1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.6.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).
18 Description: Fix replay if there is an un-replied request and open
19 Details : In some cases, older replay request will revert the
20 mcd->mcd_last_xid on MDS which is used to record the client's
23 Severity : enhancement
25 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
27 Severity : enhancement
29 Description: Add rhel5 support to HEAD.
31 Serverity : enhancement
33 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
38 Description: Don't allow skipping OSTs if index has been specified.
39 Details : Don't allow skipping OSTs if index has been specified, make locking
40 in internal create lots better.
44 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
45 Details : access to bitfield in structure is always rounded to long
46 and this produce problem with not atomic change any bit.
50 Description: Lustre make rpms failed.
51 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
52 builds Lustre rpms from tarball.
55 Frequency : rare on shutdown ost
57 Description: If llog cancel was not send before clean_exports phase, this can
58 produce deadlock in llog code.
59 Details : If llog thread has last reference to obd and call class_import_put
60 this produce deadlock because llog_cleanup_commit_master wait when
61 last llog_commit_thread exited, but this never success because was
62 called from llog_commit_thread.
66 Description: allow userland application know is lost one of stripes.
67 Details : fill lvb_blocks with error code on ost and return it to
68 application if error flag found.
72 Description: NULL lov_tgts causing MDS oops
73 Details : more safe checks for NULL lov_tgts for avoid oops.
75 Severity : enhancement
77 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
81 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
82 Details : If an OST is down the MDS will hang indefinitely in
83 obd_statfs_rqset() waiting for the statfs data. While for
84 MDS QOS usage of statfs, it should not stuck in waiting.
86 Severity : enhancement
88 Description: remote_acl support
89 Details : Support ACL-based permission check for remote user.
90 Support setfacl/getfacl for remote user with the utils
91 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
92 the system "{s,g}etfacl" utils.
94 Severity : enhancement
96 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
98 Severity : enhancement
100 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
104 Description: make mgs_setparam() handle fsname containing dash
105 Details : fsname containing a dash does not work with lctl conf_param
107 Severity : enhancement
109 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
111 Severity : enhancement
113 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
117 Frequency : rare, at shutdown
118 Description: access already free / zero obd_namespace.
119 Details : if client_disconnect_export was called without force flag set,
120 and exist connect request in flight, this can produce access to
121 NULL pointer (or already free pointer) when connect_interpret
122 store ocd flags in obd_namespace.
126 Frequency : only at startup
127 Description: not alloc memory with spinlock held.
128 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
129 if any spinlock held.
131 Severity : enhancement
133 Description: make lustre randomly fail allocating memory
134 Details : Make lustre randomly failed allocating memory for testing purpose.
136 Severity : enhancement
138 Description: lost problems with lov objid file
139 Details : Fixes some scability and access to not inited memory problems
140 in work with lov objdid file.
145 Description: lfs find does not continue on file error
146 Details : Continue other files processing when a file/dir is absent.
150 Description: Inconsistent usage of lustre_pack_reply()
151 Details : Standardize the usage of lustre_pack_reply() such that it
152 always generate a CERROR on failure.
155 Frequency : occasional
157 Description: MDS hang or stay in waiting lock
158 Details : If client receive lock with CBPENDING flag ldlm need send lock
159 cancel as separate rpc, to avoid situation when cancel request
160 can't processed due all i/o threads stay in wait lock.
164 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
168 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
172 Description: Update to SLES9 kernel-2.6.5-7.287.3.
176 Description: Update to RHEL4 latest kernel.
178 Severity : enhancement
180 Description: Build SLES10 patchless client fails
181 Details : The configure was broken by run ./configure with
182 --with-linux-obj=.... argument for patchless client. When the
183 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
184 can't search header adequately. Use absolute path such as
185 -I($LINUX)/include instead.
189 Description: interrupt oig_wait produce painc on resend.
190 Details : brw_redo_request can be used for resend requests from ptlrpcd and
191 private set, and this produce situation when rq_ptlrpcd_data not
192 copyed to new allocated request and triggered LBUG on assert
193 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
194 ptlrpcd set if request is changed and can be safety changed to use
199 Description: LASSERT_{REQ,REP}SWAB macros are buggy
200 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
201 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
202 these macros and replace them with their difinitions instead.
206 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
207 Details : Update which_patch & target file for SLES10 latest kernel.
210 Description: Remove client patches from SLES 10 kernel.
211 Details : This causes SLES 10 clients to behave as patchless clients
212 even on a Lustre-patched (server) kernel.
214 Severity : enhancement
216 Description: self-adjustable client's lru lists
217 Details : use adaptive algorithm for managing client cached locks lru
218 lists according to current server load, other client's work
219 pattern, memory activities, etc. Both, server and client
220 side namespaces provide number of proc tunables for controlling
223 Severity : enhancement
225 Description: light-weight GSS support
226 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
227 and reduce performance overhead.
229 Severity : enhancement
231 Description: Linux keyring support
232 Details : Support using service of Linux keyring for Lustre GSS internal
233 context refresh/cache mechanism.
237 Description: Fix errors in lfs documentation
238 Details : Fixes man pages
242 Description: don't use GFP_* in generic Lustre code.
243 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
247 Description: obdclass is limited by single OBD_ALLOC(idarray)
248 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
252 Description: warnings with build patchless client with vanila 2.6.19 and up
253 Details : change old ctl_table style and replace ctl_table/ctl_table_header
254 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
258 Frequency : Only for SLES
259 Description: sanity_quota fail test_1
260 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
261 file, which makes TSTID[2] inunique.
265 Frequency : Only for SLES9
266 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
267 Details : Update target/ChangeLog/which_patch .
271 Frequency : Only for test_10 in sanity-quota.sh
272 Description: a bug in quota test code
273 Details : checking a test flag in a wrong place
277 Frequency : Only for test_13 in sanity-quota.sh
278 Description: a bug in quota test script
279 Details : shouldn't check overfull allocated quota
283 Description: jbd statistics
284 Details : Port older jbd statistics patch for sles10
286 Severity : enhancement
288 Description: Add printing inode info into message about error in writepage.
293 Description: reply_lock_interpret crash due to race with it and lock cancel.
294 Details : Do not replay locks that are being cancelled. Do not reference
295 locks by their address during replay, just by their handle.
299 Frequency : When flocks are used.
300 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
301 Details : Flock locks might destroy just granted lock if it could be merged
302 with another existing flock, this is done in completion handler,
303 so teach ldlm_cli_enquque_fini that this is a valid case for
309 Description: Oops in read and write path when failing to allocate lock.
310 Details : Check if lock allocation failed and return error back.
314 Description: lstripe command fails for valid OST index
315 Details : The stripe offset is compared to lov->desc.ld_tgt_count
316 instead of lov->desc.ld_active_tgt_count.
320 Description: sanity.sh failed test 103
321 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
326 Description: df doesn't work properly if diskfs blocksize != 4K
327 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
331 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
332 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
336 Description: obd_health_check_timeout too short
337 Details : set obd_health_check_timeout as 1.5x of obd_timeout
341 Description: llapi_file_create() does not allow some changes
342 Details : add llapi_file_open() that allows specifying the mode and
343 open flags, and also returns an open file handle.
347 Description: merge and cleanup kernel patches.
348 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
352 Description: Add journal checksum support.(Kernel part)
353 Details : The journal checksum feature adds two new flags i.e
354 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
355 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
356 indicates that the commit block contains the checksum for
357 the blocks described by the descriptor blocks. Now commit
358 record can be sent to disk without waiting for descriptor
359 blocks to be written to disk. This behavior is controlled
360 using JBD2_FEATURE_ASYNC_COMMIT flag.
364 Description: OSS needs mutliple precreate threads
365 Details : Add ability to start more than one create thread per OSS.
369 Description: Sanity.sh test_65k failed: setstripe should have succeeded
370 Details : execute lfs setstripe on client
374 Description: mds_obd_create error creating tmp object
375 Details : When the user sets quota on root, llog will be affected and can't
376 create files and write files.
379 Frequency : Always on ia64 patchless client, and possibly others.
381 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
382 Details : This allows the patchless client to be loaded on architectures
387 Description: RedHat Update kernel for RHEL5
388 Details : Add the kernel config file for RHEL5.
392 Description: RedHat Update kernel for RHEL5
393 Details : Modify the kernel config file more closer RHEL5.
397 Description: Build failure against Centos5 (RHEL5)
398 Details : Define PAGE_SIZE when it isn't present.
402 Description: test_7 on sanity-quota.sh can't be used on separate machines
403 Details : add facet to handle it
407 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
408 Details : with NFS, the anon dentry's parent was set to itself in
409 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
410 corresponding dentry other than use rec->ur_name.
412 Severity : enhancement
414 Description: lfs setstripe enhancement
415 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
419 Description: enable data checksumming by default
420 Details : enable checksum by default, allow --disable-checksum
421 configure option and "-o nochecksum" mount option. Checksums
422 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
426 Description: lustre support for RHEL5
427 Details : Add support for RHEL5.
431 Description: Client eviction due to failover config
432 Details : after a connection loss, the lustre client should attempt to
433 reconnect to the last active server first before trying the
434 other potential connections.
438 Description: when mds and osts use different quota unit(32bit and 64bit),
439 quota will be released repeatly.
440 Details : void sending multiple quota reqs to mds, which will keep the
441 status between the reqs.
445 Description: osts not allocated evenly to files
446 Details : change the condition to increase offset_idx
449 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
451 Description: Data corruption for OSTs that are formatted larger than 2TB
453 Details : When generating the bio request for lustre file writes the
454 sector number would overflow a temporary variable before being
455 used for the IO. The data reads correctly from Lustre (which
456 will overflow in a similar manner) but other file data or
457 filesystem metadata may be corrupted in some cases.
461 Description: Tune the kernel for good SCSI performance.
462 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
463 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
468 Description: rewrite ext2-derived code in obdclass/uuid.c
469 Details : rewrite inherited code (uuid parsing code from ext2 utils)
470 from scratch preserving functionality.
474 Description: "lfs find -obd UUID" prints directories
475 Details : "lfs find -obd UUID" will return all directory names instead
476 of just file names. It is incorrect because the directories
477 do not reside on the OSTs.
481 Description: use i_size_read and i_size_write in 2.6 port
482 Details : replace inode->i_size access with i_size_read/write()
485 Frequency : when removing large files
487 Description: scheduling issue during removal of large Lustre files
488 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
489 It causes scheduling issues when removing large files (17TB in the
493 Frequency : only with liblustre clients on XT3
495 Description: evictions taking too long
496 Details : allow llrd to evict clients directly on OSTs
499 Frequency : only on ppc
501 Description: /proc/fs/lustre/devices broken on ppc
502 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
503 We should make sure the type of 'index' is loff_t and then cast
504 explicitly as needed below. Do not assign an explicitly cast
508 Frequency : only for rhel5
510 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
511 Details : Modify the target file & which_kernel.
513 Severity : enhancement
515 Description: omit set fsid for export NFS
516 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
517 as FS_REQUIRES_DEV for avoid problems with generate fsid.
521 Frequency : Always, for kernels after 2.6.16
522 Description: Fix warning idr_remove called for id=.. which is not allocated.
523 Details : Last kernels save old s_dev before kill super and not allow
524 to restore from callback - restore it before call kill_anon_super.
528 Description: buffer overruns could theoretically occur
529 Details : llapi_semantic_traverse() modifies the "path" argument by
530 appending values to the end of the origin string, and a
531 overrun may occur. Adding buffer overrun check in liblustreapi.
535 Description: Fix error on 'ls .' at the top of the Lustre mount.
536 Details : Don't revalidate dentry if it is a root dentry.
540 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
541 Details : Modify vm-tunables-rhel4.patch.
545 Description: Kernel config for 2.6.18-vanilla.
546 Details : Modify targets/2.6-vanilla.target.in.
547 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
548 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
549 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
550 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
554 Description: improve handling recoverable errors
555 Details : if request processig with error which can be recoverable on server
556 request should be resend, otherwise page released from cache and
561 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
562 Details : Modify target file & which_patch.
563 A flaw was found in the IA32 system call emulation provided
564 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
565 value could be stored in the %RAX register, which could trigger an
566 out-of-bounds system call table access. An untrusted local user
567 could exploit this flaw to run code in the kernel
568 (ie a root privilege escalation). (CVE-2007-4573).
572 Description: O_DIRECT bypasses client statistics.
573 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
574 read_ahead_stats were updated. copied the stats section from
575 osc_send_oap_rpc() into async_internal().
579 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
583 Description: change order of libsysio includes
584 Details : '#include sysio.h' should always come before '#include xtio.h'
588 Description: handle "serious error: objid * already exists" more gracefully
589 Details : If LAST_ID value on disk is smaller than the objects existing in
590 the O/0/d* directories, it indicates disk corruption and causes an
591 LBUG(). If the object is 0-length, then we should use the existing
592 object. This will help to avoid a full fsck in most cases.
594 Severity : enhancement
596 Description: adapt the lustre_config script to support the upgrade case
597 Details : Add "-u" option for lustre_config script to support upgrading 1.4
598 server targets to 1.6 in parallel.
602 Description: To avoid grant space > avaible space when the disk is almost
603 full. Without this patch you might see the error "grant XXXX >
604 available" or some LBUG about grant, when the disk is almost
606 Details : In filter_check_grant, for non_grant cache write, we should
607 check the left space by if (*left > ungranted + bytes), instead
608 of (*left > ungranted), because only we are sure the left space
609 is enough for another "bytes", then the ungrant space should be
610 increase. In client, we should update cl_avail_grant only there
611 is OBD_MD_FLGRANT in the reply.
615 Description: Update RHEL 4 kernel to fix local root privilege escalation.
616 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
617 described in CVE-2007-4573. This problem could allow untrusted
618 local users to gain root access.
621 Frequency : when using O_DIRECT and quotas
623 Description: Incorrect file ownership on O_DIRECT output files
624 Details : block usage reported by 'lfs quota' does not take into account
625 files that have been written with O_DIRECT.
630 Description: touch file failed when fs is not full
631 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
632 otherwise we can get ENOSP while fs is not full.
636 Description: parallel lock callbacks
637 Details : Instead of sending blocking and completion callbacks as separated
638 requests, adding them to a set and sending in parallel.
641 Frequency : only for Cray XT3
642 Bugzilla : 12829/13455
643 Description: Changing primary group doesn't change the group lustre assigns to
645 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
646 overridden with the primary group provided by the group upcall,
647 whereas the supplied fsgid can be trusted if it is in the list of
648 supplementary groups returned by the group upcall.
650 Severity : enhancement
652 Description: Allow masking D_WARNING, D_ERROR messages from console
653 Details : Console messages can now be disabled via lnet.printk.
657 Description: User code with malformed file open parameter crashes client node
658 Details : Before packing join_file req, all the related reference should be
659 checked carefully in case some malformed flags cause fake
660 join_file req on client.
664 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
665 Details : ldlm_completion_ast() assumes that a lock is granted when the req
666 mode is equal to the granted mode. However, it should also check
667 that LDLM_FL_CP_REQD is not set.
671 Description: Heavy nfs access might result in deadlocks
672 Details : After ELC code landed, it is now improper to enqueue any mds
673 locks under och_sem, because enqueue might want to decide to
674 cancel open locks for same inode we are holding och_sem for.
678 Description: Client eviction while running blogbench
679 Details : A lot of unlink operations with concurrent I/O can lead to a
680 deadlock causing evictions. To address the problem, the number of
681 oustanding OST_DESTROY requests is now throttled to
682 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
683 callbacks are processed in priority.
687 Description: enable ACLs on MDS by default
688 Details : ACLs must be enabled on MDS by default.
691 Frequency : PPC/PPC64 only
693 Description: conflicts between asm-ppc64/types.h and lustre_types.h
694 Details : fix duplicated definitions between asm-ppc64/types.h and
695 lustre_types.h on PPC.
697 --------------------------------------------------------------------------------
699 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
701 * Support for kernels:
702 2.6.5-7.283 (SLES 9),
703 2.6.9-55.EL (RHEL 4),
704 2.6.16.46-0.14 (SLES 10),
705 2.6.18.8 vanilla (kernel.org)
706 * Client support for unpatched kernels:
707 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
708 2.6.16 - 2.6.22 vanilla (kernel.org)
709 * Due to recently discovered recovery problems, we do not recommend
710 using patchless RHEL 4 clients with this or any earlier release.
711 * Recommended e2fsprogs version: 1.39.cfs8
712 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
713 * Starting with this release, the ldiskfs backing filesystem required
714 by Lustre is now in its own package, lustre-ldiskfs. This package
715 should be installed. It is versioned separately from Lustre and
716 may be released separately in future.
720 Description: block reactivating mgc import until all deactivates complete
721 Details : Fix race when failing back MDT/MGS to itself (testing)
724 Frequency : at statup only
726 Description: mds_lov_synchronize race leads to various problems
727 Details : simultaneous MDT->OST connections at startup can cause the
728 sync to abort, leaving the OSC in a bad state.
730 Severity : enhancement
732 Description: add optional extra BUILD_VERSION info
733 Details : add a new environment variable (namely LUSTRE_VERS) which allows
734 to override the lustre version.
