount (2712)
3746 - ptlrpc_check_set allows timedout requests to complete (2714)
3747 - wait for inflight reqs when ptlrpcd finishes (2710)
3748 - make sure unregistered services are removed from the srv_list
3749 - reset bulk XID's when resending them (caught by 1138 test)
3750 - unregister_bulk after timeout
3751 - fix lconf error (2694)
3752 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3753 - readahead locks pages, leaves pending causing memory pressure (2673)
3754 - increase OST request buffers to 4096 on large machines (2729)
3755 - fix up permission of existing directories in simple_mkdir (2661)
3756 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3757 - don't assign transno to errored transactions (2742)
3758 - don't delete objects on OST if given a bogus objid from MDS (2751)
3759 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3760 - if rq_no_resend, then timeout request after recovery (2432)
3761 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3762 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3763 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3764 - MDS recovery shouldn't delete objects that it has given out (2730)
3765 - if enqueue arrives after completion, don't clobber LVB (2819)
3766 - don't unlock pages twice when trigger_group_io returns error (2814)
3767 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3768 - don't write pages to disk if there was an error (1450)
3769 - don't ping imports that have recovery disabled (2676)
3770 - take buffered bytes into account when balancing socknal conn (2817)
3771 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3772 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3773 - remove little-endian swabbing of llog records (1987)
3774 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3775 - retry reposting request buffers if they fail (1191)
3776 - grow extent at grant time to avoid granting a revoked lock (2809)
3777 - lock revoke doesn't evict page if covered by a second lock (2765)
3778 - disable VM readahead to avoid reading outside lock extents (2805)
3780 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3781 - updated kernel patches for hp-2.4.20 kernel (2681)
3783 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3786 - fix truncated write corruption (2366)
3787 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3789 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3790 - ensure len(uuid) < 37 in lmc (1171)
3791 - fix ia64 OOPS in llog_test (2255)
3792 - zero end of page at obdfilter for partial page writes (2648)
3793 - don't leave stale dentries around after renames (bug 2428)
3794 - fix timeouts when evicting a client with a single lock held (2642)
3795 - set deadline for the initial HELLO message to drain (2634)
3796 - print out dotted-quad IP addresses in the socknal (2302)
3798 - additional debugging for MDS client eviction problem (2443)
3799 - fix mkfsoptions support for osts (2603, 2604)
3801 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3804 - add series for the vanilla 2.6.0 kernel
3805 - add series for the vanilla 2.4.24 kernel
3806 - add series for a cray x86/64 UL kernel drop
3807 - fix xattr patches for the vanilla 2.4.19 series
3809 - generate true UUIDs in lmc (1171)
3810 - have portals stack dumping break in UML (2466)
3811 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3812 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3813 - wake callback waiting threads on client eviction (2460)
3814 - Add --ptldebug and --subsystem to lmc (1719)
3815 - update assertion to allow safe interrupt allocation
3816 - set rq_no_resend for cancel requests (2432)
3817 - recalculate ptlrpcd timeout after resend (2494)
3818 - call vfs_rmdir when removing pending directories (2368)
3819 - fix renaming a file to itself (2429)
3820 - lmc creates a default one-stripe lov (2454)
3821 - expand procfs space to handle large clusters (2326)
3822 - increase UML stack to avoid overflow
3823 - update lconf's list of debug and subsystem masks
3824 - fix lfs find --obd (2510)
3825 - /proc tunable for disabling filter read caching (2591)
3826 - stop rpm packages from altering slapd.conf (2301)
3827 - disable nagle in the socknal under 0conf (2578)
3828 - choose mds inode size based on stripe count (2572)
3829 - fix kernel-source rpm problems (2516)
3831 - add --disable-doc to avoid pdf generation (2421)
3832 - update documentation, tests, type-os, comments
3833 - avoid format warnings on ia64
3834 - remove the TOE NAL
3835 - tiny code cleanups by removing unused fields
3837 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3840 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3841 - remove the most blatant lies from BUILDING file (2371)
3842 - change default debug level to reasonable production setting
3843 - reduce client side cache size to reduce cache flush time
3844 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3845 - make TCP zerocopy and pinger support enabled by default (2476)
3846 - sync writes completed after process exits caused crashes (2319)
3847 - maintain correct mount count on the MDS (2356)
3848 - backout 1557, because 2316 wasn't really fixed
3849 - better file I/O statistics gathering in /proc
3850 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3851 - another kernel patch to fix zero-copy TCP function export
3852 - don't take duplicate lock when processing re-sent getattr (2420)
3853 - lctl uses obd_self_export instead of creating new conn (2353)
3854 - MDS/OST recovery case which requires object creation asserted (2425)
3855 - move lfs from /usr/sbin to /usr/bin in packages
3856 - fix race between mds_client_add and mds_client_free (2417)
3857 - use kmalloc instead of slabs in portals (2430)
3858 - don't create duplicate records when a failover MDS is present (2442)
3859 - remove unnecessary mount age check (2332)
3860 - don't remove directory inodes from locks prematurely (2451)
3861 - don't break if MDS service name is the same as hostname (2103)
3862 - fix races in client write RPC generation when cache full (2482)
3864 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3867 - remove now-unused request->rq_obd (278)
3868 - if an allocation fails, print out how much memory we've used (1933)
3869 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3870 - add the "configurable stack size" patch to most series files (1256)
3871 - ability to write large log records, for 100+ OST configs (2306)
3872 - fix NULL deref when filter_prep fails (2314)
3873 - fix operator precedence error in filter_sync
3874 - dynamic allocation of socknal TX descriptors (2315)
3875 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3876 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3877 - ensure that CWARN messages in Portals always get to the syslog
3878 - __init/__exit are not for prototype decls (ldlm_init/exit)
3879 - x86-64 compile warning fixes
3880 - fix gateway LMC keyword conflict (2318)
3881 - fix MDS lock inversions in getattr/reint paths (1844)
3882 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3883 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3884 - DLM race condition prevented some lock evictions (2328)
3885 - ENOMEM detection and retry on socknal sends (2230)
3886 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3887 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3888 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3889 - better error messages when a client is rejected during recovery (1505)
3890 - avoid cancelling locks which were never granted, after failure (2330)
3891 - fix i_sem/journal inversion in mds_client_add (2333)
3892 - fix truncate/getattr lock cycle deadlock (2334)
3893 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3894 - fix two rare exit paths which could leak an l_lock() ref (2321)
3895 - fencepost error in MDS/OST orphan recovery (2226)
3896 - make log record alignment 8 bytes (1988)
3897 - lstripe now fails when requested offset > ost_count (2237)
3898 - ensure that all kernel series have a complete list.h (1607)
3899 - fix crashes in special-file operations (2316)
3900 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3902 - allow configurable automake binary, for testing new versions
3903 - small update to the lfs documentation
3905 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3907 * fix negative export reference count in fsfilt_sync (2312)
3909 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3910 * release candidate 0.9.1
3912 - orphans are moved into the PENDING directory for possible recovery
3913 - replayed opens now open by fid for orphan/rename safety (1042)
3914 - last close of an orphan inode generates a transno (683)
3915 - chdir() and mount() now pin the directory entry (1020)
3916 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3917 - discard very old requests without processing them (1502)
3918 - remove some common, well-understood CERRORs (1505)
3919 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3920 - clear "grant" flags in OST replies until OST grant code lands (1644)
3921 - fix read performance by not clobbering i_blksize on client (1598)
3922 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3923 - make LRU size a /proc tunable, clears locks when reduced (707)
3924 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3925 - server locks take references on exports now (1558)
3926 - build fixes for 2.4.20-rh trees (1663)
3927 - return an error from lov_create if all OSCs are inactive (1751)
3928 - fix import levels when a reconnect happens without a timeout (1597)
3929 - exit early from mds_open if we get a lookup error (1749)
3930 - partial page read at EOF wouldn't wait for disk before sending (1642)
3931 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3932 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3933 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3934 - write/truncate lock inversion (1639)
3935 - Don't auto-load obdclass, portals modules during cleanup (1495)
3936 - fix timestamps from jumping to "now" (1763)
3937 - extra journal assertions (1648)
3938 - add an extra multiunlink test (1771)
3939 - fix read_record/write_record API (1776)
3940 - fix leak of offset_extent, possible incorrect i_size later (1772)
3941 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3942 - Debugging for the kqswnal_get_idle_tx problems (1820)
3943 - Allow recovery to be attempted multiple times (1536)
3944 - Write out MDS last_rcvd file after it is first created (1600)
3945 - Fix tx_descriptor leak in failed transmit situations (1827)
3946 - ext3 journaling fixes for assertion failure after IO error (1871)