737 Frequency : 2.6.18 servers only
739 Description: ll_kern_mount() doesn't release the module reference
740 Details : The ldiskfs module reference count never drops down to 0
741 because ll_kern_mount() doesn't release the module reference.
746 Description: server LBUG when using old ost_num_threads parameter
747 Details : Accept the old ost_num_threads parameter but warn that it
748 is deprecated, and fix an off-by-one error that caused an LBUG.
753 Description: Transient SCSI error results in persistent IO issue
754 Details : iobuf->dr_error is not reinitialized to 0 between two
758 Frequency : sometimes when underlying device returns I/O errors
760 Description: OSTs not going read-only during write failures
761 Details : OSTs are not remounted read-only when the journal commit threads
762 get I/O errors because fsfilt_ext3 calls journal_start/stop()
763 instead of the ext3 wrappers.
767 Description: poor connect scaling with increasing client count
768 Details : Don't run filter_grant_sanity_check for more than 100 exports
769 to improve scaling for large numbers of clients.
772 Frequency : SLES10 only
774 Description: sanity-quota.sh quotacheck failed: rc = -22
775 Details : Quotas cannot be enabled on SLES10.
778 Frequency : liblustre clients only
780 Description: getdirentries does not give error when run on compute nodes
781 Details : getdirentries does not fail when the size specified as an argument
782 is too small to contain at least one entry
784 Severity : enhancement
786 Description: Add LNET router traceability for debug purposes
787 Details : If a checksum failure occurs with a router as part of the
788 IO path, the NID of the last router that forwarded the bulk data
789 is printed so it can be identified.
794 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
795 Details : Due to a race condition, liblustre clients were occasionally
798 Severity : enhancement
800 Description: lfs setstripe use optional parameters instead of postional
803 Severity : enhancement
805 Description: Nanosecond timestamp support for ldiskfs
806 Details : The on-disk ldiskfs filesystem has added support for nanosecond
807 resolution timestamps. There is not yet support for this at
808 the Lustre filesystem level.
811 Frequency : during server recovery
813 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
814 Details : request with rq_no_resend flag not awake l_wait_event if they get a
818 Frequency : nfs export on patchless client
820 Description: connectathon hang when test nfs export over patchless client
821 Details : Disconnected dentry cannot be found with lookup, so we do not need
822 to unhash it or make it invalid
825 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
826 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
827 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
832 Description: open req refcounting wrong on reconnect
833 Details : If reconnect happened between getting open reply from server and
834 call to mdc_set_replay_data in ll_file_open, we will schedule
835 replay for unreferenced request that we are about to free.
836 Subsequent close will crash in variety of ways.
837 Check that request is still eligible for replay in
838 mdc_set_replay_data().
843 Description: disable writes to filesystem when reading health_check file
844 Details : the default for reading the health_check proc file has changed
845 to NOT do a journal transaction and write to disk, because this
846 can cause reads of the /proc file to hang and block HA state
847 checking on a healthy but otherwise heavily loaded system. It
848 is possible to return to the previous behaviour during configure
849 with --enable-health-write.
851 Severity : enhancement
853 Description: 64-bit inode version
854 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
855 to the inode. This will be required for version-based recovery.
860 Description: MDS fails to start if a duplicate client export is detected
861 Details : in some rare cases it was possible for a client to connect to
862 an MDS multiple times. Upon recovery the MDS would detect this
863 and fail during startup. Handle this more gracefully.
865 Severity : enhancement
867 Description: Add -o localflock option to simulate old noflock
869 Details : This will achieve local-only flock/fcntl locks
875 Description: log_commit_thread vs filter_destroy race leads to crash
876 Details : Take import reference before releasing llog record semaphore
881 Description: Wrong request locking in request set processing
882 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
883 imp_delayed_list, in this place should be used imp_lock.
886 Frequency : when reconnection
888 Description: Grant Leak when osc reconnect to OST
889 Details : When osc reconnect ost, OST(filter) should check whether it
890 should grant more space to client by comparing fed_grant and
891 cl_avail_grant, and return the granted space to client instead
892 of "new granted" space, because client will call osc_init_grant
893 to update the client grant space info.
896 Frequency : when client reconnect to OST
898 Description: Grant Leak when osc do resend and replay bulk write
899 Details : When osc reconnect to OST, OST(filter)should clear grant info of
900 bulk write request, because the grant info while be sync between
901 OSC and OST when reconnect, and we should ignore the grant info
902 these of resend/replay write req.
907 Description: Grant space more than avaiable left space sometimes.
908 Details : When then OST is about to be full, if two bulk writing from
909 different clients came to OST. Accord the avaliable space of the
910 OST, the first req should be permitted, and the second one
911 should be denied by ENOSPC. But if the seconde arrived before
912 the first one is commited. The OST might wrongly permit second
913 writing, which will cause grant space > avaiable space.
916 Frequency : when client is evicted
918 Description: Grant might be wrongly erased when osc is evicted by OST
919 Details : when the import is evicted by server, it will fork another
920 thread ptlrpc_invalidate_import_thread to invalidate the
921 import, where the grant will be set to 0. While the original
922 thread will update the grant it got when connecting. So if
923 the former happened latter, the grant will be wrongly errased
924 because of this race.
929 Description: Checking Stale with correct fid
930 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
931 to check whether it is stale, because sometimes, we want the
932 enqueue happened anyway, and op_data.fid2 will not be initialized.
934 Severity : enhancement
936 Description: update patchless client
937 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
940 Frequency : only with 2.4 kernel
942 Description: random memory corruption
943 Details : size of struct ll_inode_info is to big for union inode.u and this
944 can be cause of random memory corruption.
949 Description: Memory leak in recovery
950 Details : Lov_mds_md was not free in an error handler in mds_create_object.
951 It should also check obd_fail before fsfilt_start, otherwise if
952 fsfilt_start return -EROFS,(failover mds during mds recovery).
953 then the req will return with repmsg->transno = 0 and rc = EROFS.
954 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
955 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
956 be freed no matter whether fsfilt_commit success or not.
959 Frequency : only with huge count clients
961 Description: Prevents from taking the superblock lock in llap_from_page for
963 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
964 for llap_from_page call in ll_removepage prevents from taking the
965 superblock lock for a soon died page.
970 Description: Not check open intent error before release open handle
971 Details : in some rare cases, the open intent error is not checked before
972 release open handle, which may cause
973 ASSERTION(open_req->rq_transno != 0), because it tries to release
974 the failed open handle.
979 Description: Set cat log bitmap only after create log success.
980 Details : in some rare cases, the cat log bitmap is set too early. and it
981 should be set only after create log success.
985 Description: Accessing a block bevice can re-enable I/O when Lustre is
986 tearing down a device.
987 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
992 Description: service threads may hog cpus when there are a lot of requests
993 Details : Insert cond_resched to give other threads a chance to use some CPU
998 Description: the cat log was not initialized in recovery
999 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1000 unlink log on mds will not be initialized until mds post
1001 recovery. And also in mds post recovery, the unlink log will
1002 initialization will be done asynchronausly, so there will be race
1003 between add unlink log and unlink log initialization.
1007 Description: brw_stats were being printed incorrectly
1008 Details : brw_stats were being printed as log2 but all of them were not
1009 recorded as log2. Also remove some code duplication arising from
1010 filter_tally_{read,write}.
1014 Frequency : rare, only in recovery.
1015 Description: ASSERTION(req->rq_type != LI_POISON) failed
1016 Details : imp_lock should be held while iterating over imp_sending_list for
1017 prevent destroy request after get timeout in ptlrpc_queue_wait.
1021 Description: replay-single.sh test 52 fails
1022 Details : A lock's skiplist need to be cleanup when it being unlinked
1023 from its resource list.
1027 Description: Short directio read returns full requested size rather than
1029 Details : Direct I/O operations should return actual amount of bytes
1030 transferred rather than requested size.
1032 Severity : enhancement
1034 Description: metadata RPC reduction (e.g. for rm performance)
1035 Details : decrease the amount of synchronous RPC between clients and servers
1036 by canceling conflicing lock before the operation on the client
1037 and packing thier handles into the main operation RPC to server.
1039 Severity : enhancement
1041 Description: Async OSC create to avoid the blocking unnecessarily.
1042 Details : If a OST has no remain object, system will block on the creating
1043 when need to create a new object on this OST. Now, ways use
1044 pre-created objects when available, instead of blocking on an
1045 empty osc while others are not empty. If we must block, we block
1046 for the shortest possible period of time.
1048 Severity : enhancement
1050 Description: refine locking for avoid write wrong info into lov_objid file
1051 Details : fix possible races with add new target and write/update data in
1054 --------------------------------------------------------------------------------
1056 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1061 Frequency : on some architectures
1063 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1064 Details : Uninitialized flags sometimes cause configuration commands to
1068 Frequency : patchless clients only
1070 Description: missing __iget() symbol export
1071 Details : The __iget() symbol export is missing. To avoid the need for
1072 this on patchless clients the deathrow inode reaper is turned
1073 off, and we depend on the VM to clean up old inodes. This
1074 dependency was during via the fix for bug 12181.
1076 --------------------------------------------------------------------------------
1078 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1080 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1081 INTEROPERATE with older versions automatically. In many cases a
1082 special upgrade step is needed. Please read the
1083 user documentation before upgrading any part of a live system.
1084 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1085 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1086 * WARNING: Lustre configuration and startup changes are required with
1087 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1093 Severity : enhancement
1095 Description: MountConf
1096 Details : Lustre configuration is now managed via mkfs and mount
1097 commands instead of lmc and lconf. New obd types (MGS, MGC)
1098 are added for dynamic configuration management. See
1099 https://mail.clusterfs.com/wikis/lustre/MountConf for
1102 Severity : enhancement
1104 Description: dynamic OST addition
1105 Details : OSTs can now be added to a live filesystem
1107 Severity : enhancement
1109 Description: startup order invariance
1110 Details : MDTs and OSTs can be started in any order. Clients only
1111 require the MDT to complete startup.
1113 Severity : enhancement
1115 Description: parallel, asynchronous orphan cleanup
1116 Details : orphan cleanup is now performed in separate threads for each
1117 OST, allowing parallel non-blocking operation.
1119 Severity : enhancement
1121 Description: optimized stripe assignment
1122 Details : stripe assignments are now made based on ost space available,
1123 ost previous usage, and OSS previous usage, in order to try
1124 to optimize storage space and networking resources.
1126 Severity : enhancement
1128 Description: Permanently set tunables
1129 Details : All writable /proc/fs/lustre tunables can now be permanently
1130 set on a per-server basis, at mkfs time or on a live system.
1132 Severity : enhancement
1134 Description: Lustre message v2
1135 Details : Add lustre message format v2.
1137 Severity : enhancement
1139 Description: client OST exclusion list
1140 Details : Clients can be started with a list of OSTs that should be
1141 declared "inactive" for known non-responsive OSTs.
1145 Description: SPEC SFS validation failure on NFS v2 over lustre.
1146 Details : Changes the blocksize for regular files to be 2x RPC size,
1147 and not depend on stripe size.
1149 Severity : enhancement
1151 Description: Multiple MD RPCs in flight.
1152 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1153 To avoid overly-overloading MDS, introduce a limit on number of
1154 MDS RPCs in flight for a single client and add /proc controls
1155 to adjust this limit.
1157 Severity : enhancement
1159 Description: client read/write statistics
1160 Details : Add client read/write call usage stats for performance
1161 analysis of user processes.
1162 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1163 file access. extents_stats shows chunk size distribution.
1164 extents_stats_per_process show chunk size distribution per
1167 Severity : enhancement
1169 Description: mds statistics
1170 Details : Add detailed mds operations statistics in
1171 /proc/fs/lustre/mds/*/stats.
1175 Description: Failure of copying files with lustre special EAs.
1176 Details : Client side always return success for setxattr call for lustre
1177 special xattr (currently only "trusted.lov").
1182 Description: Refcount LNET uuids
1183 Details : The global LNET uuid list grew linearly with every startup;
1184 refcount repeated list entries instead of always adding to
1187 Severity : enhancement
1189 Description: Dynamic service threads
1190 Details : Within a small range, start extra service threads
1191 automatically when the request queue builds up.
1194 Frequency : mixed-endian client/server environments
1196 Description: mixed-endian crashes
1197 Details : The new msg_v2 system had some failures in mixed-endian
1200 Severity : enhancement
1202 Description: Easy OST removal
1203 Details : OSTs can be permanently deactivated with e.g. 'lctl
1204 conf_param lustre-OST0001.osc.active=0'
1206 Severity : enhancement
1208 Description: MGS proc entries
1209 Details : Added basic proc entries for the MGS showing what filesystems
1212 Severity : enhancement
1214 Description: provide MGS failover
1215 Details : Added config lock reacquisition after MGS server failover.
1217 Severity : enhancement
1219 Description: add Linux 2.4 support
1220 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1224 Description: a large application tries to do I/O to the same resource and dies
1225 in the middle of it.
1226 Details : Check the req->rq_arrival time after the call to
1227 ost_brw_lock_get(), but before we do anything about
1228 processing it & sending the BULK transfer request. This
1229 should help move old stale pending locks off the queue as
1230 quickly as obd_timeout.
1233 Frequency : when an incorrect nid is specified during startup
1235 Description: ptlrpc connect to non-existant node causes kernel crash
1236 Details : LNET can't be re-entered from an event callback, which
1237 happened when we expire a message after the export has been
1238 cleaned up. Instead, hand the zombie cleanup off to another
1241 Severity : enhancement
1243 Description: plain/inodebits lock performance improvement
1244 Details : Grouping plain/inodebits in granted list by their request modes
1245 and bits policy, thus improving the performance of search through
1249 Frequency : only if OST filesystem is corrupted
1251 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1252 Details : for a short time RPCs with bulk IO are in the replay list,
1253 but replay of bulk IOs is unimplemented. If the OST filesystem
1254 is corrupted due to disk cache incoherency and then replay is
1255 started it is possible to trip an assertion. Avoid putting
1256 committed RPCs into the replay list at all to avoid this issue.
1259 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1261 Description: System hang on startup
1262 Details : This bug allowed the liblustre (e.g. catamount) client to
1263 return to the app before handling all startup RPCs. This
1264 could leave the node unresponsive to lustre network traffic
1265 and manifested as a server ptllnd timeout.
1267 Severity : enhancement
1269 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1270 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1271 boolean to control whether to print peer debug info when a
1272 client's RPC times out.
1275 Frequency : only for kernels with patches from Lustre below 1.4.3
1277 Description: Remove old rdonly API
1278 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1281 Frequency : only for devices with external journals
1283 Description: Set external device read-only also
1284 Details : During a commanded failover stop, we set the disk device
1285 read-only while the server shuts down. We now also set any
1286 external journal device read-only at the same time.
1289 Frequency : when upgrading from 1.4 while trying to change parameters
1291 Description: The wrong (new) MDC name was used when setting parameters for
1292 upgraded MDT's. Also allows changing of OSC (and MDC)
1293 parameters if --writeconf is specified at tunefs upgrade time.
1296 Frequency : when setting specific ost indicies
1298 Description: QOS code breaks on skipped indicies
1299 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1300 created with --index need not be sequential.
1302 Severity : enhancement
1304 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1305 Details : The uninit_groups feature works in conjunction with the kernel
1306 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1307 up the pass1 processing of e2fsck. This is a read-only feature
1308 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1309 mount filesystems that have had this feature enabled.
1311 Severity : enhancement
1313 Description: Improve multi-block allocation algorithm to avoid fragmentation
1314 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1315 allocation locality and avoid filesystem fragmentation.
1317 ------------------------------------------------------------------------------
1319 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1321 * Support for kernels:
1322 2.4.21-47.0.1.EL (RHEL 3)
1323 2.6.5-7.283 (SLES 9)
1324 2.6.9-42.0.10.EL (RHEL 4)
1325 2.6.12.6 vanilla (kernel.org)
1326 2.6.16.27-0.9 (SLES 10)
1327 * Recommended e2fsprogs version: 1.39.cfs5
1329 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1333 Frequency : occasional, depends on client load and configuration
1334 Bugzilla : 12181, 12203
1335 Description: data loss for recently-modified files
1337 Details : In some cases it is possible that recently written or created
1338 files may not be written to disk in a timely manner (this should
1339 normally be within 30s unless client IO load is very high).
1340 The problem appears as zero-length files or files that are a
1341 multiple of 1MB in size after a client crash or client eviction
1342 that are missing data at the end of the file.
1344 This problem is more likely to be hit on clients where files are
1345 repeatedly created and unlinked in the same directory, clients
1346 have a large amount of RAM, have many CPUs, the filesystem has
1347 many OSTs, the clients are rebooted frequently, and/or the files
1348 are not accessed by other nodes after being written.
1350 The presence of the problem can be detected by looking at
1351 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1352 smaller than the second (nr_unused) then dirty files will not
1353 be flushed automatically to disk. "sync; sleep 10" should be
1354 run several times on the node before unmounting it to update
1355 Lustre (this is also safe to run on nodes without this problem).
1357 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1358 that can cause this same problem, so customers using that kernel
1359 also need to update the kernel in addition to Lustre. In order
1360 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1362 It is normal that files written just before a client crash (less
1363 than 30s) may not yet have been flushed to disk, even for local
1367 Frequency : frequent on thin XT3 nodes
1369 Description: UUID collision on thin XT3 Linux nodes
1370 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1371 randomly, since we relied on an insufficiently-seeded PRNG.
1376 Description: OSS hangs after "All ost request buffers busy"
1377 Details : A deadlock between quota and journal operations caused OSS
1378 hangs after printing "All ost request buffers busy."
1381 Frequency : always on liblustre builds
1383 Description: Cleanup compiler warnings on liblustre
1386 Frequency : always on liblustre builds on XT3
1388 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1392 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1393 > 32000 subdirectories
1394 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1395 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1396 subdirectory count crosses 32000. This will aid e2fsck to
1397 correctly handle more than 32000 subdirectories.
1400 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1402 Description: System hang on startup
1403 Details : This bug allowed the liblustre (e.g. catamount) client to
1404 return to the app before handling all startup RPCs. This
1405 could leave the node unresponsive to lustre network traffic
1406 and manifested as a server ptllnd timeout.
1408 Severity : enhancement
1410 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1411 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1412 boolean to control whether to print peer debug info when a
1413 client's RPC times out.
1418 Description: make O_SYNC working on 2.6 kernels
1419 Details : 2.6 kernels use different method for mark pages for write,
1420 so need add a code to lustre for O_SYNC work.
1425 Description: Failure to close file and release space on NFS
1426 Details : Put inode details into lock acquired in ll_intent_file_open.
1427 Use mdc_intent_lock in ll_intent_open to properly
1428 detect all kind of errors unhandled by mdc_enqueue.
1433 Description: proc file read during shutdown sometimes raced obd removal,
1435 Details : Add lock to prevent obd access after proc file removal.
1438 Frequency : Only for files larger than 4GB on 32-bit clients.
1440 Description: improperly doing page alignment of locks
1441 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1442 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1447 Description: RPCs being resent when they shouldn't be
1448 Details : Some RPCs that should not be resent are being resent. This
1449 can cause inconsistencies in the RPC state machine. Do not
1450 resend such requests.
1453 Frequency : rare, only with NFS export
1455 Description: Crash on NFS re-export node
1456 Details : under very unusual load conditions an assertion is hit in
1457 ll_intent_file_open()
1460 Frequency : only if OST filesystem is corrupted
1462 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1463 Details : for a short time RPCs with bulk IO are in the replay list,
1464 but replay of bulk IOs is unimplemented. If the OST filesystem
1465 is corrupted due to disk cache incoherency and then replay is
1466 started it is possible to trip an assertion. Avoid putting
1467 committed RPCs into the replay list at all to avoid this issue.
1472 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1473 Details : Large single O_DIRECT read and write calls can fail to allocate
1474 a sufficiently large buffer to process the request. In case of
1475 allocation failure the allocation is retried with a smaller
1476 buffer and broken into smaller requests.
1478 Severity : enhancement
1480 Description: Add -o localflock option to simulate old noflock behaviour.
1481 Details : This will achieve local-only flock/fcntl locks coherentness.
1486 Description: versioning check is incomplete
1487 Details : Checking the version difference of client vs. server, report
1488 error if the gap is too big.
1493 Description: add support PG_writeback bit
1494 Details : add support for PG_writeback bit for Lustre, for more carefull
1495 work with page cache in 2.6 kernel. This also fix some deadlocks
1496 and remove hack for work O_SYNC with 2.6 kernel.
1498 Severity : enhancement
1500 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1501 Details : The uninit_groups feature works in conjunction with the kernel
1502 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1503 up the pass1 processing of e2fsck. This is a read-only feature
1504 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1505 mount filesystems that have had this feature enabled.
1507 Severity : enhancement
1509 Description: Improve multi-block allocation algorithm to avoid fragmentation
1510 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1511 allocation locality and avoid filesystem fragmentation.
1513 ------------------------------------------------------------------------------
1515 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1517 * Support for kernels:
1518 2.6.9-42.0.3.EL (RHEL 4)
1519 2.6.5-7.276 (SLES 9)
1520 2.4.21-47.0.1.EL (RHEL 3)
1521 2.6.12.6 vanilla (kernel.org)
1522 2.6.16.21-0.8 (SLES10)
1523 * Recommended e2fsprogs version: 1.39.cfs2-0
1525 * The backwards-compatible /proc/sys/portals symlink has been removed
1526 in this release. Before upgrading, please ensure that you change
1527 any configuration scripts or /etc/sysctl.conf files that access
1528 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1529 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1530 in advance of the upgrade on any system running Lustre 1.4.6 or
1531 newer, since /proc/sys/lnet was added in that version.
1532 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1536 Frequency : only when quota is used
1538 Description: avoid scanning export list for quota master
1539 Details : Change the algorithms to avoid scanning export list in order
1540 to improve the efficiency.
1543 Frequency : MDS failover only, very rarely
1545 Description: "went back in time" messages on mds failover
1546 Details : The greatest transno may be lost when the current operation
1547 finishes with an error (transno==0) and the client's last_rcvd
1548 record is over-written. Save the greatest transno in the
1549 mds_last_transno for this case.
1552 Frequency : always for specific kernels and striping counts
1554 Description: client may get "Matching packet too big" without ACL support
1555 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1556 when trying to access files in certain configurations. The
1557 clients should in fact be denied when mounting because they do
1558 not understand ACLs.
1561 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1563 Description: many clients connecting with IO in progress causes connect timeouts
1564 Details : Avoid synchronous journal commits to avoid delays caused by many
1565 clients connecting/disconnecting when bulk IO is in progress.
1566 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1567 OST_IO_PORTAL to avoid delays behind potentially many pending
1571 Frequency : occasionally with multiple writers to a single file
1573 Description: shared writes to file may result in wrong size reported by stat()
1574 Details : Allow growing of kms when extent lock is cancelled
1577 Frequency : always with random mmap IO to multi-striped file
1579 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1580 Details : Only if the hole is at the end of OST object so that kms is too
1581 small. Fix is to increase kms accordingly in ll_nopage.
1584 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1586 Description: writes to a missing object would leak memory on the OST
1587 Details : If there is an inconsistency between the MDS and OST filesystems,
1588 such that the MDS references an object that doesn't exist, writes
1589 to that object will leak memory due to incorrect cleanup in the
1590 error handling path, eventually running out of memory on the OST.
1595 Description: Creating too long symlink causes lustre errors
1596 Details : Check symlink and name lengths before sending requests to MDS.
1599 Frequency : only if flock is enabled (not on by default)
1601 Description: posix locks not released on fd closure on 2.6.9+
1602 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1603 kernels, this caused such locks not to be released on fd close and
1604 then assertions on fs unmount about still used locks.
1607 Frequency : MDS failover only, very rarely
1609 Description: clients may get ASSERTION(granted_lock != NULL)
1610 Details : When request was taking a long time, and a client was resending
1611 a getattr by name lock request. The were multiple lock requests
1612 with the same client lock handle and
1613 mds_getattr_name->fixup_handle_for_resent_request found one of the
1614 lock handles but later failed with ASSERTION(granted_lock != NULL).
1619 Description: handle->h_buffer_credits > 0, assertion failure
1620 Details : h_buffer_credits is zero after truncate, causing assertion
1621 failure. This patch extends the transaction or creates a new
1625 Frequency : NFS re-export or patchless client
1626 Bugzilla : 11179, 10796
1627 Description: Crash on NFS re-export node (__d_move)
1628 Details : We do not want to hash the dentry if we don't have a lock.
1629 But if this dentry is later used in d_move, we'd hit uninitialised
1630 list head d_hash, so we just do this to init d_hash field but
1631 leave dentry unhashed.
1634 Frequency : NFS re-export or patchless client
1636 Description: NFS exports has problem with symbolic link
1637 Details : lustre client didn't properly install dentry when re-exported
1638 to NFS or running patchless client.
1641 Frequency : NFS re-export or patchless client
1643 Description: Various nfs/patchless fixes.
1644 Details : fixes reuse disconected alias for lookup process - this fixes
1645 warning "find_exported_dentry: npd != pd",
1646 fix permission error with open files at nfs.
1647 fix apply umask when do revalidate.
1650 Frequency : occasional
1652 Description: Crash on NFS re-export node
1653 Details : calling clear_page() on the wrong pointer triggered oops in
1654 generic_mapping_read().
1657 Frequency : rarely, using O_DIRECT IO
1659 Description: unaligned directio crashes client with LASSERT
1660 Details : check for unaligned buffers before trying any requests.
1663 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1665 Description: stale data returned from RAID cache
1666 Details : If only a small amount of IO is done to the RAID device before
1667 reading it again it is possible to get stale data from the RAID
1668 cache instead of reading it from disk.
1671 Frequency : always for sles10 kernel
1673 Description: sles10 support
1674 Details : ll_follow_link: compile fixes and using of nd_set_link
1675 under newer kernels.
1678 Frequency : depends on arch, kernel and compiler version, always on sles10
1681 Description: recursive or deep enough symlinks cause stack overflow
1682 Details : getting rid of large stack-allocated variable in
1686 Frequency : depends on hardware
1688 Description: lustre write performance loss in the SLES10 kernel
1689 Details : the performance loss is caused by using of write barriers in the
1690 ext3 code. The SLES10 kernel turns barrier support on by
1691 default. The fix is to undo that change for ldiskfs.
1693 ------------------------------------------------------------------------------
1695 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1697 * Support for kernels:
1698 2.6.9-42.0.3EL (RHEL 4)
1699 2.6.5-7.276 (SLES 9)
1700 2.4.21-47.0.1.EL (RHEL 3)
1701 2.6.12.6 vanilla (kernel.org)
1705 Frequency : quota enabled and large files being deleted
1707 Description: releasing more than 4GB of quota at once hangs OST
1708 Details : If a user deletes more than 4GB of files on a single OST it
1709 will cause the OST to spin in an infinite loop. Release
1710 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1715 Description: statfs data retrieved from /proc may be stale or zero
1716 Details : When reading per-device statfs data from /proc, in the
1717 {kbytes,files}_{total,free,avail} files, it may appear
1718 as zero or be out of date.
1721 Frequency : systems with MD RAID1 external journal devices
1723 Description: lconf's call to blkid is confused by RAID1 journal devices
1724 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1725 of returning all block devices that match the journal UUID.
1728 Frequency : always, for aggregate stripe size over 4GB
1730 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1731 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1732 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1735 Frequency : always if "lfs find" used on a local file/directory
1737 Description: "lfs find" segfaults if used on a local file/directory
1738 Details : The case where a directory component was not specified wasn't
1739 handled correctly. Handle this properly.
1742 Frequency : always on ppc64
1744 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1745 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1746 perform endianness conversion.
1749 Frequency : rarely (truncate to non-zero file size after write under load)
1750 Bugzilla : 10730, 10687
1751 Description: Files padded with zeros to next 4K multiple
1752 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1753 it is possible that files that are truncated to a non-zero size
1754 immediately after being written are filled with zero bytes beyond
1755 the truncated size. No file data is lost.
1757 Severity : enhancement
1759 Description: Allow recovery/failover for liblustre clients.
1760 Details : liblustre clients were unaware of failover configurations until
1763 Severity : enhancement
1765 Description: user file locks should fail when not mounting with flock option
1766 Details : Set up an error-returning stub in ll_file_operations.lock field
1767 to prevent incorrect behaviour when client is mounted without
1768 flock option. Also, set up properly f_op->flock field for
1772 Frequency : always on ia64
1774 Description: "lfs df" loops on printing out MDS statfs information
1775 Details : The obd_ioctl_data was not initialized and in some systems
1776 this caused a failure during the ioctl that did not return
1777 an error. Initialize the struct and return an error on failure.
1780 Frequency : SLES 9 only
1782 Description: Error of copying files with lustre special EAs as root
1783 Details : Client side always return success for setxattr call for lustre
1784 special xattr (currently only "trusted.lov").
1787 Frequency : rarely on clusters with both ia64+i386 clients
1789 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1790 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1791 clients are doing concurrent writes to the same file it is
1792 possible that the ia64 clients may LASSERT because the OST
1793 extent locks are not PAGE_SIZE aligned. Ensure that grown
1794 locks are always aligned on the request boundary.
1797 Frequency : specific use, occasional
1799 Description: Overwriting in use executable truncates on-disk binary image
1800 Details : If one node attempts to overwrite an executable in use by
1801 another node, we now correctly return ETXTBSY instead of
1802 truncating the file.
1804 Severity : enhancement
1806 Description: Async OSC create to avoid the blocking unnecessarily.
1807 Details : If a OST has no remain object, system will block on the creating
1808 when need to create a new object on this OST. Now, ways use
1809 pre-created objects when available, instead of blocking on an
1810 empty osc while others are not empty. If we must block, we block
1811 for the shortest possible period of time.
1816 Description: chmod on Lustre root is propagated to other clients
1817 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1818 invalid by the follow_mount time.
1821 Frequency : liblustre clients only
1823 Description: Race in 'instant cancel' lock handling could lead to such locks
1824 never to be granted in case of SMP MDS
1825 Details : Do not destroy not yet granted but cbpending locks in
1829 Frequency : replay/resend of open
1831 Description: non null lock assetion failure in mds_intent_policy
1832 Details : Trying to replay/resend lockless open requests resulted in
1833 mds_open() returning 0 with no lock. Now it sets a flag if
1834 a lock is going to be returned.
1836 Severity : enhancement
1838 Description: Checksum enhancements
1839 Details : New checksum enhancements allow for resending RPCs that failed
1842 Severity : enhancement
1844 Description: Tunables on number of dirty pages in cacche
1845 Details : Allow to set limit on number of dirty pages cached.
1850 Description: client crash on unmount - lock still has references
1851 Details : In some error handling cases it was possible to leak a lock
1852 reference on a client while accessing a file. This was not
1853 harmful to the client during operation, but would cause the
1854 client to crash when the filesystem is unmounted.
1857 Frequency : specific case, rare
1859 Description: ETXTBSY on mds though file not in use by client
1860 Details : ETXTBSY is no longer incorrectly returned when attempting to
1861 chmod or chown a directory that the user previously tried to
1862 execute or a currently-executing binary.
1865 Frequency : extremely rare except on liblustre-based clients
1867 Description: Lustre space not freed when files are deleted
1868 Details : Clean up open-unlinked files after client eviction. Previously
1869 the unlink was skipped and the files remained as orphans.
1874 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1875 Details : In some cases it was possible to send a blocking callback to
1876 a client doing a glimpse, even though that client didn't get
1877 a lock granted. When the glimpse lock is cancelled on the OST
1878 the freed lock is left on the waiting list and corrupted the list.
1881 Frequency : all core dumps
1883 Description: Broke core dumps to lustre
1884 Details : Negative dentry may be unhashed if parent does not have UPDATE
1885 lock, but some callers, e.g. do_coredump, expect dentry to be
1886 hashed after successful create, hash it in ll_create_it.
1888 ------------------------------------------------------------------------------
1890 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1892 * Support for kernels:
1893 2.6.9-42.0.2.EL (RHEL 4)
1894 2.6.5-7.276 (SLES 9)
1895 2.4.21-40.EL (RHEL 3)
1896 2.6.12.6 vanilla (kernel.org)
1900 Frequency : always on RHEL 3
1902 Description: Number of open files grows over time
1903 Details : The number of open files grows over time, whether or not
1904 Lustre is started. This was due to a filp leak introduced
1905 by one of our kernel patches.
1907 ------------------------------------------------------------------------------
1909 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1911 * Support for kernels:
1912 2.6.9-42.EL (RHEL 4)
1913 2.6.5-7.276 (SLES 9)
1914 2.4.21-40.EL (RHEL 3)
1915 2.6.12.6 vanilla (kernel.org)
1920 Bugzilla : 5719, 9635, 9792, 9684
1921 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1922 Details : If a server is under heavy load and cannot reply to new
1923 connection requests before the client resends the (re)connect,
1924 the connection handling code can behave badly if two service
1925 threads are concurrently handing separate (re)connections from
1926 the same client. Add better locking to the connection handling
1927 code, and ensure that only a single connection will be processed
1928 for a given client UUID, even if the lock is dropped.
1930 Severity : enhancement
1932 Description: add TCP zero-copy support to kernel
1933 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1934 sends if the hardware supports scatter-gather and checksumming.
1935 This allows socklnd to do client-write and server-read more
1936 efficiently and reduce CPU utilization from skbuf copying.
1939 Frequency : only if NFS exporting from client
1941 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1942 Details : If setting attributes on a file created under NFS that had
1943 never been opened it would be possible to oops the client
1944 if the file had no objects.
1947 Frequency : always for liblustre
1949 Description: liblustre client does MDS+OSTs setattr RPC for each write
1950 Details : When doing a write from a liblustre client, the client
1951 incorrectly issued an RPC to the MDS and each OST the file was
1952 striped over in order to update the timestamps. When writing
1953 with small chunks and many clients this could overwhelm the MDS
1954 with RPCs. In all cases it would slow down the write because
1955 these RPCs are unnecessary.
1957 Severity : enhancement
1959 Description: allow number of MDS service threads to be changed at module load
1960 Details : It is now possible to change the number of MDS service threads
1961 running. Adding "options mds mds_num_threads={N}" to the MDS's
1962 /etc/modprobe.conf will set the number of threads for the next
1963 time Lustre is restarted (assuming the "mds" module is also
1964 reloaded at that time). The default number of threads will
1965 stay the same, 32 for most systems.