3947 - class_export_put() on freed export after completion AST error (1896)
3948 - Fix revalidate looping in VFS (1322)
3949 - Don't access a freed export during MDS_REINT timeout (1521)
3950 - Add open-unlink recovery support on the MDS (1673,1764)
3951 - Return an error if no MDS data was read from last_rcvd (1946)
3952 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3953 - Don't setup a disk device that doesn't match exported UUID (317)
3954 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3955 - avoid committing NULL handle in force close
3956 - local.sh is now a one-stripe LOV configuration
3957 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3958 - fix ext3 htree duplicate directory entry corruption (1516)
3959 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3960 - update to new LOV EA format (2097)
3961 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3963 2003-06-15 Phil Schwan <phil@clusterfs.com>
3966 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3967 - per-import/export recovery handling (958, 931, 959)
3968 - multiple last-rcvd slots, for serving multiple FSes (949)
3969 - connections are again shared between multiple imp/exports (963, 964)
3970 - "umount -f" would hang if any requests needed to be sent (393, 978)
3971 - avoid pinning large req buffer by copying for queued messages (989)
3972 - add "uuid" to "lctl device" command to help upcalls (991)
3973 - "open" RPCs with transnos would confuse recovery counters (1037)
3974 - do proper endian conversion of all wire messages (288, 340, 891)
3975 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3976 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3977 - fix ptlrpc_connection leak in target_handle_connect (1174)
3978 - fix import refcounting bug in OST and MDS cleanup (1134)
3979 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3980 - fix very unlikely obd_types race condition (501)
3981 - remove osc_open hack for echo_client (1187)
3982 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3983 - a failure in read_inode2 leads to deadlock (1139)
3984 - cancel ack-locks as soon as transaction is committed (1072)
3985 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3986 - make sure to commitrw after any preprw to avoid deadlock (1162)
3987 - failing to execute a file in a lustre FS would lock inode (1203)
3988 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3989 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3990 - obd_uuid2tgt would walk off the end of the list (1255)
3991 - on IA64 the osc would give portals incorrect bulk size (1258)
3992 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3993 - fix lock inversion caused by new llite matching code (1282)
3994 - limit the number of dirty pages on a client to 10MB (1286)
3995 - timed out locks were not being corrected cancelled (1289)
3996 - fix O_DIRECT above 4GB on IA-32 (1292)
3997 * major user-visible changes
3998 - fail out/fail over policy now controlled by the upcall (993)
4000 - add OBD_PING to check server availability and failure (954)
4001 - lustre messages are now sent in sending host order (288, 340, 891)
4002 - add eadatalen to MDS getattr reply (340)
4003 - OST read replies may contain second buffer, with per-page status (593)
4005 2003-03-11 Phil Schwan <phil@clusterfs.com>
4008 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4009 - failed open()s could cause deadlock; fixed (867, 869)
4010 - stop cancelling OST locks when files are closed (481)
4011 - overlapping XID spaces caused network corruption (851, 853)
4012 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4013 - setattr_raw added, to do single-RPC, server-side setattrs
4014 - lmc/lconf syntax change for OST UUIDs
4015 - fix crashy race condition between ptlrpc_free_req and osc_close
4016 - don't use request in mdc_enqueue if we hit a timeout (889)
4017 - don't set the inode i_size for regular files from the MDS (896)
4018 - handle out of order completion AST (842)
4019 - don't LBUG if a lock request times out after receiving AST (913)
4020 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4021 - if a bad lock AST arrives, send an error instead of dropping entirely
4022 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4023 - fix leak in bulk IO when only partially completed (899, 900, 926)
4024 - fix O_DIRECT for ia64 (55)
4025 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4026 - C-z after timeout could hang a process forever; fixed (977)
4028 - client-side I/O cache (678, 924, 929, 941, 970)
4030 - READPAGE and SETATTRs which don't take server-side locks get
4033 2003-02-11 Phil Schwan <phil@clusterfs.com>
4036 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4037 - Fix fsfilt_extN_readpage() to read a full page of directory
4038 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4039 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4040 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4041 - Most intent operations were removed, in favour of a new RPC mode
4042 that does a single RPC to the server and bypasses most of the VFS
4043 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4044 - Aggressively cancel local locks on DLM servers
4045 - mds_reint_unlink sends EA to the client if it's the last nlink.