1970 Description: OST crash if filesystem is unformatted or corrupt
1971 Details : If an OST is started on a device that has never been formatted
1972 or if the filesystem is corrupt and cannot even mount then the
1973 error handling cleanup routines would dereference a NULL pointer.
1978 Description: NULL pointer deref in llap_from_page.
1979 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1980 incorrect) mapping if the page was truncated/reclaimed while it was
1981 searched for. Check for this condition and skip such pages when
1982 doing readahead. Introduce extra check to llap_from_page() to
1983 verify page->mapping->host is non-NULL (so page is not anonymous).
1986 Frequency : Sometimes when using sys_sendfile
1988 Description: "page not covered by a lock" warnings from ll_readpage
1989 Details : sendfile called ll_readpage without right page locks present.
1990 Now we introduced ll_file_sendfile that does necessary locking
1991 around call to generic_file_sendfile() much like we do in
1995 Frequency : with certain MDS communication failures at client mount time
1997 Description: NULL pointer deref after failed client mount
1998 Details : a client connection request may delayed by the network layer
1999 and not be sent until after the PTLRPC layer has timed out the
2000 request. If the client fails the mount immediately it will try
2001 to clean up before the network times out the request. Add a
2002 reference from the request import to the obd device and delay
2003 the cleanup until the network drops the request.
2006 Frequency : occasionally during client (re)connect
2008 Description: assertion failure during client (re)connect
2009 Details : processing a client connection request may be delayed by the
2010 client or server longer than the client connect timeout. This
2011 causes the client to resend the connection request. If the
2012 original connection request is replied in this interval, the
2013 client may trip an assertion failure in ptlrpc_connect_interpret()
2014 which thought it would be the only running connect process.
2017 Frequency : only with obd_echo servers and clients that are rebooted
2019 Description: kernel BUG accessing uninitialized data structure
2020 Details : When running an obd_echo server it did not start the ping_evictor
2021 thread, and when a client was evicted an uninitialized data
2022 structure was accessed. Start the ping_evictor in the RPC
2023 service startup instead of the OBD startup.
2025 Severity : enhancement
2026 Bugzilla : 10193 (patchless)
2027 Description: Remove dependency on various unexported kernel interfaces.
2028 Details : No longer need reparent_to_init, exit_mm, exit_files,
2029 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2032 Frequency : rare (only users of deprecated and unsupported LDAP config)
2034 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2035 Details : LDAP apparently contains 'lustreName' attributes instead of
2036 'name'. A simple remapping of the name is sufficient.
2039 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2041 Description: waiting_locks_callback trips kernel BUG if client is evicted
2042 Details : Running with the dump_on_timeout debug flag turned on makes
2043 it possible that the waiting_locks_callback() can try to dump
2044 the Lustre kernel debug logs from an interrupt handler. Defer
2045 this log dumping to the expired_lock_main() thread.
2047 Severity : enhancement
2049 Description: Support NFS exporting on 2.6 kernels.
2050 Details : Implement non-rawops metadata methods for NFS server to use without
2051 changing NFS server code.
2054 Frequency : very rare (synthetic metadata workload only)
2056 Description: two racing renames might cause an MDS thread to deadlock
2057 Details : Running the "racer" program may cause one MDS thread to rename
2058 a file from being the source of a rename to being the target of
2059 a rename at exactly the same time that another thread is doing
2060 so, and the second thread has already enqueued these locks after
2061 doing a lookup of the target and is trying to relock them in
2062 order. Ensure that we don't try to re-lock the same resource.
2065 Frequency : only very large systems with liblustre clients
2067 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2068 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2069 all OSTs in parallel. This allows the eviction of stale liblustre
2070 clients to proceed much faster than if they were done in series,
2071 and also offers similar improvements for other set_info RPCs.
2076 Description: excessive CPU usage during initial read phase on client
2077 Details : During the initial read phase on a client, it would agressively
2078 retry readahead on the file, consuming too much CPU and impacting
2079 performance (since 1.4.5.8). Improve the readahead algorithm
2080 to avoid this, and also improve some other common cases (read
2081 of small files in particular, where "small" is files smaller than
2082 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2087 Description: MDS crash when receiving packet with unknown intent.
2088 Details : Do not LBUG in unknown intent case, just return -EFAULT
2090 Severity : enhancement
2091 Bugzilla : 9293, 9385
2092 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2093 Details : Do not serialize getattr (non-intent version) and statfs.
2096 Frequency : occasional, when OST network is overloaded/intermittent
2098 Description: client evicted by OST after bulk IO timeout
2099 Details : If a client sends a bulk IO request (read or write) the OST
2100 may evict the client if it is unresposive to its data GET/PUT
2101 request. This is incorrect if the network is overloaded (takes
2102 too long to transfer the RPC data) or dropped the OST GET/PUT
2103 request. There is no need to evict the client at all, since
2104 the pinger and/or lock callbacks will handle this, and the
2105 client can restart the bulk request.
2108 Frequency : Always when mmapping file with no objects
2110 Description: client crashes when mmapping file with no objects
2111 Details : Check that we actually have objects in a file before doing any
2112 operations on objects in ll_vm_open, ll_vm_close and
2118 Description: Request leak when working with deleted CWD
2119 Details : Introduce advanced request refcount tracking for requests
2120 referenced from lustre intent.
2122 Severity : Enhancement
2124 Description: Cache open file handles on client.
2125 Details : MDS now will return special lock along with openhandle, if
2126 requested and client is allowed to hold openhandle, even if unused,
2127 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2128 closing files for every read/write openration.
2130 Severity : Enhancement
2132 Description: Cache open negative dentries on client when possible.
2133 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2134 negative dentries on lock revocation.
2139 Description: Remounting a client read-only wasn't possible with a zconf mount
2140 Details : It wasn't possible to remount a client read-only with llmount.
2142 Severity : enhancement
2143 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2144 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2145 MPICH is included to add Lustre-specific ADIO interfaces.
2146 This is based closely on the UFS ADIO layer and only differs
2147 in file creation, in order to allow the OST striping to be set.
2148 This is user-contributed code and not supported by CFS.
2153 Description: extended inode attributes (immutable, append-only) work improperly
2154 when 2.4 and 2.6 kernels are used on client/server or vice versa
2155 Details : Introduce kernel-independent values for these flags.
2157 Severity : enhancement
2160 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2161 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2162 granularity. Now, Fractional MB granularity is be supported,
2163 this is very useful for low memory system.
2165 Severity : enhancement
2167 Description: Getattr by fid
2168 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2169 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2170 have locks and avoids getattr rpc requests for it on every stat.
2173 Frequency : Always, for filesystems larger than 2TB
2175 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2176 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2177 filesystems larger than 2TB to be mounted with mballoc enabled.
2180 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2182 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2183 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2184 filesystems larger than 2TB is mounted then the kernel will
2185 silently corrupt the start of the filesystem. CONFIG_LBD is
2186 enabled for all CFS-supported kernels, but the possibility of
2187 this happening with a modified kernel config exists.
2189 Severity : enhancement
2191 Description: add client O_DIRECT support for 2.6 kernels
2192 Details : It is now possible to do O_DIRECT reads and writes to files
2193 in the Lustre client mountpoint on 2.6 kernel clients.
2195 Severity : enhancement
2197 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2198 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2199 parallel, not waiting for response from every OST before sending
2200 a rpc to the next OST.
2205 Description: setattr vs write race when updating file timestamps
2206 Details : Client processes that update a file timestamp into the past
2207 right after writing to the file (e.g. tar) it is possible that
2208 the updated file modification time can be reset to the current
2209 time due to a race between processing the setattr and write RPC.
2211 Severity : enhancement
2213 Description: Bring 'lfs find' closer in line with regular Linux find.
2214 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2215 -print0 options and obtains all the needed info through the lustre
2218 Severity : enhancement
2220 Description: support up to 1024 configured devices on one node
2221 Details : change obd_dev array from statically allocated to dynamically
2222 allocated structs as they are first used to reduce memory usage
2227 Description: Flush dirty partially truncated pages during truncate
2228 Details : Immediatelly flush partially truncated pages in filter_setattr,
2229 this way we completely avoid having any pages in page cache on OST
2230 and can retire ugly workarounds during writes to flush such pages.
2235 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2236 Details : protect inode from truncation within vfs_unlink() context
2237 just take a reference before calling vfs_unlink() and release it
2238 when parent's i_sem is free.
2243 Description: last_id value checked outside lock on OST caused LASSERT failure
2244 Details : If there were multiple MDS->OST object precreate requests in
2245 flight, it was possible that the OST's last object id was checked
2246 outside a lock and incorrectly tripped an assertion. Move checks
2247 inside locks, and discard old precreate requests.
2250 Frequency : always, if extents are used on OSTs
2252 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2253 Details : OSTs using the extents format would not zero the high 16 bits of
2254 the index physical block number. This is not a problem for any
2255 OST filesystems smaller than 16TB, and no kernels support ext3
2256 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2257 new/modified files) and can be fixed for existing filesystems
2258 with e2fsprogs-1.39-cfs1.
2263 Description: import connection selection may be incorrect if timer wraps
2264 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2265 import connections to be ignored if the 32-bit jiffies counter
2266 wraps. Use a 64-bit jiffies counter.
2269 Frequency : very large clusters immediately after boot
2271 Description: LNET request buffers exhausted under heavy short-term load
2272 Details : If a large number of client requests are generated on a service
2273 that has previously never seen so many requests it is possible
2274 that the request buffer growth cannot keep up with the spike in
2275 demand. Instead of dropping incoming requests, they are held in
2276 the LND until the RPC service can accept more requests.
2279 Frequency : Sometimes during replay
2281 Description: Assertion failure in ll_local_open after replay.
2282 Details : If replay happened on an open request reply before we were able
2283 to set replay handler, reply will become not swabbed tripping the
2284 assertion in ll_local_open. Now we set the handler right after
2285 recognising of open request
2288 Frequency : very rare
2290 Description: kernel reports "badness in vsnprintf"
2291 Details : Reading from the "recovery_status" /proc file in small chunks
2292 may cause a negative length in lprocfs_obd_rd_recovery_status()
2293 call to vsnprintf() (which is otherwise harmless). Exit early
2294 if there is no more space in the output buffer.
2296 Severity : enhancement
2298 Description: clear OBD RPC statistics by writing to them
2299 Details : It is now possible to clear the OBD RPC statistics by writing
2300 to the "stats" file.
2305 Description: Client mtime is not the same on different clients after utimes
2306 Details : In some cases, the client was using the utimes() syscall on
2307 a file cached on another node. The clients now validate the
2308 ctime from the MDS + OSTs to determine which one is right.
2313 Description: Inability to activate failout mode
2314 Details : lconf script incorrectly assumed that in pythong string's numeric
2315 value is used in comparisons.
2318 Frequency : always with multiple stripes per file
2320 Description: Inefficient object allocation for mutli-stripe files
2321 Details : When selecting which OSTs to stripe files over, for files with
2322 a stripe count that divides evenly into the number of OSTs,
2323 the MDS is always picking the same starting OST for each file.
2324 Return the OST selection heuristic to the original design.
2329 Description: mount failures may take full timeout to return an error
2330 Details : Under some heavy load conditions it is possible that a
2331 failed mount can wait for the full obd_timeout interval,
2332 possibly several minutes, before reporting an error.
2333 Instead return an error as soon as the status is known.
2335 Frequency : quota enabled and large files being deleted
2337 Description: releasing more than 4GB of quota at once hangs OST
2338 Details : If a user deletes more than 4GB of files on a single OST it
2339 will cause the OST to spin in an infinite loop. Release
2340 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2345 Description: statfs data retrieved from /proc may be stale or zero
2346 Details : When reading per-device statfs data from /proc, in the
2347 {kbytes,files}_{total,free,avail} files, it may appear
2348 as zero or be out of date.
2351 Frequency : systems with MD RAID1 external journal devices
2353 Description: lconf's call to blkid is confused by RAID1 journal devices
2354 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2355 of returning all block devices that match the journal UUID.
2358 Frequency : always, for aggregate stripe size over 4GB
2360 Description: assertion fails when trying to use 4GB stripe size
2361 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2362 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2365 Frequency : always on ppc64
2367 Description: the first write on an ext3 filesystem with mballoc got stuck
2368 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2369 perform endianness conversion.
2371 ------------------------------------------------------------------------------
2373 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2375 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2376 INTEROPERATE with older versions automatically. Please read the
2377 user documentation before upgrading any part of a live system.
2378 * WARNING: Lustre networking configuration changes are required with
2379 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2382 * Support for kernels:
2383 2.6.9-22.0.2.EL (RHEL 4)
2384 2.6.5-7.244 (SLES 9)
2385 2.6.12.6 vanilla (kernel.org)
2388 Severity : enhancement
2389 Bugzilla : 7981/8208
2390 Description: Introduced Lustre Networking (LNET)
2391 Details : LNET is new networking infrastructure for Lustre, it includes
2392 a reorganized network configuration mode (see the user
2393 documentation for full details) as well as support for routing
2394 between different network fabrics. Lustre Networking Devices
2395 (LNDs) for the supported network fabrics have also been
2396 created for this new infrastructure.
2398 Severity : enhancement
2399 Description: Introduced Access control lists
2400 Details : clients can set ACLs on files and directories in order to have
2401 more fine-grained permissions than the standard Unix UGO+RWX.
2402 The MDS must be started with the "-o acl" mount option.
2404 Severity : enhancement
2405 Description: Introduced filesystem quotas
2406 Details : Administrators may now establish per-user quotas on the
2409 Severity : enhancement
2411 Description: Configuration change for the XT3
2412 The PTLLND is now used to run Lustre over Portals on the XT3
2413 The configure option(s) --with-cray-portals are no longer used.
2414 Rather --with-portals=<path-to-portals-includes> is used to
2415 enable building on the XT3. In addition to enable XT3 specific
2416 features the option --enable-cray-xt3 must be used.
2421 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2422 Details : A race between allocating a new llcd and re-getting the llcd_lock
2423 allowed another thread to grab newly-allocated llcd.
2425 Severity : enhancement
2427 Description: 2.6 OST async journal commit and locking fix to improve performance
2428 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2429 2.6 kernels are now async as they already were in 2.4 kernels so
2430 that they can commit concurrently with the network bulk transfer.
2431 For block-allocated files the filter allocation semaphore is held
2432 to avoid filesystem fragmentation during allocation. BKL lock
2433 removed for 2.6 xattr operations where it is no longer needed.
2438 Description: lconf incorrectly determined whether two IP networks could talk
2439 Details : In some more complicated routing and multiple-network
2440 configurations, lconf will avoid trying to make a network
2441 connection to a disjoint part of the IP space. It was doing the
2442 math incorrectly for one set of cases.
2447 Description: Fix for potential infinite loop processing records in an llog.
2448 Details : If an llog record is corrupted/zeroed, it is possible to loop
2449 forever in llog_process(). Validate the llog record length
2450 and skip the remainder of the block on error.
2453 Frequency : occasional (liblustre only)
2455 Description: liblustre could not open files whose last component is a symlink
2456 Details : sysio_path_walk() would incorrectly pass the open intent to
2457 intermediate path components.
2460 Frequency : rare (liblustre only with non-standard tuning)
2461 Bugzilla : 7201 (7350)
2462 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2463 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2464 after it has been released, only do this for referenced locks
2466 Severity : enhancement
2468 Description: specifying an (invalid) directory default stripe_size of -1
2469 would reset the directory default striping
2470 Details : stripe_size -1 was used internally to signal directory stripe
2471 removal, now use "all default" to signal dir stripe removal
2472 as a directory striping of "all default" is not useful
2475 Frequency : common for large clusters running liblustre clients
2477 Description: doing an ls when liblustre clients are running is slow
2478 Details : sending a glimpse AST to a liblustre client waits for every AST
2479 to time out, as liblustre clients will not respond. Since they
2480 cannot cache data we refresh the OST lock LVB from disk instead.
2482 Severity : enhancement
2484 Description: doing an ls at the same time as file IO can be slow
2485 Details : enqueue and other "small" requests can be blocked behind many
2486 large IO requests. Create a new OST IO portal for non-IO
2487 requests so they can be processed faster.
2490 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2492 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2493 created file with mode 000
2494 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2495 allow an ftruncate() as a non-root user to a file with mode 000.
2496 Lustre now respects this flag to disable mode checks when
2497 truncating a file owned by the user
2500 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2503 Description: Revoking locks from clients that went dead or catatonic might take
2505 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2506 cancellation of such locks instant on servers without waiting for
2507 any reply from clients. Clients drops these locks when cancel
2508 notification from server is received without replying.
2511 Frequency : liblustre-only, when liblustre client dies or becomes busy
2513 Description: Doing ls on Linux clients can take a long time with active
2515 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2516 granting such locks to it in the first place if possible. Locks
2517 are taken by proxy on the OST during the read or write and
2518 dropped immediately afterward. Add connect flags handling, do
2519 not grant locks to liblustre clients for glimpse ASTs.
2521 Severity : enhancement
2523 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2524 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2525 sequential reads and behaves badly with applications doing random
2526 reads. Improve it by reading ahead at least read region, and
2527 avoiding excessive large RPC for small reads.