4046 client uses that EA to unlink OST objects.
4047 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4048 - recursive symlinks were fixed (439)
4049 - fixed NULL deref in DEBUG_REQ
4050 - filter_update_lastobjid no longer calls sync, which annoyed extN
4051 - fixed multi-client small-writes to a single file problem (445)
4052 - fixed mtime updates during file writes (607)
4053 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4054 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4055 - fixed Linux OST size reporting problem (444, 656)
4056 - OST now updates object mtime with writes or setattr (607, 619)
4057 - client verifies file size before zeroing page past EOF (445)
4058 - OST now writes last allocated objid to disk with allocation (108)
4059 - LOV on echo now works (409)
4061 - mds_reint_unlink sends a new buffer, with the EA included. this
4062 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4063 set if a regular file was being unlinked, and it was the last link
4064 - use PtlGet from the target for bulk writes (315)
4065 - OST now updates object mtime with writes or setattr (607, 619)
4066 - LDLM now has a grant-time callback to revalidate locked items, if
4068 - Many MDS operations were reorganized to combat race conditions
4070 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4071 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4073 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4076 - Fully reactivate OST imports after reconnection (512, others)
4077 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4078 - More graceful error handling for truncating on dead OST (515)
4079 - Don't error out unless we're actually accessing dead stripes (474)
4080 - Fix garbage sizes when stripes are missing (410)
4081 - LRU counters were broken, causing constant lock purge (433, 432)
4082 - garbage on read from stripes with failed OSTs (441)
4083 - mark OSCs as active before reconnecting during recovery (438)
4084 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4085 - lfind did not preserve OST order in output (443)
4086 - symlinks cause hung clients, incorrect data (439)
4087 - stop dereferencing request after dropping refcount (457)
4088 - don't LASSERT(spin_is_locked) on non-SMP (455)
4089 - fixes for many rename() bugs
4090 - fstat didn't correctly synchronize attributes (399)
4091 - server must handle lock cancellation during blocking AST prep (487)
4092 - bulk descriptors were free()d too soon (511)
4093 - fix paths in lconf, which would load incorrect modules (451, 507)
4094 - fix confusing lconf 'host not found' error message (386)
4095 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4096 - fix race condition in mdc_blocking_ast() for inode access (526)
4097 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4098 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4099 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4100 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4101 - fix lstripe in several ways (kernel side) (527)
4102 - fix request leak in ldlm_cli_enqueue (262)
4103 - incorrect OSC was marked inactive after OST failure
4104 - call mds_fs_cleanup before unmounting filesystem (524)
4105 - fix races between taking ns_lock and ldlm_lock_change_resource
4106 - fix races updating LOV export open file list
4107 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4108 - fix recovery NULL deref in ldlm_cli_cancel_unused
4109 - fix some DLM races by using new hash table for lock handles (419)
4110 - permit the client to specify desired inodes, at replay
4111 - duplicate requests when we queue them for replay reintegration
4112 - fix last_rcvd offset calculation
4113 - sync after each recovered transaction, so we always make progress
4114 - never, not always, ERESTART requests without transnos
4115 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4117 - skip replay if the MDS says that the client is already connected
4118 - don't check for a recovery-enabled export to match lctl's UUID
4119 - don't INC_USE_COUNT for phantom exports
4120 - don't crash when cleaning up phantom exports (567)
4121 - don't double-finish or set replay data for errored mdc_open requests
4122 - abort requests when they time out, so we don't get old replies
4123 - send/receive replies for AST messages again
4124 - if the client says that it doesn't have the lock, cancel it on the
4126 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4127 mark it as destroyed, it will all work out when decref is called