2529 Severity : enhancement
2531 Description: Creating more than 1000 files for a single job may cause a load
2532 imbalance on the OSTs if there are also a large number of OSTs.
2533 Details : qos_prep_create() uses an OST index reseed value that is an
2534 even multiple of the number of available OSTs so that if the
2535 reseed happens in the middle of the object allocation it will
2536 still utilize the OSTs as uniformly as possible.
2541 Description: OST or MDS may oops in ping_evictor_main()
2542 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2543 but doesn't restart at beginning of obd_exports_timed list
2546 Severity : enhancement
2548 Description: improve by-nid export eviction on the MDS and OST
2549 Details : allow multiple exports with the same NID to be evicted at one
2550 time without re-searching the exports list.
2553 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2555 Description: MDS may oops in groups_free()
2556 Details : in rare race conditions a newly allocated group_info struct is
2557 freed again, and this can be NULL. The 2.4 compatibility code
2558 for groups_free() checked for a NULL pointer, but 2.6 did not.
2561 Frequency : common for liblustre clients doing little filesystem IO
2562 Bugzilla : 9352, 7313
2563 Description: server may evict liblustre clients accessing contended locks
2564 Details : if a client is granted a lock or receives a completion AST
2565 with a blocking AST already set it would not reply to the AST
2566 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2569 Frequency : lfs setstripe, only systems with more than 160 OSTs
2571 Description: unable to set striping with a starting offset beyond OST 160
2572 Details : llapi_create_file() incorrectly limited the starting stripe
2573 index to the maximum single-file stripe count.
2576 Frequency : LDAP users only
2578 Description: lconf did not handle in-kernel recovery with LDAP properly
2579 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2581 Severity : enhancement
2583 Description: bind OST threads to NUMA nodes to improve performance
2584 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2585 node and do their allocations there to localize memory access
2587 Severity : enhancement
2589 Description: llmount can determine client NID directly from Myrinet (GM)
2590 Details : the client NID code from gmnalnid was moved directly into
2591 llmount, removing the need to use this or specifying the
2592 client NID explicitly when mounting GM clients with zeroconf
2595 Frequency : if client is started with down MDS
2597 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2598 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2599 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2600 ptlrpc_import_delayed_req() because we didn't check if the
2601 request was interrupted, and we also didn't break out of the
2602 event loop if there was no timeout
2607 Description: data loss during non-page-aligned writes to a single file from
2608 both multiple nodes and multiple threads on one node at same time
2609 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2610 inconsistency led to false short-reads, that were cached and later
2611 used by ->prepare_write() to fill in partially written page,
2612 leading to data loss.
2615 Frequency : always, if lconf --abort_recovery used
2617 Description: lconf --abort_recovery fails with 'Operation not supported'
2618 Details : lconf was attempting to abort recovery on the MDT device and not
2621 Severity : enhancement
2623 Description: remove cleanup logs
2624 Details : replace lconf-generated cleanup logs with lustre internal
2625 cleanup routines. Eliminates the need for client-cleanup and
2628 Severity : enhancement
2630 Description: add support for EAs (user and system) on lustre filesystems
2631 Details : it is now possible to store extended attributes in the Lustre
2632 client filesystem, and with the user_xattr mount option it
2633 is possible to allow users to store EAs on their files also
2635 Severity : enhancement
2637 Description: Add possibility (config option) to show minimal available OST free
2639 Details : When compiled with --enable-mindf configure option, statfs(2)
2640 (and so, df) will return least minimal free space available from
2641 all OSTs as amount of free space on FS, instead of summary of
2642 free spaces of all OSTs.
2644 Severity : enhancement
2646 Description: do not expand extent locks acquired on OST-side
2647 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2648 by server: they are not cached anyway.
2651 Frequency : when mmap is used/binaries executed from Lustre
2653 Description: Unmmap pages before throwing them away from read cache.
2654 Details : llap_shrink cache now attempts to unmap pages before discarding
2655 them (if unmapping failed - do not discard). SLES9 kernel has
2656 extra checks that trigger if this unmapping is not done first.
2661 Description: lconf didn't resolve symlinks before checking to see whether a
2662 given mountpoint was already in use
2665 Frequency : when migrating failover services
2666 Bugzilla : 6395, 9514
2667 Description: When migrating a subset of services from a node (e.g. failback
2668 from a failover service node) the remaining services would
2669 time out and evict clients.
2670 Details : lconf --force (implied by --failover) sets the global obd_timeout
2671 to 5 seconds in order to quickly disconnect, but this caused
2672 other RPCs to time out too quickly. Do not change the global
2673 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2675 Severity : enhancement
2676 Frequency : if MDS is started with down OST
2677 Bugzilla : 9439,5706
2678 Description: Allow startup/shutdown of an MDS without depending on the
2679 availability of the OSTs.
2680 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2681 Add appropriate locking and lov-osc refcounts for safe
2682 cleaning. Add osc abort_inflight calls in case the
2683 synchronize never started.
2686 Frequency : occasional (Cray XT3 only)
2688 Description: root not authorized to access files in CRAY_PORTALS environment
2689 Details : The client process capabilities were not honoured on the MDS in
2690 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2691 been accessed by an authorized user then root was able to access
2692 the file on the local client also. The root user capabilities
2693 are now allowed on the MDS, as this environment has secure UID.
2696 Frequency : occasional
2698 Description: ldiskfs "too long searching" message happens too often
2699 Details : A debugging message (otherwise harmless) prints too often on
2700 the OST console. This has been reduced to only happen when
2701 there are fragmentation problems on the filesystem.
2706 Description: Division by zero in statfs when all OSCs are inactive
2707 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2708 lov_statfs divides by value returned by lov_get_stripecnt().
2712 Bugzilla : 9489, 3273
2713 Description: First write from each client to each OST was only 4kB in size,
2714 to initialize client writeback cache, which caused sub-optimal
2715 RPCs and poor layout on disk for the first writen file.
2716 Details : Clients now request an initial cache grant at (re)connect time
2717 and so that they can start streaming writes to the cache right
2718 away and always do full-sized RPCs if there is enough data.
2719 If the OST is rebooted the client also re-establishes its grant
2720 so that client cached writes will be honoured under the grant.
2725 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2726 Details : Now I/O RPCs go to different portal number and (presumably) fast
2727 lock requests (and glimses) and other RPCs get their own service
2728 threads pool that should be able to service those RPCs
2731 Severity : enhancement
2733 Description: Ability to exchange lustre version between client and servers and
2734 issue warnings at client side if client is too old. Also for
2735 liblustre clients there is ability to refuse connection of too old
2737 Details : New 'version' field is added to connect data structure that is
2738 filled with version info. That info is later checked by server and
2742 Frequency : rare, liblustre only.
2743 Bugzilla : 9296, 9581
2744 Description: Two simultaneous writes from liblustre at offset within same page
2745 might proceed at the same time overwriting eachother with stale
2747 Details : I/O lock withing llu_file_prwv was released too early, before data
2748 actually was hitting the wire. Extended lock-holding time until
2749 server acknowledges receiving data.
2752 Frequency : extremely rare. Never observed in practice.
2754 Description: avoid generating lustre_handle cookie of 0.
2755 Details : class_handle_hash() generates handle cookies by incrementing
2756 global counter, and can hit 0 occasionaly (this is unlikely, but
2757 not impossible, because initial value of cookie counter is
2758 selected randonly). Value of 0 is used as a sentinel meaning
2759 "unassigned handle" --- avoid it. Also coalesce two critical
2760 sections in this function into one.
2762 Severity : enhancement
2764 Description: allow liblustre clients to delegate truncate locking to OST
2765 Details : To avoid overhead of locking, liblustre client instructs OST to
2766 take extent lock in ost_punch() on client's behalf. New connection
2767 flag is added to handle backward compatibility.
2769 Severity : enhancement
2770 Bugzilla : 4928, 7341, 9758
2771 Description: allow number of OST service threads to be specified
2772 Details : a module parameter allows the number of OST service threads
2773 to be specified via "options ost ost_num_threads={N}" in the
2774 OSS's /etc/modules.conf or /etc/modprobe.conf.
2778 Bugzilla : 6146, 9635, 9895
2779 Description: servers crash with bad pointer in target_handle_connect()
2780 Details : In rare cases when a client is reconnecting it was possible that
2781 the connection request was the last reference for that export.
2782 We would temporarily drop the export reference and get a new
2783 one, but this may have been the last reference and the export
2784 was just destroyed. Get new reference before dropping old one.
2786 Severity : enhancement
2787 Frequency : if client is started with failover MDS
2789 Description: Allow multiple MDS hostnames in the mount command
2790 Details : Try to read the configuration from all specified MDS
2791 hostnames during a client mount in case the "primary"
2794 Severity : enhancement
2796 Description: Stop sending data to evicted clients as soon as possible.
2797 Details : Check if the client we are about to send or are sending data to
2798 was evicted already. (Check is done every second of waiting,
2799 for which l_wait_event interface was extended to allow checking
2800 of exit condition at specified intervals).
2803 Frequency : rare, normally only when NFS exporting is done from client
2805 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2807 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2810 Frequency : always, if srand() is called before liblustre initialization
2812 Description: Liblustre uses system PRNG disturbing its usage by user application
2813 Details : Introduce internal to lustre fast and high-quality PRNG for
2814 lustre usage and make liblustre and some other places in generic
2815 lustre code to use it.
2817 Severity : enhancement
2818 Bugzilla : 9477, 9557, 9870
2819 Description: Verify that the MDS configuration logs are updated when xml is
2820 Details : Check if the .xml configuration logs are newer than the config
2821 logs stored on the MDS and report an error if this is the case.
2822 Request --write-conf, or allow starting with --old_conf.
2824 Severity : enhancement
2826 Description: Handle symlinks in the path when checking if Lustre is mounted.
2827 Details : Resolve intermediate symlinks when checking if a client has
2828 mounted a filesystem to avoid duplicate client mounts.
2833 Description: lconf can hit an error exception but still return success.
2834 Details : The lconf command catches the Command error exception at the top
2835 level script context and will exit with the associated exit
2836 status, but doesn't ensure that this exit status is non-zero.
2841 Description: failure of ptlrpc thread startup can cause oops
2842 Details : Starting a ptlrpc service thread can fail if there are a large
2843 number of threads or the server memory is very fragmented.
2844 Handle this without oopsing.
2847 Frequency : always, only if liblustre and non-default acceptor port was used
2849 Description: liblustre cannot connect to servers with non-default acceptor port
2850 Details : tcpnal_set_default_params() was not called and was therefore
2851 ignoring the environment varaible TCPNAL_PORT, as well as other
2852 TCPNAL_ environment variables
2857 Description: two objects could be created on the same OST for a single file
2858 Details : If an OST is down, in some cases it was possible to create two
2859 objects on a single OST for a single file. No problems other
2860 than potential performance impact and spurious error messages.
2864 Bugzilla : 5681, 9562
2865 Description: Client may oops in ll_unhash_aliases
2866 Details : Client dcache may become inconsistent in race condition.
2867 In some cases "getcwd" can fail if the current directory is
2873 Description: Inode refcounting problems in NFS export code
2874 Details : link_raw functions used to call d_instantiate without obtaining
2875 extra inode reference first.
2879 Bugzilla : 9942, 9903
2880 Description: Referencing freed requests leading to crash, memleaks with NFS.
2881 Details : We used to require that call to ll_revalidate_it was always
2882 followed by ll_lookup_it. Also with revalidate_special() it is
2883 possible to call ll_revalidate_it() twice for the same dentry
2884 even if first occurence returned success. This fix changes semantic
2885 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2886 reference on a request referred from the intent.
2887 ll_intent_release() then releases such a request.
2890 Frequency : rare, normally benchmark loads only
2892 Description: unlinked inodes were kept in memory on the client
2893 Details : If a client is repeatedly creating and unlinking files it
2894 can accumulate a lot of stale inodes in the inode slab cache.
2895 If there is no other client load running this can cause the
2896 client node to run out of memory. Instead flush old inodes
2897 from client cache that have the same inode number as a new inode.
2900 Frequency : SLES9 2.6.5 kernel and long filenames only
2901 Bugzilla : 9969, 10379
2902 Description: utime reports stale NFS file handle
2903 Details : SLES9 uses out-of-dentry names in some cases, which confused
2904 the lustre dentry revalidation. Change it to always use the
2908 Frequency : rare, unless heavy write-truncate concurrency is continuous
2909 Bugzilla : 4180, 6984, 7171, 9963, 9331
2910 Description: OST becomes very slow and/or deadlocked during object unlink
2911 Details : filter_destroy() was holding onto the parent directory lock
2912 while truncating+unlinking objects. For very large objects this
2913 may block other threads for a long time and slow overall OST
2914 responsiveness. It may also be possible to get a lock ordering
2915 deadlock in this case, or run out of journal credits because of
2916 the combined truncate+unlink. Solution is to do object truncate
2917 first in one transaction without parent lock, and then do the
2918 final unlink in a new transaction with the parent lock. This
2919 reduces the lock hold time dramatically.
2922 Frequency : rare, 2.4 kernels only
2924 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2925 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2926 before unmounting their filesystem in order to avoid deadlock.
2927 The kernel_locked() function in 2.4 kernels only checks whether
2928 the kernel lock is held, not whether it is this process that is
2929 holding it as 2.6 kernels do.
2934 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2935 Details : The client ptlrpc code may be trying to reconnect to a down
2936 server before a previous connection attempt has timed out.
2937 Increase the reconnect interval to be longer than the connection
2938 timeout interval to avoid sending duplicate connections to
2942 Frequency : echo_client brw_test command
2944 Description: fix echo_client to work with OST preallocated code
2945 Details : OST preallocation code (5137) didn't take echo_client IO path
2946 into account: echo_client calls filter methods outside of any
2947 OST thread and, hence, there is no per-thread preallocated
2948 pages and buffers to use. Solution: hijack pga pages for IO. As
2949 a byproduct, this avoids unnecessary data copying.
2953 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2954 Description: Client can oops in mdc_commit_close() after open replay
2955 Details : It was possible for the MDS to return an open request with no
2956 transaction number in mds_finish_transno() if the client was
2957 evicted, but without actually returning an error. Clients
2958 would later try to replay that open and may trip an assertion
2959 Simplify the client close codepath, and always return an error
2960 from the MDS in case the open is not successful.
2963 Frequency : rare, 2.6 OSTs only
2965 Description: OST may deadlock under high load on fragmented files
2966 Details : If there was a heavy load and highly-fragmented OST filesystems
2967 it was possible to have all the OST threads deadlock waiting on
2968 allocation of biovecs, because the biovecs were not released
2969 until the entire RPC IO was completed. Instead, release biovecs
2970 as soon as they are complete to ensure forward IO progress.
2972 Severity : enhancement
2974 Description: Support for specifying external journal device at mount
2975 Details : If an OST or MDS device is formatted with an external journal
2976 device, this device major/minor is stored in the ext3 superblock
2977 and may not be valid for failover. Allow detecting and
2978 specifying the external journal at mount time.
2983 Description: Mounting an MDS with pending unlinked files may cause oops
2984 Details : target_finish_recovery() calls mds_postrecov() which returned
2985 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2986 considers this an error and immediately begins cleaning up the
2987 lov, just after starting the mds_lov process
2989 Severity : enhancement
2991 Description: Implement 'lfs df' to report actual free space on per-OST basis
2992 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2993 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2994 usage of objects; '-h' to report in human readable format.
2996 ------------------------------------------------------------------------------
2998 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3005 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3006 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3007 at startup during orphan unlinking. Instead, skip buddy update
3008 before setup, as it will be regenerated after recovery is complete.
3013 Description: If an OST is inactive, its locks might reference stale inodes.
3014 Details : lov_change_cbdata() must iterate over all namespaces, even if
3015 they are inactive to clear inode references from the lock.
3017 Severity : enhancement
3018 Frequency : occasional, if non-standard max_dirty_mb used
3020 Description: Client will block write RPCs if not enough grant
3021 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3022 then the client will block writes while waiting for another RPC
3023 to complete instead of consuming its dirty limit. With change
3024 we get improved performance when max_dirty_mb is small.
3026 Severity : enhancement
3027 Bugzilla : 3389, 6253
3028 Description: Add support for supplementary groups on the MDS.
3029 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3030 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3031 lookups for user supplementary groups into a cache.
3035 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3036 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3037 for liblustre O_EXCL is enforced by MDS.
3041 Description: readdir never returns NULL in liblustre.
3042 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3043 dentry in struct dirent.
3047 Frequency : liblustre only, depends on application IO pattern
3048 Description: liblustre clients evicted if not contacting servers
3049 Details : Don't put liblustre clients into the ping_evictor list, so
3050 they will not be evicted by the pinger ever.
3052 Severity : enhancement
3054 Description: Add ability to evict clients by NID from MDS.
3055 Details : By echoing "nid:$NID" string into
3056 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3057 $NID would be instantly evicted from this MDS and from all active
3058 OSTs connected to it.
3062 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3063 Details : lookup_it_finish() used to query file size from OSTs that was not
3068 Description: service threads change working directory to that of init
3069 Details : Starting lustre service threads may pin the working directory
3070 of the parent thread, making that filesystem busy. Threads
3071 now change to the working directory of init to avoid this.