4128 - fix module use counts (22, 581)
4130 - ASTs now expect a reply (server cancels lock on error reply)
4132 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4135 - fix many simultaneous client startup (392)
4136 - fix dentry->d_it clobbering
4137 - credentials weren't being shipped for readdir/getattr operations
4138 - remove invalid assertions triggered during some concurrent MD
4140 - proper Lustre versions added (336, 389)
4141 - fix memory leak for create error case (398)
4142 - fix LOV locking bug that would get cli/srv out of sync
4143 - fix echo client over LOV (409)
4144 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4145 - fix double-O_EXCL intent crash (424)
4146 - avoid sending multiple lock CANCELs (352)
4148 - MDS can do multi-client recovery (modulo bugs in new code)
4150 - many updates, edits, cleanups
4152 2002-11-18 Phil Schwan <phil@clusterfs.com>
4155 - fix null d_it dereference (346)
4156 - fix full OST/dbench hang (333)
4157 - fix permission problem with file removal (286)
4158 - fix removal of OSCs from LOV when they fail
4159 - fix NULL deref during bulk timeout (214)
4160 - fix problems related to multiple filesystems on one MDS (241)
4161 - fixed serious subtle metadata locking bugs
4162 - free locks on clients when inodes are removed due to memory
4164 - fix inode pointer in lock data (285)
4165 - partial support for multiple MDS on a single host (241)
4166 - data locks weren't cancelled at clear_inode time (290, 311)
4167 - intent locks could lead to unbounded lock growth (205)
4168 - added a maximum lock count, an LRU list, and a flusher
4169 - fix multiple rename (365)
4170 - properly abstracted the echo client
4171 - OSC locked 1 byte too many; fixed
4172 - rewrote brw callback code:
4173 - fixed recovery bugs related to LOVs (306)
4174 - fixed too-many-pages-in-one-write crash (191)
4175 - fixed (again) crash in sync_io_timeout (214)
4176 - probably fixed callback-related race (385)
4178 - Add capability to MDS protocol
4179 - LDLM cancellations and callbacks on different portals
4181 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4184 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4185 - defer open object destruction to close time (601981, 138)
4186 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4187 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4188 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4189 - don't call obd_setattr in ll_file_release for destroyed objects
4190 * protocol change to lustre_msg: move |version| and add |flags|
4191 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4192 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4193 * added replay of create, unlink, link and rename operations during
4194 MDS failover; recovery should be much more robust now
4195 * remove failed OSCs from LOVs (only lov_create uses this so far)
4196 * the lustre-HOWTO was brought (more) up to date (582544)
4198 2002-10-23 Phil Schwan <phil@clusterfs.com>
4201 - in-use dentries weren't being reused properly (617851)
4202 - prevent multiple LDLM setup (599178)
4203 - fix LOV size calculations for truncate (617853)
4204 - fix client handling of MDS intent errors (POSIX)
4205 - fix permission bug in lovstripe.c test (624321)
4206 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4207 - truncate past end of file could corrupt data
4208 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4209 - a race in recovery could return ETIMEDOUT to apps (623947)
4210 - building outside the source directory was fixed
4211 * the lustre-HOWTO was brought (more) up to date (582544)
4212 * major progress was made on recovery functionality
4214 2002-10-10 Phil Schwan <phil@clusterfs.com>
4217 - recovery deadlock fix
4218 - rm -rf causes LBUG fix (617817)
4219 - file open by multiple tasks fix (618962)
4220 - directory permissions bugs (602707 and 620007)
4221 - journal_stop fixed with locking (611313)
4222 - O_APPEND failures resolved (618273, perhaps 614459)
4223 - lconf PATH fix (619770)
4224 - IA64 build fix (621450)
4225 - RPC buffer sizes scale with amount of memory
4227 2002-10-01 Phil Schwan <phil@clusterfs.com>
4230 - locks would be cancelled without throwing away data pages,
4231 resulting in inconsistent data (605627)
4232 - inode attributes were not always being refreshed (605627, 612449)
4233 - lconf now continues to cleanup after lctl reports an error
4234 - MDS now enforces user permissions (602707)
4235 - lprocfs cleanup fixed, but not yet enabled (614157)