3075 Frequency : during shutdown only
3076 Description: shutdown with a failed MDS or OST can cause unmount to hang
3077 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3082 Frequency : 2.6 only
3083 Description: chmod/chown may include an extra supplementary group
3084 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3085 supplementary group and if none is specified this is used.
3088 Bugzilla : 5479 (6816)
3090 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3091 Details : If lookup is in progress on a file that is unlinked we might try
3092 to revalidate the inode and fail in revalidate after lookup is
3093 complete and ll_file_open() enqueues the open again but
3094 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3097 Frequency : always, if lconf --abort_recovery used
3099 Description: lconf --abort_recovery fails with 'Operation not supported'
3100 Details : lconf was attempting to abort recovery on the MDT device and not
3103 ------------------------------------------------------------------------------
3105 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3110 Frequency : rare (only unsupported configurations with a node running as an
3112 Bugzilla : 6514, 5137
3113 Description: Mounting a Lustre file system on a node running as an OST could
3115 Details : OSTs now preallocates memory needed to write out data at
3116 startup, instead of when needed, to avoid having to
3117 allocate memory in possibly low memory situations.
3118 Specifically, if the file system is mounted on on OST,
3119 memory pressure could force it to try to write out data,
3120 which it needed to allocate memory to do. Due to the low
3121 memory, it would be unable to do so and the node would
3122 become unresponsive.
3124 Severity : enhancement
3126 Description: Addition of lconf --service command line option
3127 Details : lconf now accepts a '--service <arg>' option, which is
3128 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3130 Severity : enhancement
3132 Description: Failover mode is now the default for OSTs.
3133 Details : By default, OSTs will now run in failover mode. To return to
3134 the old behaviour, add '--failout' to the lmc line for OSTs.
3136 Severity : enhancement
3138 Description: Health checks are now provided for MDS and OSTs
3139 Details : Additional detailed health check information on MSD and OSTs
3140 is now provided through the procfs health_check value.
3143 Frequency : occasional, depends on IO load
3145 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3146 after numerous create/delete cycles
3147 Details : The ext3 inode allocation policy would not allocate new inodes
3148 very well on the OSTs because there are no new directories
3149 being created. Instead we look for groups with free space if
3150 the parent directories are nearly full.
3155 Description: Network or server problems during mount may cause partially
3156 mounted clients instead of returning an error.
3157 Details : The config llog parsing code may overwrite the error return
3158 code during mount error handling, returning success instead
3164 Description: MDS can fail to allocate large reply buffers
3165 Details : After long uptimes the MDS can fail to allocate large reply
3166 buffers (e.g. zconf client mount config records) due to memory
3167 fragmentation or consumption by the buffer cache. Preallocate
3168 some large reply buffers so that these replies can be sent even
3169 under memory pressure.
3173 Frequency : rare (liblustre)
3174 Description: fsx running with liblustre complained that using truncate() to
3175 extend the file doesn't work. This patch corrects that issue.
3176 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3177 ATTR_SIZE and lsm use in llu_setattr_raw.
3181 Frequency : rare, only 2.6 kernels
3182 Description: Unusual file access patterns on the MDS may result in inode
3183 data being lost in very rare circumstances.
3184 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3185 code in the 2.6 kernel caused the inode and/or EA data not to
3186 be read from disk, causing single-file corruption.
3190 Frequency : rare, only 2.6 filesystems using extents
3191 Description: Heavy concurrent write and delete load may cause data corruption.
3192 Details : It was possible under high-load situations to have an extent
3193 metadata block in the block device cache from a just-unlinked
3194 file overwrite a newly-allocated data block. We now unmap any
3195 metadata buffers that alias just-allocated data blocks.
3199 Frequency : filesystems with default stripe_count larger than 77
3200 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3201 Details : lconf specifies an inode size of 4096 bytes when the default
3202 stripe_count is larger than 77. This conflicts with the default
3203 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3204 this case to avoid pinning too much memory for large EAs.
3206 ------------------------------------------------------------------------------
3208 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3213 Frequency : rare (extremely heavy IO load with hundreds of clients)
3215 Description: Client is evicted, gets IO error writing to file
3216 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3217 caused clients to be evicted for AST timeouts. The fixes in
3218 bug 5192 mean we no longer need to have such short AST timeouts
3219 so ldlm_timeout has been increased.
3222 Frequency : occasional during --force or --failover shutdown under load
3223 Bugzilla : 5949, 4834
3224 Description: Server oops/LBUG if stopped with --force or --failover under load
3225 Details : a collection of import/export refcount and cleanup ordering
3226 issues fixed for safer force cleanup
3229 Frequency : only filesystems larger than 120 OSTs
3230 Bugzilla : 5990, 6223
3231 Description: lfs getstripe would oops on a very large filesystem
3232 Details : lov_getconfig used kfree on vmalloc'd memory
3235 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3236 Bugzilla : 6242, 6243
3237 Description: reading from files that had been truncated to a non-zero size
3238 but never opened returned no data
3239 Details : ll_file_read() reads zeros from no-object files to EOF
3244 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3245 Details : There's little harm in aborting MDS/OSS recovery and letting it
3246 try again, so I removed the LASSERT and return an error instead.
3248 Severity : enhancement
3250 Description: New debugging infrastructure for tracking down data corruption
3251 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3252 (b) cover more of the client-side code path, and (c) try to narrow
3253 down where problems occurred
3257 Bugzilla : 3819, 4364, 4397, 6313
3258 Description: Racing close and eviction MDS could cause assertion in mds_close
3259 Details : It was possible to get multiple mfd references during close and
3260 client eviction, leading to one thread referencing a freed mfd.
3262 Severity: : enhancement
3263 Bugzilla : 3262, 6359
3264 Description: Attempts to reconnect to servers are now more aggressive.
3265 Details : This builds on the enhanced upcall-less recovery that was added
3266 in 1.4.2. When trying to reconnect to servers, clients will
3267 now try each server in the failover group every 10 seconds. By
3268 default, clients would previously try one server every 25 seconds.
3273 Description: After recovery, certain operations trigger a failed
3274 assertion on a client.
3275 Details : Failing over an mds, using lconf -d --failover, while a
3276 client was doing a readdir() call would cause the client to
3277 LBUG after recovery completed and the readdir() was resent.
3279 Severity : enhancement
3281 Description: Default groups are now added by lconf
3282 Details : You can now run lconf --group <servicename> without having to
3283 manually add groups with lmc.
3286 Frequency : occasional
3288 Description: Nodes with an elan id of 0 trigger a failed assertion
3291 Frequency : always when accessing e.g. tty/console device nodes
3293 Description: tty and some other devices nodes cannot be used on lustre
3294 Details : file's private_data field is used by device data and lustre
3295 values in there got lost. New field was added to struct file to
3296 store fs-specific private data.
3299 Frequency : when exporting Lustre via NFS
3301 Description: NFSD failed occasionally when looking up a path component
3302 Details : NFSD is looking up ".." which was broken in ext3 directories
3303 that had grown large enough to become hashed.
3306 Frequency : Clusters with multiple interfaces not on the same subnet
3308 Description: Nodes will repeatedly try to reconnect to an interface which it
3309 cannot reach and report an error to the log.
3310 Details : Extra peer list entries will be created by lconf with some peers
3311 unreachable. lconf now validates the peer before adding it.
3314 Frequency : Only if a default stripe is set on the filesystem root.
3316 Description: Setting a default stripe on the filesystem root prevented the
3317 filesystem from being remounted.
3318 Details : The client was sending extra request flags in the root getattr
3319 request and did not allocate a reply buffer for the dir EA.
3322 Frequency : occasional, higher if lots of files are accessed by one client
3323 Bugzilla : 6159, 6097
3324 Description: Client trips assertion regarding lsm mismatch/magic
3325 Details : While revalidating inodes the VFS looks up inodes with ifind()
3326 and in rare cases can find an inode that is being freed.
3327 The ll_test_inode() code will free the lsm during ifind()
3328 when it finds an existing inode and then the VFS later attaches
3329 this free lsm to a new inode.
3333 Bugzilla : 6422, 7030
3334 Description: MDS deadlock between mkdir and client eviction
3335 Details : Creating a new file via mkdir or mknod (starting a transaction
3336 and getting the ns lock) can deadlock with client eviction
3337 (gets ns lock and trying to finish a synchronous transaction).
3340 Frequency : occasional
3341 Description: While starting a server, the fsfilt_ext3 module could not be
3343 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3344 kernels. Previously, lconf would still use the ext3 name
3345 when trying to load modules. Now, it will correctly use
3346 ext3 on 2.4 and ldiskfs on 2.6.
3348 Severity : enhancement
3349 Description: The default stripe count has been changed to 1
3350 Details : The interpretation of the default stripe count (0, to lfs
3351 or lmc) has been changed to mean striping across a single
3352 OST, rather than all available. For general usage we have
3353 found a stripe count of 1 or 2 works best.
3355 Severity : enhancement
3356 Description: Add support for compiling against Cray portals.
3357 Details : Conditional compiling for some areas that are different
3361 Frequency : occasional
3362 Bugzilla : 6409, 6834
3363 Description: Creating files with an explicit stripe count may lead to
3364 a failed assertion on the MDS
3365 Details : If some OSTs are full or unavailable, creating files may
3366 trigger a failed assertion on the MDS. Now, Lustre will
3367 try to use other servers or return an error to the
3371 Frequency : occasional
3373 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3374 caused lock timeout during readahead (since 1.4.2).
3375 Details : Processes doing readahead might match a lock that hasn't been
3376 granted yet if there are overlapping and conflicting lock
3377 requests. The readahead process waits on ungranted lock
3378 (original lock is CBPENDING), while OST waits for that process
3379 to cancel CBPENDING read lock and eventually evicts client.
3381 Severity : enhancement
3383 Description: Initial enabling of flock support for clients
3384 Details : Implements fcntl advisory locking and file status functions.
3385 This feature is provided as an optional mount flag (default
3386 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3387 locking are implemented yet, and those that are are not guaranteed
3388 to be completely correct in production environments.
3389 mount -t lustre -o [flock|noflock] ...
3392 Frequency : occasional
3394 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3395 assertion in the ext3 JBD (journaling) layer.
3396 Details : The b_committed_data struct is protected by the big kernel lock
3397 in 2.4 kernels, serializing journal_commit_transaction() and
3398 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3399 access to this struct. In 2.6 kernels there is finer grained
3400 locking to improve SMP performance of the JBD layer.
3404 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3409 Description: merge and cleanup kernel patches.
3412 -----------------------------------------------------------------------------
3414 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3416 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3417 versions of Lustre. Please update all servers and clients to
3418 version 1.4.2 or later at the same time. You must also run
3419 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3420 to update the configuration logs.
3422 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3423 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3424 - swab llog records of type '0' so we get proper header size/idx (5861)
3425 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3426 - fix rename of one directory over another leaking an inode (5953)
3427 - avoid SetPageDirty on 2.6 (5981)
3428 - don't re-add just-being-destroyed locks to the waiting list (5653)
3429 - when creating new directories, inherit the parent's custom
3430 striping settings if present parent (3048)
3431 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3432 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3433 - don't hold client locks on temporary worklist from l_lru (5666)
3434 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3435 - automatically evict dead clients (5921)
3436 - Update file size properly in create+truncate+fstat case (6196)
3437 - Do not unhash mountpoint dentries, do not allow removal of
3439 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3440 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3441 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3442 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3443 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3444 - Correctly handle reads of files with no objects (6243)
3445 - lctl recover will also mark a device active if deactivate used (5933)
3447 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3448 - allow --write-conf on an MDS with different nettype than client (5619)
3449 - don't write config llogs to MDS for mounts not from that MDS (5617)
3450 - lconf should create multiple TCP connections from a client (5201)
3451 - init scripts are now turned off by default; run chkconfig --on
3452 lustre and chkconfig --on lustrefs to use them
3453 - upcalls are no longer needed for clients to recover to failover
3455 - add --abort-recovery option to lconf to abort recovery on device
3457 - add support for an arbitrary number of OSTs (3026)
3458 - Quota support protocol changes.
3459 - forward compatibility changes to wire structs (6007)
3460 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3461 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3462 - improved "lustre status" script
3463 - initialize blocksize for non-regular files (6062)
3464 - added --disable-server and --disable-client configure options (5782)
3465 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3466 - Vanilla 2.4.29 support
3467 - increase maximum number of obd devices to 520 (6242)
3468 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3469 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3470 SLES 9 2.6.5 kernels
3471 - limit stripes per file to 160 (the maximum EA size) (6093)
3473 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3476 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3477 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3478 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3479 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3480 - small ext3 extents cleanups and fixes (5733)
3481 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3482 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3483 - avoid needless client->OST connect, fix handle mismatch (5317)
3484 - fix DLM error path that led to out-of-sync client, long delays (5779)
3485 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3486 - fix several locking issues related to i_size (5492,5624,5654,5672)
3487 - don't move pending lock onto export if it is already evicted (5683)
3488 - fix kernel oops when creating .foo in unlinked directory (5548)
3489 - fix deadlock in obdfilter statistics vs. object create (5811)
3490 - use time_{before,after} to avoid timer jiffies wrap (5882)
3491 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3492 - Do not leak request if server was not able to process it (5154)
3493 - If mds_open unable to find parent dir, make that negative lookup(5154)
3494 - don't create new directories with extent-mapping (5909, 5936)
3496 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3497 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3498 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3499 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3500 - support for RHEL4 kernel on i686 (5773)
3501 - provide error messages when incompatible logs are encountered (5898)
3503 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3504 * version 1.4.0.10 (1.4.1 release candidate 1)
3506 - don't keep a lock reference when lock is not granted (4238)
3507 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3508 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3509 - drop import inflight refcount on signal_completed_replay error (5255)
3510 - unlock page after async write error during send (3677)
3511 - handle missing objects in filter_preprw_read properly (5265)
3512 - no transno return for symlink open, don't save no-trasno open (3440)
3513 - don't try to complete elan receive that already failed (4012)
3514 - free RPC server reply state on error (5406)
3515 - clean up thread from ptlrpc_start_thread() on error (5160)
3516 - readahead could read extra page into cache that wasn't ejected (5388)
3517 - prevent races in class_attach/setup/cleanup/detach (5260)
3518 - don't dereference de->d_inode after l_dput of de (5458)
3519 - use "int" for stripe value returned from lock_to_stripe (5544)
3520 - mballoc allocation and error-checking fixes in 2.6 (5504)
3521 - block device patches to fix I/O request sizes in 2.6 (5482)
3522 - look up hostnames for IB nals (5602)
3523 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3524 - don't start multiple acceptors for the same port (5277)
3525 - fix incorrect LASSERT in mds_getattr_name (5635)
3526 - export a proc file for general "ping" checking (5628)
3527 - fix "lfs check" to not block when the MDS is down (5628)
3529 - service request history (4965)
3530 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3531 - create an "evict_client" /proc entry on OSTs, like the MDS has
3532 - fix mount usage message, return errors per mount(8) (5168)
3533 - change grep [] to grep "[]" in tests so they work in more UMLs
3534 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3535 - remove ancient LOV_MAGIC_V0 EA support (5047)
3536 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3537 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3538 - allow readahead allocations to fail when low on memory (5383)
3539 - mmap locking landed again, after considerable improvement (2828)
3540 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3542 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3545 - send OST transaction number in read/write reply to free req (4966)
3546 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3547 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3548 - the watchdog thread now runs as interruptible (5246)
3549 - flock/lockf fixes (but it's still disabled, pending 5135)
3550 - don't use EXT3 constants in llite code (5094)
3551 - memory shortage at startup could cause assertion (5176)
3553 - reorganization of lov code
3554 - single portals codebase
3556 - add extents/mballoc support (5025)
3557 - direct I/O reads in the obdfilter (4048)
3558 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3560 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3563 - send OST transaction number in read/write reply to free req (4966)
3564 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3565 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3566 - the watchdog thread now runs as interruptible (5246)
3567 - handle missing objects in filter_preprw_read properly (5265)
3568 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3569 - drop import inflight refcount on signal_completed_replay error (5255)
3570 - unlock page after async write error during send (3677)
3571 - return original error code on reconstructed replies (3761)
3572 - no transno return for symlink open, don't save no-trasno open (3440)
3574 - add pid to ldlm debugging output (4922)
3575 - bump the watchdog timeouts -- we can't handle 30sec yet
3576 - extra debugging for orphan dentry/inode bug (5259)
3578 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3581 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3582 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3583 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3584 - fix NULL dereference in /proc/sys/portals/routes (4827)
3585 - allow failed mdc_close() operations to be interrupted (4561)
3586 - stop precreate on OST before MDS would time out on it (4778)
3587 - don't send partial-page writes before EOF from client (4410)
3588 - discard client grant for sub-page writes on large-page clients (4520)
3589 - don't free dentries not owned by NFS code, check generation (4806)
3590 - fix lsm leak if mds_create_objects() fails (4801)
3591 - limit debug_daemon file size, always print CERROR messages (4789)
3592 - use transno after validating reply (3892)
3593 - process timed out requests if import state changes (3754)
3594 - update mtime on OST during writes, return in glimpse (4829)
3595 - add mkfsoptions to LDAP (4679)
3596 - use ->max_readahead method instead of zapping global ra (5039)
3597 - don't interrupt __l_wait_event() during strace
3599 - add software watchdogs to catch hung threads quickly (4941)
3600 - make lustrefs init script start after nfs is mounted
3601 - fix CWARN/ERROR duplication (4930)
3602 - return async write errors to application if possible (2248)
3603 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3604 - print NAL number in %x format (4645)
3605 - update barely-supported suse-2.4.21-171 series (4842)
3606 - support for sles 9 %post scripts
3607 - support for building 2.6 kernel-source packages
3608 - support for sles km_* packages
3610 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3613 - ignore -ENOENT errors in osc_destroy (3639)
3614 - notify osc create thread that OSC is being cleaned up (4600)
3615 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3616 - reconstruct ost_handle() like mds_handle() (4657)
3617 - create a new thread to do import eviction to avoid deadlock (3969)
3618 - let lconf resolve symlinked-to devices (4629)
3619 - don't unlink "objects" from directory with default EA (4554)
3620 - hold socknal file ref over connect in case target is down (4394)
3621 - allow more than 32000 subdirectories in a single directory (3244)
3622 - fix blocks count for O_DIRECT writes (3751)
3623 - OST returns ENOSPC from object create when no space left (4539)
3624 - don't send truncate RPC if file size isn't changing (4410)
3625 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3626 - bind to privileged port in socknal and tcpnal (3689)
3628 - rate limit CERROR/CWARN console message to avoid overload (4519)
3629 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3630 - basic mmap support (3918)
3631 - kernel patch series update from b1_4 (4711)
3633 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3636 - avoid crash during MDS cleanup with OST shut down (2775)
3637 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3638 - don't use bad inodes on the MDS (3744)
3639 - dynamic object preallocation to improve recovery speed (4236)
3640 - don't hold spinlock over lock dumping or change debug flags (4401)
3641 - don't zero obd_dev when it is force cleaned (3651)
3642 - print grants to console if they go negative (4431)
3643 - "lctl deactivate" will stop automatic recovery attempts (3406)
3644 - look for existing locks in ldlm_handle_enqueue() (3764)
3645 - don't resolve lock handle twice in recovery avoiding race (4401)
3646 - revalidate should check working dir is a directory (4134)
3648 - don't always mark "slow" obdfilter messages as errors (4418)
3650 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3653 - don't close LustreDB during write_conf until it is done (3860)
3654 - fix typo in lconf for_each_profile (3821)
3655 - allow dumping logs from multiple threads at one time (3820)
3656 - don't allow multiple threads in OSC recovery (3812)
3657 - fix debug_size parameters (3864)
3658 - fix mds_postrecov to initialize import for llog ctxt (3121)
3659 - replace config semaphore with spinlock (3306)
3660 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3661 - don't allow enqueue to complete on a destroyed export (3822)
3662 - down write_lock before checking llog header bitmap (3825)
3663 - recover from lock replay timeout (3764)
3664 - up llog sem before sending rpc (3652)
3665 - reduce ns lock hold times when setting kms (3267)
3666 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3667 - fix NULL deref and obd_dev leak on setup error (3312)
3668 - replace some LBUG about llog ops with error handling (3841)
3669 - don't match INVALID dentries from d_lookup and spin (3784)
3670 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3671 - fix invalid assertion in ptlrpc_set_wait (3880)
3673 - add libwrap support for the TCP acceptor (3996)
3674 - add /proc/sys/portals/routes for non-root route listing (3994)
3675 - allow setting MDS UUID in .xml (2580)
3676 - print the stack of a process that LBUGs (4228)
3678 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3681 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3682 - make sure to unset replay flag from failed open requests (3440)
3683 - if default stripe count is 0, use OST count for inode size (3636)
3684 - update parent mtime/ctime on client for create/unlink (2611)
3685 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3686 - free recovery state on server during a forced cleanup (3571)
3687 - unregister_reply for resent reqs (3063)
3688 - loop back devices mounting and status check on 2.6 (3563)
3689 - fix resource-creation race that can provoke i_size == 0 (3513)
3690 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3691 - more debugging for page-accounting assertion (3746)
3692 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3693 - avoid deadlock after precreation failure (3758)
3694 - fix race and lock order deadlock in orphan handling (3450, 3750)
3695 - add validity checks when grabbing inodes from l_ast_data (3599)
3697 - add /proc/.../recovery_status to obdfilter (3428)
3698 - lightweight CDEBUG infrastructure, debug daemon (3668)
3699 - change default OSC RPC parameters to be better on small clusters
3700 - turn off OST read cache for files smaller than 32MB
3701 - install man pages and include them in rpms (3100)
3702 - add new init script for (un)mounting lustre filesystems (2593)
3703 - run chkconfig in %post for init scripts (3701)
3704 - drop scimac NAL (unmaintained)
3706 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3709 - clean kiobufs before and after use (3485)
3710 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3711 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3712 - change init script to fail more clearly if not run as root (1528)
3713 - allow clients to reconnect during replay (1742)
3714 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3715 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3716 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3717 - fix cond_resched (3554)
3718 - only evict unfinished clients after recovery (3515)
3719 - allow bulk resend, prevent data loss (3570)
3720 - dynamic ptlrpc request buffer allocation (2102)
3721 - don't allow unlinking open directory if it isn't empty (2904)
3722 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3723 - remove extraneous obd dereference causing LASSERT failure (3334)
3724 - don't use get_cycles() when creating temp. files on the mds (3156)
3725 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3726 - handle EEXIST for set-stripe, set proper directory name (3336)
3728 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3729 - fix ksocknal_fmb_callback() error messages (2918)
3731 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3734 - don't copy lvb into (possibly NULL) reply on error (2983)
3735 - don't deref dentry after dput, don't free lvb on error (2922)
3736 - use the kms to determine writeback rpc length (2947)
3737 - increment oti_logcookies when osc is inactive (2948)
3738 - update client's i_blocks count via lvb messages (2543)
3739 - handle intent open/close of special files properly (1557)
3740 - mount MDS with errors=remount-ro, like obdfilter (2009)
3741 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3742 - don't use cancelling-locks' kms values (2947)
3743 - use highest lock extent for kms, not last one (2925)
3744 - don't dereference ERR_PTR() dentry in error handling path (3107)
3745 - fix thread race in portals_debug_dumplog() (3122)
3746 - create lprocfs device entries at setup instead of at attach (1519)
3747 - common AST error handler, don't evict client on completion race (3145)
3748 - zero nameidata in detach_mnt in 2.6 (3118)
3749 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3750 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3751 - initialize RPC timeout timer earlier for 2.6 (3219)
3752 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3753 - print nal/nid for unknown nid (3258)
3754 - additional checks for oscc recovery before doing precreate (3284)
3755 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3756 - don't crash in mdc_close for bad permissions on open (3285)
3757 - zero i_rdev for non-device files (3147)
3758 - clear page->private before handing to FS, better assertion (3119)
3759 - tune the read pipeline (3236)
3760 - fix incorrect decref of invalidated dentry (2350)
3761 - provide read-ahead stats and refine rpc in flight stats (3328)
3762 - don't hold journal transaction open across create RPC (3313)
3763 - update atime on MDS at close time (3265)
3764 - close LDAP connection when recovering to avoid server load (3315)
3765 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3766 - don't leak open file on MDS after open resend (3325)
3767 - serialize filter_precreate and filter_destroy_precreated (3329)
3768 - loop device shouldn't call sync_dev() for nul device (3092)
3769 - clear page cache after eviction (2766)
3770 - resynchronize MDS->OST in background (2824)
3771 - refuse to mount the same filesystem twice on same mountpoint (3394)
3772 - allow llmount to create routes for mounting behind routers (3320)
3773 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3774 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3775 - fix rare problem with rename on htree directories (3417)
3777 - allow default OST striping configuration per directory (1414)
3778 - fix compilation for qswnal for 2.6 kernels (3125)
3779 - increase maximum number of MDS request buffers for large systems
3780 - change liblustreapi to be useful for external progs like lfsck (3098)
3781 - increase local configuration timeout for slow disks (3353)
3782 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3784 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3787 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3788 - don't overwrite extent policy data in reply if lock was blocked (2901)
3789 - drop filter export grants atomically with removal from device (2663)
3790 - del obd_self_export from work_list in class_disconnect_exports (2908)
3791 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3792 - swab reply message in mdc_close, other PPC fixes (2464)
3793 - fix destroying of named logs (2325)
3794 - overwrite old logs when running lconf --write_conf (2264)
3795 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3796 - fix race in target_handle_connect (2898)
3797 - mds_reint_create() should take same inode create lock (2926)
3798 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3799 - don't close files for self_export to avoid uninitialized obd (2936)
3800 - allow MDS with the same name as client node (2939)
3801 - hold dentry reference for closed log files for unlink (2325)
3802 - reserve space for all logs during transactions (2059)
3803 - don't evict page beyond end of stripe extent (2925)
3804 - don't oops on a deleted current working directory (2399)
3805 - handle hard links to targets without a parent properly (2517)
3806 - don't dereference NULL lock when racing during eviction (2867)
3807 - don't grow lock extents when lots of conflicting locks (2919)
3809 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3812 - account for cache space usage on clients to avoid data loss (974)
3813 - lfsck support in lustre kernel code (2349)
3814 - reduce journal credits needed for BRW writes (2370)
3815 - orphan handling to avoid losing space on client/server crashes
3816 - ptlrpcd can be blocked, stopping ALL progress (2477)
3817 - use lock value blocks to assist in proper KMS, faster stat (1021)
3818 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3819 - recovery for initial connections (2355)
3820 - fixes for mds_cleanup_orphans (1934)
3821 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3822 - block all file creations until orphan recovery completes (1901)
3823 - client remove rq_connection from request struct (2423)
3824 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3825 - recovery timer race (2670)
3826 - mdc_close recovey bug (2532)
3827 - ptlrpc cleanup bug (2710)
3828 - mds timeout on local locks (2588)
3829 - namespace lock held during RPCs (2431)
3830 - handle interrupted sync write properly (2503)
3831 - don't try to handle a message that hasn't been replied to (2699)
3832 - client assert failure during cleanup after abort recovery (2701)
3833 - leak mdc device after failed mount (2712)
3834 - ptlrpc_check_set allows timedout requests to complete (2714)
3835 - wait for inflight reqs when ptlrpcd finishes (2710)
3836 - make sure unregistered services are removed from the srv_list
3837 - reset bulk XID's when resending them (caught by 1138 test)
3838 - unregister_bulk after timeout
3839 - fix lconf error (2694)
3840 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3841 - readahead locks pages, leaves pending causing memory pressure (2673)
3842 - increase OST request buffers to 4096 on large machines (2729)
3843 - fix up permission of existing directories in simple_mkdir (2661)
3844 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3845 - don't assign transno to errored transactions (2742)
3846 - don't delete objects on OST if given a bogus objid from MDS (2751)
3847 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3848 - if rq_no_resend, then timeout request after recovery (2432)
3849 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3850 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3851 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3852 - MDS recovery shouldn't delete objects that it has given out (2730)
3853 - if enqueue arrives after completion, don't clobber LVB (2819)
3854 - don't unlock pages twice when trigger_group_io returns error (2814)
3855 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3856 - don't write pages to disk if there was an error (1450)
3857 - don't ping imports that have recovery disabled (2676)
3858 - take buffered bytes into account when balancing socknal conn (2817)
3859 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3860 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3861 - remove little-endian swabbing of llog records (1987)
3862 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3863 - retry reposting request buffers if they fail (1191)
3864 - grow extent at grant time to avoid granting a revoked lock (2809)
3865 - lock revoke doesn't evict page if covered by a second lock (2765)
3866 - disable VM readahead to avoid reading outside lock extents (2805)
3868 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3869 - updated kernel patches for hp-2.4.20 kernel (2681)
3871 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3874 - fix truncated write corruption (2366)
3875 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3877 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3878 - ensure len(uuid) < 37 in lmc (1171)
3879 - fix ia64 OOPS in llog_test (2255)
3880 - zero end of page at obdfilter for partial page writes (2648)
3881 - don't leave stale dentries around after renames (bug 2428)
3882 - fix timeouts when evicting a client with a single lock held (2642)
3883 - set deadline for the initial HELLO message to drain (2634)
3884 - print out dotted-quad IP addresses in the socknal (2302)
3886 - additional debugging for MDS client eviction problem (2443)
3887 - fix mkfsoptions support for osts (2603, 2604)
3889 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3892 - add series for the vanilla 2.6.0 kernel
3893 - add series for the vanilla 2.4.24 kernel
3894 - add series for a cray x86/64 UL kernel drop
3895 - fix xattr patches for the vanilla 2.4.19 series
3897 - generate true UUIDs in lmc (1171)
3898 - have portals stack dumping break in UML (2466)
3899 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3900 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3901 - wake callback waiting threads on client eviction (2460)
3902 - Add --ptldebug and --subsystem to lmc (1719)
3903 - update assertion to allow safe interrupt allocation
3904 - set rq_no_resend for cancel requests (2432)
3905 - recalculate ptlrpcd timeout after resend (2494)
3906 - call vfs_rmdir when removing pending directories (2368)
3907 - fix renaming a file to itself (2429)
3908 - lmc creates a default one-stripe lov (2454)
3909 - expand procfs space to handle large clusters (2326)
3910 - increase UML stack to avoid overflow
3911 - update lconf's list of debug and subsystem masks
3912 - fix lfs find --obd (2510)
3913 - /proc tunable for disabling filter read caching (2591)
3914 - stop rpm packages from altering slapd.conf (2301)
3915 - disable nagle in the socknal under 0conf (2578)
3916 - choose mds inode size based on stripe count (2572)
3917 - fix kernel-source rpm problems (2516)
3919 - add --disable-doc to avoid pdf generation (2421)
3920 - update documentation, tests, type-os, comments
3921 - avoid format warnings on ia64
3922 - remove the TOE NAL
3923 - tiny code cleanups by removing unused fields
3925 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3928 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3929 - remove the most blatant lies from BUILDING file (2371)
3930 - change default debug level to reasonable production setting
3931 - reduce client side cache size to reduce cache flush time
3932 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3933 - make TCP zerocopy and pinger support enabled by default (2476)
3934 - sync writes completed after process exits caused crashes (2319)
3935 - maintain correct mount count on the MDS (2356)
3936 - backout 1557, because 2316 wasn't really fixed
3937 - better file I/O statistics gathering in /proc
3938 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3939 - another kernel patch to fix zero-copy TCP function export
3940 - don't take duplicate lock when processing re-sent getattr (2420)
3941 - lctl uses obd_self_export instead of creating new conn (2353)
3942 - MDS/OST recovery case which requires object creation asserted (2425)
3943 - move lfs from /usr/sbin to /usr/bin in packages
3944 - fix race between mds_client_add and mds_client_free (2417)
3945 - use kmalloc instead of slabs in portals (2430)
3946 - don't create duplicate records when a failover MDS is present (2442)
3947 - remove unnecessary mount age check (2332)
3948 - don't remove directory inodes from locks prematurely (2451)
3949 - don't break if MDS service name is the same as hostname (2103)
3950 - fix races in client write RPC generation when cache full (2482)
3952 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3955 - remove now-unused request->rq_obd (278)
3956 - if an allocation fails, print out how much memory we've used (1933)
3957 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3958 - add the "configurable stack size" patch to most series files (1256)
3959 - ability to write large log records, for 100+ OST configs (2306)
3960 - fix NULL deref when filter_prep fails (2314)
3961 - fix operator precedence error in filter_sync
3962 - dynamic allocation of socknal TX descriptors (2315)
3963 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3964 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3965 - ensure that CWARN messages in Portals always get to the syslog
3966 - __init/__exit are not for prototype decls (ldlm_init/exit)
3967 - x86-64 compile warning fixes
3968 - fix gateway LMC keyword conflict (2318)
3969 - fix MDS lock inversions in getattr/reint paths (1844)
3970 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3971 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3972 - DLM race condition prevented some lock evictions (2328)
3973 - ENOMEM detection and retry on socknal sends (2230)
3974 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3975 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3976 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3977 - better error messages when a client is rejected during recovery (1505)
3978 - avoid cancelling locks which were never granted, after failure (2330)
3979 - fix i_sem/journal inversion in mds_client_add (2333)
3980 - fix truncate/getattr lock cycle deadlock (2334)
3981 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3982 - fix two rare exit paths which could leak an l_lock() ref (2321)
3983 - fencepost error in MDS/OST orphan recovery (2226)
3984 - make log record alignment 8 bytes (1988)
3985 - lstripe now fails when requested offset > ost_count (2237)
3986 - ensure that all kernel series have a complete list.h (1607)