4236 - fixed infinite server hang, should a client not respond to an AST
4237 - avoid going into recovery if user calls readlink() with a buffer
4238 that's too small (613941)
4239 - AST RPCs no longer require replies (614867) -- this may be changed
4240 - don't crash server if client sends an IOV that's too big (611336)
4241 - fixed lock conversion deadlock (611892)
4242 - fixed the following of symlinks (614622)
4243 * recovery: the server can remove locks from a client that dies, other
4244 clients can make progress
4245 * more extN patch fixes
4246 * compile-time configurable ptlrpc buffer allocations
4248 - collaborative read cache document
4249 - Lustre Lite Performance CDR document-in-progress
4251 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4254 - fix typo in patch-2.4.18
4256 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4259 - clear ptlrpc request each time in handle_incoming_request()
4260 - unlink of files now destroys the object on the OST
4262 2002-09-19 Peter Braam <braam@clusterfs.com>
4264 * add hard link support
4265 * change obdfile creation method
4266 * kernel patch changed
4268 2002-09-19 Peter Braam <braam@clusterfs.com>
4271 - stack overflow bug in extN fixed
4273 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4275 * documentation updates
4276 - add man pages for config tools
4277 - update tests/README to describe testing with new config tools
4278 - finish metadata API descriptions
4279 * bug fixes and cleanups
4280 - statfs workaround for 16TB limit
4281 - LOV stripe allocation improved, can stripe on subset of OSTs
4282 - LOV file size/IO offset was wrong for files > 4GB in size
4283 - object EA data was being dropped, caused files to be unreadable
4284 - memory overflow with non-LOV OST caused memory corruption
4285 - fixed regression tests to work with new config tools, obdfilter
4286 - fixed bug when directory size became larger than 1 block
4287 - fixed bug (for single client case) when PWD was deleted
4288 - invalidate local directory pages when doing intent-based ops
4289 - avoid LDLM oops when lock callback contained bad data
4291 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4293 * documentation updates
4294 * bug fixes and cleanups
4295 - configuration tools
4298 - 64-bit compile warnings
4299 - 64-bit internal statfs data
4301 * test_brw on persistent OST devices
4303 * lprocfs (disabled)
4305 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4307 * documentation updates
4308 * bug fixes and cleanups
4309 * configuration tools
4311 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4314 * many small fixes to 0_5_4
4315 * io/network handling
4316 * thinkos in MDS operations
4318 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4321 * crucial basic fixes to 0.5.3
4322 * IOR, Iozone work over Elan
4325 2002-08-07 Phil Schwan <phil@clusterfs.com>
4326 * version 0_5_3, our first alpha
4327 * we use the new Portals iovs
4328 * documentation updates
4329 * bug fixes and cleanups
4330 * small changes in the DLM wire protocol
4332 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4333 * version 0_5_1 with some initial stability,
4334 * locking on MD and file I/O.
4335 * documentation updates
4336 * several bug fixes since 0.5.0
4337 * small changes in wire protocol
4339 2002-07-18 Phil Schwan <phil@clusterfs.com>
4341 * delivered as Lustre Light Alpha
4342 * fixed a crash after handling invalid MDS requests
4343 * fixed directory pages for architectures with non-4k pages sizes
4345 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4346 * release version v0_4_4
4347 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4349 * Fixes a number of interruption problems with OST operations.
4350 * Update documentation for portals header changes
4351 * Move all wire protocol structs/defines to lustre_idl.h
4352 * Fixes symlink length bug.
4353 * Add tcpdump to repository.
4355 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4356 * release version v0_4_3
4357 * Fixes statfs for inodes on extN.
4358 * Fixes bug in runtests which would delete /etc/hosts.
4359 * Use 64-bit object IDs wherever possible (not into VFS though)
4360 Remove ost_get_info, which is unused by lustre, and out of date.
4362 2002-07-03 Peter Braam <braam@clusterfs.com>
4363 * release version v0_4_2 Fixes a lookup error (type not passed)
4364 * move forward to head of Portals
4365 * move forward to latest Lustre kernel
4367 2002-06-25 Peter Braam <braam@clusterfs.com>
4368 * release version v0_4_1. Hopefully stable on single node use.