3987 - fix crashes in special-file operations (2316)
3988 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3990 - allow configurable automake binary, for testing new versions
3991 - small update to the lfs documentation
3993 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3995 * fix negative export reference count in fsfilt_sync (2312)
3997 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3998 * release candidate 0.9.1
4000 - orphans are moved into the PENDING directory for possible recovery
4001 - replayed opens now open by fid for orphan/rename safety (1042)
4002 - last close of an orphan inode generates a transno (683)
4003 - chdir() and mount() now pin the directory entry (1020)
4004 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4005 - discard very old requests without processing them (1502)
4006 - remove some common, well-understood CERRORs (1505)
4007 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4008 - clear "grant" flags in OST replies until OST grant code lands (1644)
4009 - fix read performance by not clobbering i_blksize on client (1598)
4010 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4011 - make LRU size a /proc tunable, clears locks when reduced (707)
4012 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4013 - server locks take references on exports now (1558)
4014 - build fixes for 2.4.20-rh trees (1663)
4015 - return an error from lov_create if all OSCs are inactive (1751)
4016 - fix import levels when a reconnect happens without a timeout (1597)
4017 - exit early from mds_open if we get a lookup error (1749)
4018 - partial page read at EOF wouldn't wait for disk before sending (1642)
4019 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4020 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4021 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4022 - write/truncate lock inversion (1639)
4023 - Don't auto-load obdclass, portals modules during cleanup (1495)
4024 - fix timestamps from jumping to "now" (1763)
4025 - extra journal assertions (1648)
4026 - add an extra multiunlink test (1771)
4027 - fix read_record/write_record API (1776)
4028 - fix leak of offset_extent, possible incorrect i_size later (1772)
4029 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4030 - Debugging for the kqswnal_get_idle_tx problems (1820)
4031 - Allow recovery to be attempted multiple times (1536)
4032 - Write out MDS last_rcvd file after it is first created (1600)
4033 - Fix tx_descriptor leak in failed transmit situations (1827)
4034 - ext3 journaling fixes for assertion failure after IO error (1871)
4035 - class_export_put() on freed export after completion AST error (1896)
4036 - Fix revalidate looping in VFS (1322)
4037 - Don't access a freed export during MDS_REINT timeout (1521)
4038 - Add open-unlink recovery support on the MDS (1673,1764)
4039 - Return an error if no MDS data was read from last_rcvd (1946)
4040 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4041 - Don't setup a disk device that doesn't match exported UUID (317)
4042 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4043 - avoid committing NULL handle in force close
4044 - local.sh is now a one-stripe LOV configuration
4045 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4046 - fix ext3 htree duplicate directory entry corruption (1516)
4047 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4048 - update to new LOV EA format (2097)
4049 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4051 2003-06-15 Phil Schwan <phil@clusterfs.com>
4054 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4055 - per-import/export recovery handling (958, 931, 959)
4056 - multiple last-rcvd slots, for serving multiple FSes (949)
4057 - connections are again shared between multiple imp/exports (963, 964)
4058 - "umount -f" would hang if any requests needed to be sent (393, 978)
4059 - avoid pinning large req buffer by copying for queued messages (989)
4060 - add "uuid" to "lctl device" command to help upcalls (991)
4061 - "open" RPCs with transnos would confuse recovery counters (1037)
4062 - do proper endian conversion of all wire messages (288, 340, 891)
4063 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4064 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4065 - fix ptlrpc_connection leak in target_handle_connect (1174)
4066 - fix import refcounting bug in OST and MDS cleanup (1134)
4067 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4068 - fix very unlikely obd_types race condition (501)
4069 - remove osc_open hack for echo_client (1187)
4070 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4071 - a failure in read_inode2 leads to deadlock (1139)
4072 - cancel ack-locks as soon as transaction is committed (1072)
4073 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4074 - make sure to commitrw after any preprw to avoid deadlock (1162)
4075 - failing to execute a file in a lustre FS would lock inode (1203)
4076 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4077 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4078 - obd_uuid2tgt would walk off the end of the list (1255)
4079 - on IA64 the osc would give portals incorrect bulk size (1258)
4080 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4081 - fix lock inversion caused by new llite matching code (1282)
4082 - limit the number of dirty pages on a client to 10MB (1286)
4083 - timed out locks were not being corrected cancelled (1289)
4084 - fix O_DIRECT above 4GB on IA-32 (1292)
4085 * major user-visible changes
4086 - fail out/fail over policy now controlled by the upcall (993)
4088 - add OBD_PING to check server availability and failure (954)
4089 - lustre messages are now sent in sending host order (288, 340, 891)
4090 - add eadatalen to MDS getattr reply (340)
4091 - OST read replies may contain second buffer, with per-page status (593)
4093 2003-03-11 Phil Schwan <phil@clusterfs.com>
4096 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4097 - failed open()s could cause deadlock; fixed (867, 869)
4098 - stop cancelling OST locks when files are closed (481)
4099 - overlapping XID spaces caused network corruption (851, 853)
4100 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4101 - setattr_raw added, to do single-RPC, server-side setattrs
4102 - lmc/lconf syntax change for OST UUIDs
4103 - fix crashy race condition between ptlrpc_free_req and osc_close
4104 - don't use request in mdc_enqueue if we hit a timeout (889)
4105 - don't set the inode i_size for regular files from the MDS (896)
4106 - handle out of order completion AST (842)
4107 - don't LBUG if a lock request times out after receiving AST (913)
4108 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4109 - if a bad lock AST arrives, send an error instead of dropping entirely
4110 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4111 - fix leak in bulk IO when only partially completed (899, 900, 926)
4112 - fix O_DIRECT for ia64 (55)
4113 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4114 - C-z after timeout could hang a process forever; fixed (977)
4116 - client-side I/O cache (678, 924, 929, 941, 970)
4118 - READPAGE and SETATTRs which don't take server-side locks get
4121 2003-02-11 Phil Schwan <phil@clusterfs.com>
4124 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4125 - Fix fsfilt_extN_readpage() to read a full page of directory
4126 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4127 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4128 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4129 - Most intent operations were removed, in favour of a new RPC mode
4130 that does a single RPC to the server and bypasses most of the VFS
4131 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4132 - Aggressively cancel local locks on DLM servers
4133 - mds_reint_unlink sends EA to the client if it's the last nlink.
4134 client uses that EA to unlink OST objects.
4135 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4136 - recursive symlinks were fixed (439)
4137 - fixed NULL deref in DEBUG_REQ
4138 - filter_update_lastobjid no longer calls sync, which annoyed extN
4139 - fixed multi-client small-writes to a single file problem (445)
4140 - fixed mtime updates during file writes (607)
4141 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4142 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4143 - fixed Linux OST size reporting problem (444, 656)
4144 - OST now updates object mtime with writes or setattr (607, 619)
4145 - client verifies file size before zeroing page past EOF (445)
4146 - OST now writes last allocated objid to disk with allocation (108)
4147 - LOV on echo now works (409)
4149 - mds_reint_unlink sends a new buffer, with the EA included. this
4150 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4151 set if a regular file was being unlinked, and it was the last link
4152 - use PtlGet from the target for bulk writes (315)
4153 - OST now updates object mtime with writes or setattr (607, 619)
4154 - LDLM now has a grant-time callback to revalidate locked items, if
4156 - Many MDS operations were reorganized to combat race conditions
4158 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4159 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4161 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4164 - Fully reactivate OST imports after reconnection (512, others)
4165 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4166 - More graceful error handling for truncating on dead OST (515)
4167 - Don't error out unless we're actually accessing dead stripes (474)
4168 - Fix garbage sizes when stripes are missing (410)
4169 - LRU counters were broken, causing constant lock purge (433, 432)
4170 - garbage on read from stripes with failed OSTs (441)
4171 - mark OSCs as active before reconnecting during recovery (438)
4172 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4173 - lfind did not preserve OST order in output (443)
4174 - symlinks cause hung clients, incorrect data (439)
4175 - stop dereferencing request after dropping refcount (457)
4176 - don't LASSERT(spin_is_locked) on non-SMP (455)
4177 - fixes for many rename() bugs
4178 - fstat didn't correctly synchronize attributes (399)
4179 - server must handle lock cancellation during blocking AST prep (487)
4180 - bulk descriptors were free()d too soon (511)
4181 - fix paths in lconf, which would load incorrect modules (451, 507)
4182 - fix confusing lconf 'host not found' error message (386)
4183 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4184 - fix race condition in mdc_blocking_ast() for inode access (526)
4185 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4186 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4187 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4188 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4189 - fix lstripe in several ways (kernel side) (527)
4190 - fix request leak in ldlm_cli_enqueue (262)
4191 - incorrect OSC was marked inactive after OST failure
4192 - call mds_fs_cleanup before unmounting filesystem (524)
4193 - fix races between taking ns_lock and ldlm_lock_change_resource
4194 - fix races updating LOV export open file list
4195 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4196 - fix recovery NULL deref in ldlm_cli_cancel_unused
4197 - fix some DLM races by using new hash table for lock handles (419)
4198 - permit the client to specify desired inodes, at replay
4199 - duplicate requests when we queue them for replay reintegration
4200 - fix last_rcvd offset calculation
4201 - sync after each recovered transaction, so we always make progress
4202 - never, not always, ERESTART requests without transnos
4203 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4205 - skip replay if the MDS says that the client is already connected
4206 - don't check for a recovery-enabled export to match lctl's UUID
4207 - don't INC_USE_COUNT for phantom exports
4208 - don't crash when cleaning up phantom exports (567)
4209 - don't double-finish or set replay data for errored mdc_open requests
4210 - abort requests when they time out, so we don't get old replies
4211 - send/receive replies for AST messages again
4212 - if the client says that it doesn't have the lock, cancel it on the
4214 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4215 mark it as destroyed, it will all work out when decref is called
4216 - fix module use counts (22, 581)
4218 - ASTs now expect a reply (server cancels lock on error reply)
4220 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4223 - fix many simultaneous client startup (392)
4224 - fix dentry->d_it clobbering
4225 - credentials weren't being shipped for readdir/getattr operations
4226 - remove invalid assertions triggered during some concurrent MD
4228 - proper Lustre versions added (336, 389)
4229 - fix memory leak for create error case (398)
4230 - fix LOV locking bug that would get cli/srv out of sync
4231 - fix echo client over LOV (409)
4232 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4233 - fix double-O_EXCL intent crash (424)
4234 - avoid sending multiple lock CANCELs (352)
4236 - MDS can do multi-client recovery (modulo bugs in new code)
4238 - many updates, edits, cleanups
4240 2002-11-18 Phil Schwan <phil@clusterfs.com>
4243 - fix null d_it dereference (346)
4244 - fix full OST/dbench hang (333)
4245 - fix permission problem with file removal (286)
4246 - fix removal of OSCs from LOV when they fail
4247 - fix NULL deref during bulk timeout (214)
4248 - fix problems related to multiple filesystems on one MDS (241)
4249 - fixed serious subtle metadata locking bugs
4250 - free locks on clients when inodes are removed due to memory
4252 - fix inode pointer in lock data (285)
4253 - partial support for multiple MDS on a single host (241)
4254 - data locks weren't cancelled at clear_inode time (290, 311)
4255 - intent locks could lead to unbounded lock growth (205)
4256 - added a maximum lock count, an LRU list, and a flusher
4257 - fix multiple rename (365)
4258 - properly abstracted the echo client
4259 - OSC locked 1 byte too many; fixed
4260 - rewrote brw callback code:
4261 - fixed recovery bugs related to LOVs (306)
4262 - fixed too-many-pages-in-one-write crash (191)
4263 - fixed (again) crash in sync_io_timeout (214)
4264 - probably fixed callback-related race (385)
4266 - Add capability to MDS protocol
4267 - LDLM cancellations and callbacks on different portals
4269 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4272 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4273 - defer open object destruction to close time (601981, 138)
4274 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4275 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4276 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4277 - don't call obd_setattr in ll_file_release for destroyed objects
4278 * protocol change to lustre_msg: move |version| and add |flags|
4279 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4280 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4281 * added replay of create, unlink, link and rename operations during
4282 MDS failover; recovery should be much more robust now
4283 * remove failed OSCs from LOVs (only lov_create uses this so far)
4284 * the lustre-HOWTO was brought (more) up to date (582544)
4286 2002-10-23 Phil Schwan <phil@clusterfs.com>
4289 - in-use dentries weren't being reused properly (617851)
4290 - prevent multiple LDLM setup (599178)
4291 - fix LOV size calculations for truncate (617853)
4292 - fix client handling of MDS intent errors (POSIX)
4293 - fix permission bug in lovstripe.c test (624321)
4294 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4295 - truncate past end of file could corrupt data
4296 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4297 - a race in recovery could return ETIMEDOUT to apps (623947)
4298 - building outside the source directory was fixed
4299 * the lustre-HOWTO was brought (more) up to date (582544)
4300 * major progress was made on recovery functionality
4302 2002-10-10 Phil Schwan <phil@clusterfs.com>
4305 - recovery deadlock fix
4306 - rm -rf causes LBUG fix (617817)
4307 - file open by multiple tasks fix (618962)
4308 - directory permissions bugs (602707 and 620007)
4309 - journal_stop fixed with locking (611313)
4310 - O_APPEND failures resolved (618273, perhaps 614459)
4311 - lconf PATH fix (619770)
4312 - IA64 build fix (621450)
4313 - RPC buffer sizes scale with amount of memory
4315 2002-10-01 Phil Schwan <phil@clusterfs.com>
4318 - locks would be cancelled without throwing away data pages,
4319 resulting in inconsistent data (605627)
4320 - inode attributes were not always being refreshed (605627, 612449)
4321 - lconf now continues to cleanup after lctl reports an error
4322 - MDS now enforces user permissions (602707)
4323 - lprocfs cleanup fixed, but not yet enabled (614157)
4324 - fixed infinite server hang, should a client not respond to an AST
4325 - avoid going into recovery if user calls readlink() with a buffer
4326 that's too small (613941)
4327 - AST RPCs no longer require replies (614867) -- this may be changed
4328 - don't crash server if client sends an IOV that's too big (611336)
4329 - fixed lock conversion deadlock (611892)
4330 - fixed the following of symlinks (614622)
4331 * recovery: the server can remove locks from a client that dies, other
4332 clients can make progress
4333 * more extN patch fixes
4334 * compile-time configurable ptlrpc buffer allocations
4336 - collaborative read cache document
4337 - Lustre Lite Performance CDR document-in-progress
4339 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4342 - fix typo in patch-2.4.18
4344 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4347 - clear ptlrpc request each time in handle_incoming_request()
4348 - unlink of files now destroys the object on the OST
4350 2002-09-19 Peter Braam <braam@clusterfs.com>
4352 * add hard link support
4353 * change obdfile creation method
4354 * kernel patch changed
4356 2002-09-19 Peter Braam <braam@clusterfs.com>
4359 - stack overflow bug in extN fixed
4361 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4363 * documentation updates
4364 - add man pages for config tools
4365 - update tests/README to describe testing with new config tools
4366 - finish metadata API descriptions
4367 * bug fixes and cleanups
4368 - statfs workaround for 16TB limit
4369 - LOV stripe allocation improved, can stripe on subset of OSTs
4370 - LOV file size/IO offset was wrong for files > 4GB in size
4371 - object EA data was being dropped, caused files to be unreadable
4372 - memory overflow with non-LOV OST caused memory corruption
4373 - fixed regression tests to work with new config tools, obdfilter
4374 - fixed bug when directory size became larger than 1 block
4375 - fixed bug (for single client case) when PWD was deleted
4376 - invalidate local directory pages when doing intent-based ops
4377 - avoid LDLM oops when lock callback contained bad data
4379 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4381 * documentation updates
4382 * bug fixes and cleanups
4383 - configuration tools
4386 - 64-bit compile warnings
4387 - 64-bit internal statfs data
4389 * test_brw on persistent OST devices
4391 * lprocfs (disabled)
4393 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4395 * documentation updates
4396 * bug fixes and cleanups
4397 * configuration tools
4399 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4402 * many small fixes to 0_5_4
4403 * io/network handling
4404 * thinkos in MDS operations
4406 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4409 * crucial basic fixes to 0.5.3
4410 * IOR, Iozone work over Elan
4413 2002-08-07 Phil Schwan <phil@clusterfs.com>
4414 * version 0_5_3, our first alpha
4415 * we use the new Portals iovs
4416 * documentation updates
4417 * bug fixes and cleanups
4418 * small changes in the DLM wire protocol
4420 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4421 * version 0_5_1 with some initial stability,
4422 * locking on MD and file I/O.
4423 * documentation updates
4424 * several bug fixes since 0.5.0
4425 * small changes in wire protocol
4427 2002-07-18 Phil Schwan <phil@clusterfs.com>
4429 * delivered as Lustre Light Alpha
4430 * fixed a crash after handling invalid MDS requests
4431 * fixed directory pages for architectures with non-4k pages sizes
4433 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4434 * release version v0_4_4
4435 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4437 * Fixes a number of interruption problems with OST operations.
4438 * Update documentation for portals header changes
4439 * Move all wire protocol structs/defines to lustre_idl.h
4440 * Fixes symlink length bug.
4441 * Add tcpdump to repository.
4443 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4444 * release version v0_4_3
4445 * Fixes statfs for inodes on extN.
4446 * Fixes bug in runtests which would delete /etc/hosts.
4447 * Use 64-bit object IDs wherever possible (not into VFS though)
4448 Remove ost_get_info, which is unused by lustre, and out of date.
4450 2002-07-03 Peter Braam <braam@clusterfs.com>
4451 * release version v0_4_2 Fixes a lookup error (type not passed)
4452 * move forward to head of Portals
4453 * move forward to latest Lustre kernel
4455 2002-06-25 Peter Braam <braam@clusterfs.com>
4456 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob