1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.5 (SLES 10),
6 2.6.18-53.1.13.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).
16 Frequency : rare, depends on device drivers and load
18 Description: MDS or OSS nodes crash due to stack overflow
19 Details : Code changes in 1.8.0 increased the stack usage of some functions.
20 In some cases, in conjunction with device drivers that use a lot
21 of stack the MDS (or possibly OSS) service threads could overflow
22 the stack. One change which was identified to consume additional
23 stack has been reworked to avoid the extra stack usage.
26 Frequency : occasional
28 Description: Do not fail import if osc_interpret_create gets -EAGAIN
29 Details : If osc_interpret_create got -EAGAIN it immediately exits and
30 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
31 oscc_has_objects and see OSC has no objests and call
32 oscc_internal_create to resend create request.
34 Severity : enhancement
36 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
38 Severity : enhancement
40 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
45 Description: Fix replay if there is an un-replied request and open
46 Details : In some cases, older replay request will revert the
47 mcd->mcd_last_xid on MDS which is used to record the client's
50 Severity : enhancement
52 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
54 Severity : enhancement
56 Description: Add rhel5 support to HEAD.
58 Serverity : enhancement
60 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
65 Description: Don't allow skipping OSTs if index has been specified.
66 Details : Don't allow skipping OSTs if index has been specified, make locking
67 in internal create lots better.
71 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
72 Details : access to bitfield in structure is always rounded to long
73 and this produce problem with not atomic change any bit.
77 Description: Lustre make rpms failed.
78 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
79 builds Lustre rpms from tarball.
82 Frequency : rare on shutdown ost
84 Description: If llog cancel was not send before clean_exports phase, this can
85 produce deadlock in llog code.
86 Details : If llog thread has last reference to obd and call class_import_put
87 this produce deadlock because llog_cleanup_commit_master wait when
88 last llog_commit_thread exited, but this never success because was
89 called from llog_commit_thread.
93 Description: allow userland application know is lost one of stripes.
94 Details : fill lvb_blocks with error code on ost and return it to
95 application if error flag found.
99 Description: NULL lov_tgts causing MDS oops
100 Details : more safe checks for NULL lov_tgts for avoid oops.
102 Severity : enhancement
104 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
108 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
109 Details : If an OST is down the MDS will hang indefinitely in
110 obd_statfs_rqset() waiting for the statfs data. While for
111 MDS QOS usage of statfs, it should not stuck in waiting.
113 Severity : enhancement
115 Description: remote_acl support
116 Details : Support ACL-based permission check for remote user.
117 Support setfacl/getfacl for remote user with the utils
118 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
119 the system "{s,g}etfacl" utils.
121 Severity : enhancement
123 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
125 Severity : enhancement
127 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
131 Description: make mgs_setparam() handle fsname containing dash
132 Details : fsname containing a dash does not work with lctl conf_param
134 Severity : enhancement
136 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
138 Severity : enhancement
140 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
144 Frequency : rare, at shutdown
145 Description: access already free / zero obd_namespace.
146 Details : if client_disconnect_export was called without force flag set,
147 and exist connect request in flight, this can produce access to
148 NULL pointer (or already free pointer) when connect_interpret
149 store ocd flags in obd_namespace.
153 Frequency : only at startup
154 Description: not alloc memory with spinlock held.
155 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
156 if any spinlock held.
158 Severity : enhancement
160 Description: make lustre randomly fail allocating memory
161 Details : Make lustre randomly failed allocating memory for testing purpose.
163 Severity : enhancement
165 Description: lost problems with lov objid file
166 Details : Fixes some scability and access to not inited memory problems
167 in work with lov objdid file.
172 Description: lfs find does not continue on file error
173 Details : Continue other files processing when a file/dir is absent.
177 Description: Inconsistent usage of lustre_pack_reply()
178 Details : Standardize the usage of lustre_pack_reply() such that it
179 always generate a CERROR on failure.
182 Frequency : occasional
184 Description: MDS hang or stay in waiting lock
185 Details : If client receive lock with CBPENDING flag ldlm need send lock
186 cancel as separate rpc, to avoid situation when cancel request
187 can't processed due all i/o threads stay in wait lock.
191 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
195 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
199 Description: Update to SLES9 kernel-2.6.5-7.287.3.
203 Description: Update to RHEL4 latest kernel.
205 Severity : enhancement
207 Description: Build SLES10 patchless client fails
208 Details : The configure was broken by run ./configure with
209 --with-linux-obj=.... argument for patchless client. When the
210 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
211 can't search header adequately. Use absolute path such as
212 -I($LINUX)/include instead.
216 Description: interrupt oig_wait produce painc on resend.
217 Details : brw_redo_request can be used for resend requests from ptlrpcd and
218 private set, and this produce situation when rq_ptlrpcd_data not
219 copyed to new allocated request and triggered LBUG on assert
220 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
221 ptlrpcd set if request is changed and can be safety changed to use
226 Description: LASSERT_{REQ,REP}SWAB macros are buggy
227 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
228 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
229 these macros and replace them with their difinitions instead.
233 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
234 Details : Update which_patch & target file for SLES10 latest kernel.
237 Description: Remove client patches from SLES 10 kernel.
238 Details : This causes SLES 10 clients to behave as patchless clients
239 even on a Lustre-patched (server) kernel.
241 Severity : enhancement
243 Description: self-adjustable client's lru lists
244 Details : use adaptive algorithm for managing client cached locks lru
245 lists according to current server load, other client's work
246 pattern, memory activities, etc. Both, server and client
247 side namespaces provide number of proc tunables for controlling
250 Severity : enhancement
252 Description: light-weight GSS support
253 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
254 and reduce performance overhead.
256 Severity : enhancement
258 Description: Linux keyring support
259 Details : Support using service of Linux keyring for Lustre GSS internal
260 context refresh/cache mechanism.
264 Description: Fix errors in lfs documentation
265 Details : Fixes man pages
269 Description: don't use GFP_* in generic Lustre code.
270 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
274 Description: obdclass is limited by single OBD_ALLOC(idarray)
275 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
279 Description: warnings with build patchless client with vanila 2.6.19 and up
280 Details : change old ctl_table style and replace ctl_table/ctl_table_header
281 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
285 Frequency : Only for SLES
286 Description: sanity_quota fail test_1
287 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
288 file, which makes TSTID[2] inunique.
292 Frequency : Only for SLES9
293 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
294 Details : Update target/ChangeLog/which_patch .
298 Frequency : Only for test_10 in sanity-quota.sh
299 Description: a bug in quota test code
300 Details : checking a test flag in a wrong place
304 Frequency : Only for test_13 in sanity-quota.sh
305 Description: a bug in quota test script
306 Details : shouldn't check overfull allocated quota
310 Description: jbd statistics
311 Details : Port older jbd statistics patch for sles10
313 Severity : enhancement
315 Description: Add printing inode info into message about error in writepage.
320 Description: reply_lock_interpret crash due to race with it and lock cancel.
321 Details : Do not replay locks that are being cancelled. Do not reference
322 locks by their address during replay, just by their handle.
326 Frequency : When flocks are used.
327 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
328 Details : Flock locks might destroy just granted lock if it could be merged
329 with another existing flock, this is done in completion handler,
330 so teach ldlm_cli_enquque_fini that this is a valid case for
336 Description: Oops in read and write path when failing to allocate lock.
337 Details : Check if lock allocation failed and return error back.
341 Description: lstripe command fails for valid OST index
342 Details : The stripe offset is compared to lov->desc.ld_tgt_count
343 instead of lov->desc.ld_active_tgt_count.
347 Description: sanity.sh failed test 103
348 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
353 Description: df doesn't work properly if diskfs blocksize != 4K
354 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
358 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
359 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
363 Description: obd_health_check_timeout too short
364 Details : set obd_health_check_timeout as 1.5x of obd_timeout
368 Description: llapi_file_create() does not allow some changes
369 Details : add llapi_file_open() that allows specifying the mode and
370 open flags, and also returns an open file handle.
374 Description: merge and cleanup kernel patches.
375 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
379 Description: Add journal checksum support.(Kernel part)
380 Details : The journal checksum feature adds two new flags i.e
381 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
382 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
383 indicates that the commit block contains the checksum for
384 the blocks described by the descriptor blocks. Now commit
385 record can be sent to disk without waiting for descriptor
386 blocks to be written to disk. This behavior is controlled
387 using JBD2_FEATURE_ASYNC_COMMIT flag.
391 Description: OSS needs mutliple precreate threads
392 Details : Add ability to start more than one create thread per OSS.
396 Description: Sanity.sh test_65k failed: setstripe should have succeeded
397 Details : execute lfs setstripe on client
401 Description: mds_obd_create error creating tmp object
402 Details : When the user sets quota on root, llog will be affected and can't
403 create files and write files.
406 Frequency : Always on ia64 patchless client, and possibly others.
408 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
409 Details : This allows the patchless client to be loaded on architectures
414 Description: RedHat Update kernel for RHEL5
415 Details : Add the kernel config file for RHEL5.
419 Description: RedHat Update kernel for RHEL5
420 Details : Modify the kernel config file more closer RHEL5.
424 Description: Build failure against Centos5 (RHEL5)
425 Details : Define PAGE_SIZE when it isn't present.
429 Description: test_7 on sanity-quota.sh can't be used on separate machines
430 Details : add facet to handle it
434 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
435 Details : with NFS, the anon dentry's parent was set to itself in
436 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
437 corresponding dentry other than use rec->ur_name.
439 Severity : enhancement
441 Description: lfs setstripe enhancement
442 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
446 Description: enable data checksumming by default
447 Details : enable checksum by default, allow --disable-checksum
448 configure option and "-o nochecksum" mount option. Checksums
449 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
453 Description: lustre support for RHEL5
454 Details : Add support for RHEL5.
458 Description: Client eviction due to failover config
459 Details : after a connection loss, the lustre client should attempt to
460 reconnect to the last active server first before trying the
461 other potential connections.
465 Description: when mds and osts use different quota unit(32bit and 64bit),
466 quota will be released repeatly.
467 Details : void sending multiple quota reqs to mds, which will keep the
468 status between the reqs.
472 Description: osts not allocated evenly to files
473 Details : change the condition to increase offset_idx
476 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
478 Description: Data corruption for OSTs that are formatted larger than 2TB
480 Details : When generating the bio request for lustre file writes the
481 sector number would overflow a temporary variable before being
482 used for the IO. The data reads correctly from Lustre (which
483 will overflow in a similar manner) but other file data or
484 filesystem metadata may be corrupted in some cases.
488 Description: Tune the kernel for good SCSI performance.
489 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
490 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
495 Description: rewrite ext2-derived code in obdclass/uuid.c
496 Details : rewrite inherited code (uuid parsing code from ext2 utils)
497 from scratch preserving functionality.
501 Description: "lfs find -obd UUID" prints directories
502 Details : "lfs find -obd UUID" will return all directory names instead
503 of just file names. It is incorrect because the directories
504 do not reside on the OSTs.
508 Description: use i_size_read and i_size_write in 2.6 port
509 Details : replace inode->i_size access with i_size_read/write()
512 Frequency : when removing large files
514 Description: scheduling issue during removal of large Lustre files
515 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
516 It causes scheduling issues when removing large files (17TB in the
520 Frequency : only with liblustre clients on XT3
522 Description: evictions taking too long
523 Details : allow llrd to evict clients directly on OSTs
526 Frequency : only on ppc
528 Description: /proc/fs/lustre/devices broken on ppc
529 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
530 We should make sure the type of 'index' is loff_t and then cast
531 explicitly as needed below. Do not assign an explicitly cast
535 Frequency : only for rhel5
537 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
538 Details : Modify the target file & which_kernel.
540 Severity : enhancement
542 Description: omit set fsid for export NFS
543 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
544 as FS_REQUIRES_DEV for avoid problems with generate fsid.
548 Frequency : Always, for kernels after 2.6.16
549 Description: Fix warning idr_remove called for id=.. which is not allocated.
550 Details : Last kernels save old s_dev before kill super and not allow
551 to restore from callback - restore it before call kill_anon_super.
555 Description: buffer overruns could theoretically occur
556 Details : llapi_semantic_traverse() modifies the "path" argument by
557 appending values to the end of the origin string, and a
558 overrun may occur. Adding buffer overrun check in liblustreapi.
562 Description: Fix error on 'ls .' at the top of the Lustre mount.
563 Details : Don't revalidate dentry if it is a root dentry.
567 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
568 Details : Modify vm-tunables-rhel4.patch.
572 Description: Kernel config for 2.6.18-vanilla.
573 Details : Modify targets/2.6-vanilla.target.in.
574 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
575 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
576 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
577 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
581 Description: improve handling recoverable errors
582 Details : if request processig with error which can be recoverable on server
583 request should be resend, otherwise page released from cache and
588 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
589 Details : Modify target file & which_patch.
590 A flaw was found in the IA32 system call emulation provided
591 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
592 value could be stored in the %RAX register, which could trigger an
593 out-of-bounds system call table access. An untrusted local user
594 could exploit this flaw to run code in the kernel
595 (ie a root privilege escalation). (CVE-2007-4573).
599 Description: O_DIRECT bypasses client statistics.
600 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
601 read_ahead_stats were updated. copied the stats section from
602 osc_send_oap_rpc() into async_internal().
606 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
610 Description: change order of libsysio includes
611 Details : '#include sysio.h' should always come before '#include xtio.h'
615 Description: handle "serious error: objid * already exists" more gracefully
616 Details : If LAST_ID value on disk is smaller than the objects existing in
617 the O/0/d* directories, it indicates disk corruption and causes an
618 LBUG(). If the object is 0-length, then we should use the existing
619 object. This will help to avoid a full fsck in most cases.
621 Severity : enhancement
623 Description: adapt the lustre_config script to support the upgrade case
624 Details : Add "-u" option for lustre_config script to support upgrading 1.4
625 server targets to 1.6 in parallel.
629 Description: To avoid grant space > avaible space when the disk is almost
630 full. Without this patch you might see the error "grant XXXX >
631 available" or some LBUG about grant, when the disk is almost
633 Details : In filter_check_grant, for non_grant cache write, we should
634 check the left space by if (*left > ungranted + bytes), instead
635 of (*left > ungranted), because only we are sure the left space
636 is enough for another "bytes", then the ungrant space should be
637 increase. In client, we should update cl_avail_grant only there
638 is OBD_MD_FLGRANT in the reply.
642 Description: Update RHEL 4 kernel to fix local root privilege escalation.
643 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
644 described in CVE-2007-4573. This problem could allow untrusted
645 local users to gain root access.
648 Frequency : when using O_DIRECT and quotas
650 Description: Incorrect file ownership on O_DIRECT output files
651 Details : block usage reported by 'lfs quota' does not take into account
652 files that have been written with O_DIRECT.
657 Description: touch file failed when fs is not full
658 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
659 otherwise we can get ENOSP while fs is not full.
663 Description: parallel lock callbacks
664 Details : Instead of sending blocking and completion callbacks as separated
665 requests, adding them to a set and sending in parallel.
668 Frequency : only for Cray XT3
669 Bugzilla : 12829/13455
670 Description: Changing primary group doesn't change the group lustre assigns to
672 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
673 overridden with the primary group provided by the group upcall,
674 whereas the supplied fsgid can be trusted if it is in the list of
675 supplementary groups returned by the group upcall.
677 Severity : enhancement
679 Description: Allow masking D_WARNING, D_ERROR messages from console
680 Details : Console messages can now be disabled via lnet.printk.
684 Description: User code with malformed file open parameter crashes client node
685 Details : Before packing join_file req, all the related reference should be
686 checked carefully in case some malformed flags cause fake
687 join_file req on client.
691 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
692 Details : ldlm_completion_ast() assumes that a lock is granted when the req
693 mode is equal to the granted mode. However, it should also check
694 that LDLM_FL_CP_REQD is not set.
698 Description: Heavy nfs access might result in deadlocks
699 Details : After ELC code landed, it is now improper to enqueue any mds
700 locks under och_sem, because enqueue might want to decide to
701 cancel open locks for same inode we are holding och_sem for.
705 Description: Client eviction while running blogbench
706 Details : A lot of unlink operations with concurrent I/O can lead to a
707 deadlock causing evictions. To address the problem, the number of
708 oustanding OST_DESTROY requests is now throttled to
709 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
710 callbacks are processed in priority.
714 Description: enable ACLs on MDS by default
715 Details : ACLs must be enabled on MDS by default.
718 Frequency : PPC/PPC64 only
720 Description: conflicts between asm-ppc64/types.h and lustre_types.h
721 Details : fix duplicated definitions between asm-ppc64/types.h and
722 lustre_types.h on PPC.
726 Description: better handle error messages in extents code
728 --------------------------------------------------------------------------------
730 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
732 * Support for kernels:
733 2.6.5-7.283 (SLES 9),
734 2.6.9-55.EL (RHEL 4),
735 2.6.16.46-0.14 (SLES 10),
736 2.6.18.8 vanilla (kernel.org)
737 * Client support for unpatched kernels:
738 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
739 2.6.16 - 2.6.22 vanilla (kernel.org)
740 * Due to recently discovered recovery problems, we do not recommend
741 using patchless RHEL 4 clients with this or any earlier release.
742 * Recommended e2fsprogs version: 1.39.cfs8
743 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
744 * Starting with this release, the ldiskfs backing filesystem required
745 by Lustre is now in its own package, lustre-ldiskfs. This package
746 should be installed. It is versioned separately from Lustre and
747 may be released separately in future.
751 Description: block reactivating mgc import until all deactivates complete
752 Details : Fix race when failing back MDT/MGS to itself (testing)
755 Frequency : at statup only
757 Description: mds_lov_synchronize race leads to various problems
758 Details : simultaneous MDT->OST connections at startup can cause the
759 sync to abort, leaving the OSC in a bad state.
761 Severity : enhancement
763 Description: add optional extra BUILD_VERSION info
764 Details : add a new environment variable (namely LUSTRE_VERS) which allows
765 to override the lustre version.
768 Frequency : 2.6.18 servers only
770 Description: ll_kern_mount() doesn't release the module reference
771 Details : The ldiskfs module reference count never drops down to 0
772 because ll_kern_mount() doesn't release the module reference.
777 Description: server LBUG when using old ost_num_threads parameter
778 Details : Accept the old ost_num_threads parameter but warn that it
779 is deprecated, and fix an off-by-one error that caused an LBUG.
784 Description: Transient SCSI error results in persistent IO issue
785 Details : iobuf->dr_error is not reinitialized to 0 between two
789 Frequency : sometimes when underlying device returns I/O errors
791 Description: OSTs not going read-only during write failures
792 Details : OSTs are not remounted read-only when the journal commit threads
793 get I/O errors because fsfilt_ext3 calls journal_start/stop()
794 instead of the ext3 wrappers.
798 Description: poor connect scaling with increasing client count
799 Details : Don't run filter_grant_sanity_check for more than 100 exports
800 to improve scaling for large numbers of clients.
803 Frequency : SLES10 only
805 Description: sanity-quota.sh quotacheck failed: rc = -22
806 Details : Quotas cannot be enabled on SLES10.
809 Frequency : liblustre clients only
811 Description: getdirentries does not give error when run on compute nodes
812 Details : getdirentries does not fail when the size specified as an argument
813 is too small to contain at least one entry
815 Severity : enhancement
817 Description: Add LNET router traceability for debug purposes
818 Details : If a checksum failure occurs with a router as part of the
819 IO path, the NID of the last router that forwarded the bulk data
820 is printed so it can be identified.
825 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
826 Details : Due to a race condition, liblustre clients were occasionally
829 Severity : enhancement
831 Description: lfs setstripe use optional parameters instead of postional
834 Severity : enhancement
836 Description: Nanosecond timestamp support for ldiskfs
837 Details : The on-disk ldiskfs filesystem has added support for nanosecond
838 resolution timestamps. There is not yet support for this at
839 the Lustre filesystem level.
842 Frequency : during server recovery
844 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
845 Details : request with rq_no_resend flag not awake l_wait_event if they get a
849 Frequency : nfs export on patchless client
851 Description: connectathon hang when test nfs export over patchless client
852 Details : Disconnected dentry cannot be found with lookup, so we do not need
853 to unhash it or make it invalid
856 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
857 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
858 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
863 Description: open req refcounting wrong on reconnect
864 Details : If reconnect happened between getting open reply from server and
865 call to mdc_set_replay_data in ll_file_open, we will schedule
866 replay for unreferenced request that we are about to free.
867 Subsequent close will crash in variety of ways.
868 Check that request is still eligible for replay in
869 mdc_set_replay_data().
874 Description: disable writes to filesystem when reading health_check file
875 Details : the default for reading the health_check proc file has changed
876 to NOT do a journal transaction and write to disk, because this
877 can cause reads of the /proc file to hang and block HA state
878 checking on a healthy but otherwise heavily loaded system. It
879 is possible to return to the previous behaviour during configure
880 with --enable-health-write.
882 Severity : enhancement
884 Description: 64-bit inode version
885 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
886 to the inode. This will be required for version-based recovery.
891 Description: MDS fails to start if a duplicate client export is detected
892 Details : in some rare cases it was possible for a client to connect to
893 an MDS multiple times. Upon recovery the MDS would detect this
894 and fail during startup. Handle this more gracefully.
896 Severity : enhancement
898 Description: Add -o localflock option to simulate old noflock
900 Details : This will achieve local-only flock/fcntl locks
906 Description: log_commit_thread vs filter_destroy race leads to crash
907 Details : Take import reference before releasing llog record semaphore
912 Description: Wrong request locking in request set processing
913 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
914 imp_delayed_list, in this place should be used imp_lock.
917 Frequency : when reconnection
919 Description: Grant Leak when osc reconnect to OST
920 Details : When osc reconnect ost, OST(filter) should check whether it
921 should grant more space to client by comparing fed_grant and
922 cl_avail_grant, and return the granted space to client instead
923 of "new granted" space, because client will call osc_init_grant
924 to update the client grant space info.
927 Frequency : when client reconnect to OST
929 Description: Grant Leak when osc do resend and replay bulk write
930 Details : When osc reconnect to OST, OST(filter)should clear grant info of
931 bulk write request, because the grant info while be sync between
932 OSC and OST when reconnect, and we should ignore the grant info
933 these of resend/replay write req.
938 Description: Grant space more than avaiable left space sometimes.
939 Details : When then OST is about to be full, if two bulk writing from
940 different clients came to OST. Accord the avaliable space of the
941 OST, the first req should be permitted, and the second one
942 should be denied by ENOSPC. But if the seconde arrived before
943 the first one is commited. The OST might wrongly permit second
944 writing, which will cause grant space > avaiable space.
947 Frequency : when client is evicted
949 Description: Grant might be wrongly erased when osc is evicted by OST
950 Details : when the import is evicted by server, it will fork another
951 thread ptlrpc_invalidate_import_thread to invalidate the
952 import, where the grant will be set to 0. While the original
953 thread will update the grant it got when connecting. So if
954 the former happened latter, the grant will be wrongly errased
955 because of this race.
960 Description: Checking Stale with correct fid
961 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
962 to check whether it is stale, because sometimes, we want the
963 enqueue happened anyway, and op_data.fid2 will not be initialized.
965 Severity : enhancement
967 Description: update patchless client
968 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
971 Frequency : only with 2.4 kernel
973 Description: random memory corruption
974 Details : size of struct ll_inode_info is to big for union inode.u and this
975 can be cause of random memory corruption.
980 Description: Memory leak in recovery
981 Details : Lov_mds_md was not free in an error handler in mds_create_object.
982 It should also check obd_fail before fsfilt_start, otherwise if
983 fsfilt_start return -EROFS,(failover mds during mds recovery).
984 then the req will return with repmsg->transno = 0 and rc = EROFS.
985 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
986 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
987 be freed no matter whether fsfilt_commit success or not.
990 Frequency : only with huge count clients
992 Description: Prevents from taking the superblock lock in llap_from_page for
994 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
995 for llap_from_page call in ll_removepage prevents from taking the
996 superblock lock for a soon died page.
1001 Description: Not check open intent error before release open handle
1002 Details : in some rare cases, the open intent error is not checked before
1003 release open handle, which may cause
1004 ASSERTION(open_req->rq_transno != 0), because it tries to release
1005 the failed open handle.
1010 Description: Set cat log bitmap only after create log success.
1011 Details : in some rare cases, the cat log bitmap is set too early. and it
1012 should be set only after create log success.
1016 Description: Accessing a block bevice can re-enable I/O when Lustre is
1017 tearing down a device.
1018 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1023 Description: service threads may hog cpus when there are a lot of requests
1024 Details : Insert cond_resched to give other threads a chance to use some CPU
1029 Description: the cat log was not initialized in recovery
1030 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1031 unlink log on mds will not be initialized until mds post
1032 recovery. And also in mds post recovery, the unlink log will
1033 initialization will be done asynchronausly, so there will be race
1034 between add unlink log and unlink log initialization.
1038 Description: brw_stats were being printed incorrectly
1039 Details : brw_stats were being printed as log2 but all of them were not
1040 recorded as log2. Also remove some code duplication arising from
1041 filter_tally_{read,write}.
1045 Frequency : rare, only in recovery.
1046 Description: ASSERTION(req->rq_type != LI_POISON) failed
1047 Details : imp_lock should be held while iterating over imp_sending_list for
1048 prevent destroy request after get timeout in ptlrpc_queue_wait.
1052 Description: replay-single.sh test 52 fails
1053 Details : A lock's skiplist need to be cleanup when it being unlinked
1054 from its resource list.
1058 Description: Short directio read returns full requested size rather than
1060 Details : Direct I/O operations should return actual amount of bytes
1061 transferred rather than requested size.
1063 Severity : enhancement
1065 Description: metadata RPC reduction (e.g. for rm performance)
1066 Details : decrease the amount of synchronous RPC between clients and servers
1067 by canceling conflicing lock before the operation on the client
1068 and packing thier handles into the main operation RPC to server.
1070 Severity : enhancement
1072 Description: Async OSC create to avoid the blocking unnecessarily.
1073 Details : If a OST has no remain object, system will block on the creating
1074 when need to create a new object on this OST. Now, ways use
1075 pre-created objects when available, instead of blocking on an
1076 empty osc while others are not empty. If we must block, we block
1077 for the shortest possible period of time.
1079 Severity : enhancement
1081 Description: refine locking for avoid write wrong info into lov_objid file
1082 Details : fix possible races with add new target and write/update data in
1085 --------------------------------------------------------------------------------
1087 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1092 Frequency : on some architectures
1094 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1095 Details : Uninitialized flags sometimes cause configuration commands to
1099 Frequency : patchless clients only
1101 Description: missing __iget() symbol export
1102 Details : The __iget() symbol export is missing. To avoid the need for
1103 this on patchless clients the deathrow inode reaper is turned
1104 off, and we depend on the VM to clean up old inodes. This
1105 dependency was during via the fix for bug 12181.
1107 --------------------------------------------------------------------------------
1109 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1111 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1112 INTEROPERATE with older versions automatically. In many cases a
1113 special upgrade step is needed. Please read the
1114 user documentation before upgrading any part of a live system.
1115 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1116 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1117 * WARNING: Lustre configuration and startup changes are required with
1118 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1124 Severity : enhancement
1126 Description: MountConf
1127 Details : Lustre configuration is now managed via mkfs and mount
1128 commands instead of lmc and lconf. New obd types (MGS, MGC)
1129 are added for dynamic configuration management. See
1130 https://mail.clusterfs.com/wikis/lustre/MountConf for
1133 Severity : enhancement
1135 Description: dynamic OST addition
1136 Details : OSTs can now be added to a live filesystem
1138 Severity : enhancement
1140 Description: startup order invariance
1141 Details : MDTs and OSTs can be started in any order. Clients only
1142 require the MDT to complete startup.
1144 Severity : enhancement
1146 Description: parallel, asynchronous orphan cleanup
1147 Details : orphan cleanup is now performed in separate threads for each
1148 OST, allowing parallel non-blocking operation.
1150 Severity : enhancement
1152 Description: optimized stripe assignment
1153 Details : stripe assignments are now made based on ost space available,
1154 ost previous usage, and OSS previous usage, in order to try
1155 to optimize storage space and networking resources.
1157 Severity : enhancement
1159 Description: Permanently set tunables
1160 Details : All writable /proc/fs/lustre tunables can now be permanently
1161 set on a per-server basis, at mkfs time or on a live system.
1163 Severity : enhancement
1165 Description: Lustre message v2
1166 Details : Add lustre message format v2.
1168 Severity : enhancement
1170 Description: client OST exclusion list
1171 Details : Clients can be started with a list of OSTs that should be
1172 declared "inactive" for known non-responsive OSTs.
1176 Description: SPEC SFS validation failure on NFS v2 over lustre.
1177 Details : Changes the blocksize for regular files to be 2x RPC size,
1178 and not depend on stripe size.
1180 Severity : enhancement
1182 Description: Multiple MD RPCs in flight.
1183 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1184 To avoid overly-overloading MDS, introduce a limit on number of
1185 MDS RPCs in flight for a single client and add /proc controls
1186 to adjust this limit.
1188 Severity : enhancement
1190 Description: client read/write statistics
1191 Details : Add client read/write call usage stats for performance
1192 analysis of user processes.
1193 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1194 file access. extents_stats shows chunk size distribution.
1195 extents_stats_per_process show chunk size distribution per
1198 Severity : enhancement
1200 Description: mds statistics
1201 Details : Add detailed mds operations statistics in
1202 /proc/fs/lustre/mds/*/stats.
1206 Description: Failure of copying files with lustre special EAs.
1207 Details : Client side always return success for setxattr call for lustre
1208 special xattr (currently only "trusted.lov").
1213 Description: Refcount LNET uuids
1214 Details : The global LNET uuid list grew linearly with every startup;
1215 refcount repeated list entries instead of always adding to
1218 Severity : enhancement
1220 Description: Dynamic service threads
1221 Details : Within a small range, start extra service threads
1222 automatically when the request queue builds up.
1225 Frequency : mixed-endian client/server environments
1227 Description: mixed-endian crashes
1228 Details : The new msg_v2 system had some failures in mixed-endian
1231 Severity : enhancement
1233 Description: Easy OST removal
1234 Details : OSTs can be permanently deactivated with e.g. 'lctl
1235 conf_param lustre-OST0001.osc.active=0'
1237 Severity : enhancement
1239 Description: MGS proc entries
1240 Details : Added basic proc entries for the MGS showing what filesystems
1243 Severity : enhancement
1245 Description: provide MGS failover
1246 Details : Added config lock reacquisition after MGS server failover.
1248 Severity : enhancement
1250 Description: add Linux 2.4 support
1251 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1255 Description: a large application tries to do I/O to the same resource and dies
1256 in the middle of it.
1257 Details : Check the req->rq_arrival time after the call to
1258 ost_brw_lock_get(), but before we do anything about
1259 processing it & sending the BULK transfer request. This
1260 should help move old stale pending locks off the queue as
1261 quickly as obd_timeout.
1264 Frequency : when an incorrect nid is specified during startup
1266 Description: ptlrpc connect to non-existant node causes kernel crash
1267 Details : LNET can't be re-entered from an event callback, which
1268 happened when we expire a message after the export has been
1269 cleaned up. Instead, hand the zombie cleanup off to another
1272 Severity : enhancement
1274 Description: plain/inodebits lock performance improvement
1275 Details : Grouping plain/inodebits in granted list by their request modes
1276 and bits policy, thus improving the performance of search through
1280 Frequency : only if OST filesystem is corrupted
1282 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1283 Details : for a short time RPCs with bulk IO are in the replay list,
1284 but replay of bulk IOs is unimplemented. If the OST filesystem
1285 is corrupted due to disk cache incoherency and then replay is
1286 started it is possible to trip an assertion. Avoid putting
1287 committed RPCs into the replay list at all to avoid this issue.
1290 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1292 Description: System hang on startup
1293 Details : This bug allowed the liblustre (e.g. catamount) client to
1294 return to the app before handling all startup RPCs. This
1295 could leave the node unresponsive to lustre network traffic
1296 and manifested as a server ptllnd timeout.
1298 Severity : enhancement
1300 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1301 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1302 boolean to control whether to print peer debug info when a
1303 client's RPC times out.
1306 Frequency : only for kernels with patches from Lustre below 1.4.3
1308 Description: Remove old rdonly API
1309 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1312 Frequency : only for devices with external journals
1314 Description: Set external device read-only also
1315 Details : During a commanded failover stop, we set the disk device
1316 read-only while the server shuts down. We now also set any
1317 external journal device read-only at the same time.
1320 Frequency : when upgrading from 1.4 while trying to change parameters
1322 Description: The wrong (new) MDC name was used when setting parameters for
1323 upgraded MDT's. Also allows changing of OSC (and MDC)
1324 parameters if --writeconf is specified at tunefs upgrade time.
1327 Frequency : when setting specific ost indicies
1329 Description: QOS code breaks on skipped indicies
1330 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1331 created with --index need not be sequential.
1333 Severity : enhancement
1335 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1336 Details : The uninit_groups feature works in conjunction with the kernel
1337 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1338 up the pass1 processing of e2fsck. This is a read-only feature
1339 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1340 mount filesystems that have had this feature enabled.
1342 Severity : enhancement
1344 Description: Improve multi-block allocation algorithm to avoid fragmentation
1345 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1346 allocation locality and avoid filesystem fragmentation.
1348 ------------------------------------------------------------------------------
1350 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1352 * Support for kernels:
1353 2.4.21-47.0.1.EL (RHEL 3)
1354 2.6.5-7.283 (SLES 9)
1355 2.6.9-42.0.10.EL (RHEL 4)
1356 2.6.12.6 vanilla (kernel.org)
1357 2.6.16.27-0.9 (SLES 10)
1358 * Recommended e2fsprogs version: 1.39.cfs5
1360 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1364 Frequency : occasional, depends on client load and configuration
1365 Bugzilla : 12181, 12203
1366 Description: data loss for recently-modified files
1368 Details : In some cases it is possible that recently written or created
1369 files may not be written to disk in a timely manner (this should
1370 normally be within 30s unless client IO load is very high).
1371 The problem appears as zero-length files or files that are a
1372 multiple of 1MB in size after a client crash or client eviction
1373 that are missing data at the end of the file.
1375 This problem is more likely to be hit on clients where files are
1376 repeatedly created and unlinked in the same directory, clients
1377 have a large amount of RAM, have many CPUs, the filesystem has
1378 many OSTs, the clients are rebooted frequently, and/or the files
1379 are not accessed by other nodes after being written.
1381 The presence of the problem can be detected by looking at
1382 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1383 smaller than the second (nr_unused) then dirty files will not
1384 be flushed automatically to disk. "sync; sleep 10" should be
1385 run several times on the node before unmounting it to update
1386 Lustre (this is also safe to run on nodes without this problem).
1388 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1389 that can cause this same problem, so customers using that kernel
1390 also need to update the kernel in addition to Lustre. In order
1391 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1393 It is normal that files written just before a client crash (less
1394 than 30s) may not yet have been flushed to disk, even for local
1398 Frequency : frequent on thin XT3 nodes
1400 Description: UUID collision on thin XT3 Linux nodes
1401 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1402 randomly, since we relied on an insufficiently-seeded PRNG.
1407 Description: OSS hangs after "All ost request buffers busy"
1408 Details : A deadlock between quota and journal operations caused OSS
1409 hangs after printing "All ost request buffers busy."
1412 Frequency : always on liblustre builds
1414 Description: Cleanup compiler warnings on liblustre
1417 Frequency : always on liblustre builds on XT3
1419 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1423 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1424 > 32000 subdirectories
1425 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1426 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1427 subdirectory count crosses 32000. This will aid e2fsck to
1428 correctly handle more than 32000 subdirectories.
1431 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1433 Description: System hang on startup
1434 Details : This bug allowed the liblustre (e.g. catamount) client to
1435 return to the app before handling all startup RPCs. This
1436 could leave the node unresponsive to lustre network traffic
1437 and manifested as a server ptllnd timeout.
1439 Severity : enhancement
1441 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1442 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1443 boolean to control whether to print peer debug info when a
1444 client's RPC times out.
1449 Description: make O_SYNC working on 2.6 kernels
1450 Details : 2.6 kernels use different method for mark pages for write,
1451 so need add a code to lustre for O_SYNC work.
1456 Description: Failure to close file and release space on NFS
1457 Details : Put inode details into lock acquired in ll_intent_file_open.
1458 Use mdc_intent_lock in ll_intent_open to properly
1459 detect all kind of errors unhandled by mdc_enqueue.
1464 Description: proc file read during shutdown sometimes raced obd removal,
1466 Details : Add lock to prevent obd access after proc file removal.
1469 Frequency : Only for files larger than 4GB on 32-bit clients.
1471 Description: improperly doing page alignment of locks
1472 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1473 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1478 Description: RPCs being resent when they shouldn't be
1479 Details : Some RPCs that should not be resent are being resent. This
1480 can cause inconsistencies in the RPC state machine. Do not
1481 resend such requests.
1484 Frequency : rare, only with NFS export
1486 Description: Crash on NFS re-export node
1487 Details : under very unusual load conditions an assertion is hit in
1488 ll_intent_file_open()
1491 Frequency : only if OST filesystem is corrupted
1493 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1494 Details : for a short time RPCs with bulk IO are in the replay list,
1495 but replay of bulk IOs is unimplemented. If the OST filesystem
1496 is corrupted due to disk cache incoherency and then replay is
1497 started it is possible to trip an assertion. Avoid putting
1498 committed RPCs into the replay list at all to avoid this issue.
1503 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1504 Details : Large single O_DIRECT read and write calls can fail to allocate
1505 a sufficiently large buffer to process the request. In case of
1506 allocation failure the allocation is retried with a smaller
1507 buffer and broken into smaller requests.
1509 Severity : enhancement
1511 Description: Add -o localflock option to simulate old noflock behaviour.
1512 Details : This will achieve local-only flock/fcntl locks coherentness.
1517 Description: versioning check is incomplete
1518 Details : Checking the version difference of client vs. server, report
1519 error if the gap is too big.
1524 Description: add support PG_writeback bit
1525 Details : add support for PG_writeback bit for Lustre, for more carefull
1526 work with page cache in 2.6 kernel. This also fix some deadlocks
1527 and remove hack for work O_SYNC with 2.6 kernel.
1529 Severity : enhancement
1531 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1532 Details : The uninit_groups feature works in conjunction with the kernel
1533 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1534 up the pass1 processing of e2fsck. This is a read-only feature
1535 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1536 mount filesystems that have had this feature enabled.
1538 Severity : enhancement
1540 Description: Improve multi-block allocation algorithm to avoid fragmentation
1541 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1542 allocation locality and avoid filesystem fragmentation.
1544 ------------------------------------------------------------------------------
1546 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1548 * Support for kernels:
1549 2.6.9-42.0.3.EL (RHEL 4)
1550 2.6.5-7.276 (SLES 9)
1551 2.4.21-47.0.1.EL (RHEL 3)
1552 2.6.12.6 vanilla (kernel.org)
1553 2.6.16.21-0.8 (SLES10)
1554 * Recommended e2fsprogs version: 1.39.cfs2-0
1556 * The backwards-compatible /proc/sys/portals symlink has been removed
1557 in this release. Before upgrading, please ensure that you change
1558 any configuration scripts or /etc/sysctl.conf files that access
1559 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1560 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1561 in advance of the upgrade on any system running Lustre 1.4.6 or
1562 newer, since /proc/sys/lnet was added in that version.
1563 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1567 Frequency : only when quota is used
1569 Description: avoid scanning export list for quota master
1570 Details : Change the algorithms to avoid scanning export list in order
1571 to improve the efficiency.
1574 Frequency : MDS failover only, very rarely
1576 Description: "went back in time" messages on mds failover
1577 Details : The greatest transno may be lost when the current operation
1578 finishes with an error (transno==0) and the client's last_rcvd
1579 record is over-written. Save the greatest transno in the
1580 mds_last_transno for this case.
1583 Frequency : always for specific kernels and striping counts
1585 Description: client may get "Matching packet too big" without ACL support
1586 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1587 when trying to access files in certain configurations. The
1588 clients should in fact be denied when mounting because they do
1589 not understand ACLs.
1592 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1594 Description: many clients connecting with IO in progress causes connect timeouts
1595 Details : Avoid synchronous journal commits to avoid delays caused by many
1596 clients connecting/disconnecting when bulk IO is in progress.
1597 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1598 OST_IO_PORTAL to avoid delays behind potentially many pending
1602 Frequency : occasionally with multiple writers to a single file
1604 Description: shared writes to file may result in wrong size reported by stat()
1605 Details : Allow growing of kms when extent lock is cancelled
1608 Frequency : always with random mmap IO to multi-striped file
1610 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1611 Details : Only if the hole is at the end of OST object so that kms is too
1612 small. Fix is to increase kms accordingly in ll_nopage.
1615 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1617 Description: writes to a missing object would leak memory on the OST
1618 Details : If there is an inconsistency between the MDS and OST filesystems,
1619 such that the MDS references an object that doesn't exist, writes
1620 to that object will leak memory due to incorrect cleanup in the
1621 error handling path, eventually running out of memory on the OST.
1626 Description: Creating too long symlink causes lustre errors
1627 Details : Check symlink and name lengths before sending requests to MDS.
1630 Frequency : only if flock is enabled (not on by default)
1632 Description: posix locks not released on fd closure on 2.6.9+
1633 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1634 kernels, this caused such locks not to be released on fd close and
1635 then assertions on fs unmount about still used locks.
1638 Frequency : MDS failover only, very rarely
1640 Description: clients may get ASSERTION(granted_lock != NULL)
1641 Details : When request was taking a long time, and a client was resending
1642 a getattr by name lock request. The were multiple lock requests
1643 with the same client lock handle and
1644 mds_getattr_name->fixup_handle_for_resent_request found one of the
1645 lock handles but later failed with ASSERTION(granted_lock != NULL).
1650 Description: handle->h_buffer_credits > 0, assertion failure
1651 Details : h_buffer_credits is zero after truncate, causing assertion
1652 failure. This patch extends the transaction or creates a new
1656 Frequency : NFS re-export or patchless client
1657 Bugzilla : 11179, 10796
1658 Description: Crash on NFS re-export node (__d_move)
1659 Details : We do not want to hash the dentry if we don't have a lock.
1660 But if this dentry is later used in d_move, we'd hit uninitialised
1661 list head d_hash, so we just do this to init d_hash field but
1662 leave dentry unhashed.
1665 Frequency : NFS re-export or patchless client
1667 Description: NFS exports has problem with symbolic link
1668 Details : lustre client didn't properly install dentry when re-exported
1669 to NFS or running patchless client.
1672 Frequency : NFS re-export or patchless client
1674 Description: Various nfs/patchless fixes.
1675 Details : fixes reuse disconected alias for lookup process - this fixes
1676 warning "find_exported_dentry: npd != pd",
1677 fix permission error with open files at nfs.
1678 fix apply umask when do revalidate.
1681 Frequency : occasional
1683 Description: Crash on NFS re-export node
1684 Details : calling clear_page() on the wrong pointer triggered oops in
1685 generic_mapping_read().
1688 Frequency : rarely, using O_DIRECT IO
1690 Description: unaligned directio crashes client with LASSERT
1691 Details : check for unaligned buffers before trying any requests.
1694 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1696 Description: stale data returned from RAID cache
1697 Details : If only a small amount of IO is done to the RAID device before
1698 reading it again it is possible to get stale data from the RAID
1699 cache instead of reading it from disk.
1702 Frequency : always for sles10 kernel
1704 Description: sles10 support
1705 Details : ll_follow_link: compile fixes and using of nd_set_link
1706 under newer kernels.
1709 Frequency : depends on arch, kernel and compiler version, always on sles10
1712 Description: recursive or deep enough symlinks cause stack overflow
1713 Details : getting rid of large stack-allocated variable in
1717 Frequency : depends on hardware
1719 Description: lustre write performance loss in the SLES10 kernel
1720 Details : the performance loss is caused by using of write barriers in the
1721 ext3 code. The SLES10 kernel turns barrier support on by
1722 default. The fix is to undo that change for ldiskfs.
1724 ------------------------------------------------------------------------------
1726 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1728 * Support for kernels:
1729 2.6.9-42.0.3EL (RHEL 4)
1730 2.6.5-7.276 (SLES 9)
1731 2.4.21-47.0.1.EL (RHEL 3)
1732 2.6.12.6 vanilla (kernel.org)
1736 Frequency : quota enabled and large files being deleted
1738 Description: releasing more than 4GB of quota at once hangs OST
1739 Details : If a user deletes more than 4GB of files on a single OST it
1740 will cause the OST to spin in an infinite loop. Release
1741 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1746 Description: statfs data retrieved from /proc may be stale or zero
1747 Details : When reading per-device statfs data from /proc, in the
1748 {kbytes,files}_{total,free,avail} files, it may appear
1749 as zero or be out of date.
1752 Frequency : systems with MD RAID1 external journal devices
1754 Description: lconf's call to blkid is confused by RAID1 journal devices
1755 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1756 of returning all block devices that match the journal UUID.
1759 Frequency : always, for aggregate stripe size over 4GB
1761 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1762 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1763 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1766 Frequency : always if "lfs find" used on a local file/directory
1768 Description: "lfs find" segfaults if used on a local file/directory
1769 Details : The case where a directory component was not specified wasn't
1770 handled correctly. Handle this properly.
1773 Frequency : always on ppc64
1775 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1776 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1777 perform endianness conversion.
1780 Frequency : rarely (truncate to non-zero file size after write under load)
1781 Bugzilla : 10730, 10687
1782 Description: Files padded with zeros to next 4K multiple
1783 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1784 it is possible that files that are truncated to a non-zero size
1785 immediately after being written are filled with zero bytes beyond
1786 the truncated size. No file data is lost.
1788 Severity : enhancement
1790 Description: Allow recovery/failover for liblustre clients.
1791 Details : liblustre clients were unaware of failover configurations until
1794 Severity : enhancement
1796 Description: user file locks should fail when not mounting with flock option
1797 Details : Set up an error-returning stub in ll_file_operations.lock field
1798 to prevent incorrect behaviour when client is mounted without
1799 flock option. Also, set up properly f_op->flock field for
1803 Frequency : always on ia64
1805 Description: "lfs df" loops on printing out MDS statfs information
1806 Details : The obd_ioctl_data was not initialized and in some systems
1807 this caused a failure during the ioctl that did not return
1808 an error. Initialize the struct and return an error on failure.
1811 Frequency : SLES 9 only
1813 Description: Error of copying files with lustre special EAs as root
1814 Details : Client side always return success for setxattr call for lustre
1815 special xattr (currently only "trusted.lov").
1818 Frequency : rarely on clusters with both ia64+i386 clients
1820 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1821 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1822 clients are doing concurrent writes to the same file it is
1823 possible that the ia64 clients may LASSERT because the OST
1824 extent locks are not PAGE_SIZE aligned. Ensure that grown
1825 locks are always aligned on the request boundary.
1828 Frequency : specific use, occasional
1830 Description: Overwriting in use executable truncates on-disk binary image
1831 Details : If one node attempts to overwrite an executable in use by
1832 another node, we now correctly return ETXTBSY instead of
1833 truncating the file.
1835 Severity : enhancement
1837 Description: Async OSC create to avoid the blocking unnecessarily.
1838 Details : If a OST has no remain object, system will block on the creating
1839 when need to create a new object on this OST. Now, ways use
1840 pre-created objects when available, instead of blocking on an
1841 empty osc while others are not empty. If we must block, we block
1842 for the shortest possible period of time.
1847 Description: chmod on Lustre root is propagated to other clients
1848 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1849 invalid by the follow_mount time.
1852 Frequency : liblustre clients only
1854 Description: Race in 'instant cancel' lock handling could lead to such locks
1855 never to be granted in case of SMP MDS
1856 Details : Do not destroy not yet granted but cbpending locks in
1860 Frequency : replay/resend of open
1862 Description: non null lock assetion failure in mds_intent_policy
1863 Details : Trying to replay/resend lockless open requests resulted in
1864 mds_open() returning 0 with no lock. Now it sets a flag if
1865 a lock is going to be returned.
1867 Severity : enhancement
1869 Description: Checksum enhancements
1870 Details : New checksum enhancements allow for resending RPCs that failed
1873 Severity : enhancement
1875 Description: Tunables on number of dirty pages in cacche
1876 Details : Allow to set limit on number of dirty pages cached.
1881 Description: client crash on unmount - lock still has references
1882 Details : In some error handling cases it was possible to leak a lock
1883 reference on a client while accessing a file. This was not
1884 harmful to the client during operation, but would cause the
1885 client to crash when the filesystem is unmounted.
1888 Frequency : specific case, rare
1890 Description: ETXTBSY on mds though file not in use by client
1891 Details : ETXTBSY is no longer incorrectly returned when attempting to
1892 chmod or chown a directory that the user previously tried to
1893 execute or a currently-executing binary.
1896 Frequency : extremely rare except on liblustre-based clients
1898 Description: Lustre space not freed when files are deleted
1899 Details : Clean up open-unlinked files after client eviction. Previously
1900 the unlink was skipped and the files remained as orphans.
1905 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1906 Details : In some cases it was possible to send a blocking callback to
1907 a client doing a glimpse, even though that client didn't get
1908 a lock granted. When the glimpse lock is cancelled on the OST
1909 the freed lock is left on the waiting list and corrupted the list.
1912 Frequency : all core dumps
1914 Description: Broke core dumps to lustre
1915 Details : Negative dentry may be unhashed if parent does not have UPDATE
1916 lock, but some callers, e.g. do_coredump, expect dentry to be
1917 hashed after successful create, hash it in ll_create_it.
1919 ------------------------------------------------------------------------------
1921 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1923 * Support for kernels:
1924 2.6.9-42.0.2.EL (RHEL 4)
1925 2.6.5-7.276 (SLES 9)
1926 2.4.21-40.EL (RHEL 3)
1927 2.6.12.6 vanilla (kernel.org)
1931 Frequency : always on RHEL 3
1933 Description: Number of open files grows over time
1934 Details : The number of open files grows over time, whether or not
1935 Lustre is started. This was due to a filp leak introduced
1936 by one of our kernel patches.
1938 ------------------------------------------------------------------------------
1940 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1942 * Support for kernels:
1943 2.6.9-42.EL (RHEL 4)
1944 2.6.5-7.276 (SLES 9)
1945 2.4.21-40.EL (RHEL 3)
1946 2.6.12.6 vanilla (kernel.org)
1951 Bugzilla : 5719, 9635, 9792, 9684
1952 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1953 Details : If a server is under heavy load and cannot reply to new
1954 connection requests before the client resends the (re)connect,
1955 the connection handling code can behave badly if two service
1956 threads are concurrently handing separate (re)connections from
1957 the same client. Add better locking to the connection handling
1958 code, and ensure that only a single connection will be processed
1959 for a given client UUID, even if the lock is dropped.
1961 Severity : enhancement
1963 Description: add TCP zero-copy support to kernel
1964 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1965 sends if the hardware supports scatter-gather and checksumming.
1966 This allows socklnd to do client-write and server-read more
1967 efficiently and reduce CPU utilization from skbuf copying.
1970 Frequency : only if NFS exporting from client
1972 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1973 Details : If setting attributes on a file created under NFS that had
1974 never been opened it would be possible to oops the client
1975 if the file had no objects.
1978 Frequency : always for liblustre
1980 Description: liblustre client does MDS+OSTs setattr RPC for each write
1981 Details : When doing a write from a liblustre client, the client
1982 incorrectly issued an RPC to the MDS and each OST the file was
1983 striped over in order to update the timestamps. When writing
1984 with small chunks and many clients this could overwhelm the MDS
1985 with RPCs. In all cases it would slow down the write because
1986 these RPCs are unnecessary.
1988 Severity : enhancement
1990 Description: allow number of MDS service threads to be changed at module load
1991 Details : It is now possible to change the number of MDS service threads
1992 running. Adding "options mds mds_num_threads={N}" to the MDS's
1993 /etc/modprobe.conf will set the number of threads for the next
1994 time Lustre is restarted (assuming the "mds" module is also
1995 reloaded at that time). The default number of threads will
1996 stay the same, 32 for most systems.
2001 Description: OST crash if filesystem is unformatted or corrupt
2002 Details : If an OST is started on a device that has never been formatted
2003 or if the filesystem is corrupt and cannot even mount then the
2004 error handling cleanup routines would dereference a NULL pointer.
2009 Description: NULL pointer deref in llap_from_page.
2010 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2011 incorrect) mapping if the page was truncated/reclaimed while it was
2012 searched for. Check for this condition and skip such pages when
2013 doing readahead. Introduce extra check to llap_from_page() to
2014 verify page->mapping->host is non-NULL (so page is not anonymous).
2017 Frequency : Sometimes when using sys_sendfile
2019 Description: "page not covered by a lock" warnings from ll_readpage
2020 Details : sendfile called ll_readpage without right page locks present.
2021 Now we introduced ll_file_sendfile that does necessary locking
2022 around call to generic_file_sendfile() much like we do in
2026 Frequency : with certain MDS communication failures at client mount time
2028 Description: NULL pointer deref after failed client mount
2029 Details : a client connection request may delayed by the network layer
2030 and not be sent until after the PTLRPC layer has timed out the
2031 request. If the client fails the mount immediately it will try
2032 to clean up before the network times out the request. Add a
2033 reference from the request import to the obd device and delay
2034 the cleanup until the network drops the request.
2037 Frequency : occasionally during client (re)connect
2039 Description: assertion failure during client (re)connect
2040 Details : processing a client connection request may be delayed by the
2041 client or server longer than the client connect timeout. This
2042 causes the client to resend the connection request. If the
2043 original connection request is replied in this interval, the
2044 client may trip an assertion failure in ptlrpc_connect_interpret()
2045 which thought it would be the only running connect process.
2048 Frequency : only with obd_echo servers and clients that are rebooted
2050 Description: kernel BUG accessing uninitialized data structure
2051 Details : When running an obd_echo server it did not start the ping_evictor
2052 thread, and when a client was evicted an uninitialized data
2053 structure was accessed. Start the ping_evictor in the RPC
2054 service startup instead of the OBD startup.
2056 Severity : enhancement
2057 Bugzilla : 10193 (patchless)
2058 Description: Remove dependency on various unexported kernel interfaces.
2059 Details : No longer need reparent_to_init, exit_mm, exit_files,
2060 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2063 Frequency : rare (only users of deprecated and unsupported LDAP config)
2065 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2066 Details : LDAP apparently contains 'lustreName' attributes instead of
2067 'name'. A simple remapping of the name is sufficient.
2070 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2072 Description: waiting_locks_callback trips kernel BUG if client is evicted
2073 Details : Running with the dump_on_timeout debug flag turned on makes
2074 it possible that the waiting_locks_callback() can try to dump
2075 the Lustre kernel debug logs from an interrupt handler. Defer
2076 this log dumping to the expired_lock_main() thread.
2078 Severity : enhancement
2080 Description: Support NFS exporting on 2.6 kernels.
2081 Details : Implement non-rawops metadata methods for NFS server to use without
2082 changing NFS server code.
2085 Frequency : very rare (synthetic metadata workload only)
2087 Description: two racing renames might cause an MDS thread to deadlock
2088 Details : Running the "racer" program may cause one MDS thread to rename
2089 a file from being the source of a rename to being the target of
2090 a rename at exactly the same time that another thread is doing
2091 so, and the second thread has already enqueued these locks after
2092 doing a lookup of the target and is trying to relock them in
2093 order. Ensure that we don't try to re-lock the same resource.
2096 Frequency : only very large systems with liblustre clients
2098 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2099 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2100 all OSTs in parallel. This allows the eviction of stale liblustre
2101 clients to proceed much faster than if they were done in series,
2102 and also offers similar improvements for other set_info RPCs.
2107 Description: excessive CPU usage during initial read phase on client
2108 Details : During the initial read phase on a client, it would agressively
2109 retry readahead on the file, consuming too much CPU and impacting
2110 performance (since 1.4.5.8). Improve the readahead algorithm
2111 to avoid this, and also improve some other common cases (read
2112 of small files in particular, where "small" is files smaller than
2113 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2118 Description: MDS crash when receiving packet with unknown intent.
2119 Details : Do not LBUG in unknown intent case, just return -EFAULT
2121 Severity : enhancement
2122 Bugzilla : 9293, 9385
2123 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2124 Details : Do not serialize getattr (non-intent version) and statfs.
2127 Frequency : occasional, when OST network is overloaded/intermittent
2129 Description: client evicted by OST after bulk IO timeout
2130 Details : If a client sends a bulk IO request (read or write) the OST
2131 may evict the client if it is unresposive to its data GET/PUT
2132 request. This is incorrect if the network is overloaded (takes
2133 too long to transfer the RPC data) or dropped the OST GET/PUT
2134 request. There is no need to evict the client at all, since
2135 the pinger and/or lock callbacks will handle this, and the
2136 client can restart the bulk request.
2139 Frequency : Always when mmapping file with no objects
2141 Description: client crashes when mmapping file with no objects
2142 Details : Check that we actually have objects in a file before doing any
2143 operations on objects in ll_vm_open, ll_vm_close and
2149 Description: Request leak when working with deleted CWD
2150 Details : Introduce advanced request refcount tracking for requests
2151 referenced from lustre intent.
2153 Severity : Enhancement
2155 Description: Cache open file handles on client.
2156 Details : MDS now will return special lock along with openhandle, if
2157 requested and client is allowed to hold openhandle, even if unused,
2158 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2159 closing files for every read/write openration.
2161 Severity : Enhancement
2163 Description: Cache open negative dentries on client when possible.
2164 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2165 negative dentries on lock revocation.
2170 Description: Remounting a client read-only wasn't possible with a zconf mount
2171 Details : It wasn't possible to remount a client read-only with llmount.
2173 Severity : enhancement
2174 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2175 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2176 MPICH is included to add Lustre-specific ADIO interfaces.
2177 This is based closely on the UFS ADIO layer and only differs
2178 in file creation, in order to allow the OST striping to be set.
2179 This is user-contributed code and not supported by CFS.
2184 Description: extended inode attributes (immutable, append-only) work improperly
2185 when 2.4 and 2.6 kernels are used on client/server or vice versa
2186 Details : Introduce kernel-independent values for these flags.
2188 Severity : enhancement
2191 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2192 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2193 granularity. Now, Fractional MB granularity is be supported,
2194 this is very useful for low memory system.
2196 Severity : enhancement
2198 Description: Getattr by fid
2199 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2200 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2201 have locks and avoids getattr rpc requests for it on every stat.
2204 Frequency : Always, for filesystems larger than 2TB
2206 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2207 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2208 filesystems larger than 2TB to be mounted with mballoc enabled.
2211 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2213 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2214 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2215 filesystems larger than 2TB is mounted then the kernel will
2216 silently corrupt the start of the filesystem. CONFIG_LBD is
2217 enabled for all CFS-supported kernels, but the possibility of
2218 this happening with a modified kernel config exists.
2220 Severity : enhancement
2222 Description: add client O_DIRECT support for 2.6 kernels
2223 Details : It is now possible to do O_DIRECT reads and writes to files
2224 in the Lustre client mountpoint on 2.6 kernel clients.
2226 Severity : enhancement
2228 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2229 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2230 parallel, not waiting for response from every OST before sending
2231 a rpc to the next OST.
2236 Description: setattr vs write race when updating file timestamps
2237 Details : Client processes that update a file timestamp into the past
2238 right after writing to the file (e.g. tar) it is possible that
2239 the updated file modification time can be reset to the current
2240 time due to a race between processing the setattr and write RPC.
2242 Severity : enhancement
2244 Description: Bring 'lfs find' closer in line with regular Linux find.
2245 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2246 -print0 options and obtains all the needed info through the lustre
2249 Severity : enhancement
2251 Description: support up to 1024 configured devices on one node
2252 Details : change obd_dev array from statically allocated to dynamically
2253 allocated structs as they are first used to reduce memory usage
2258 Description: Flush dirty partially truncated pages during truncate
2259 Details : Immediatelly flush partially truncated pages in filter_setattr,
2260 this way we completely avoid having any pages in page cache on OST
2261 and can retire ugly workarounds during writes to flush such pages.
2266 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2267 Details : protect inode from truncation within vfs_unlink() context
2268 just take a reference before calling vfs_unlink() and release it
2269 when parent's i_sem is free.
2274 Description: last_id value checked outside lock on OST caused LASSERT failure
2275 Details : If there were multiple MDS->OST object precreate requests in
2276 flight, it was possible that the OST's last object id was checked
2277 outside a lock and incorrectly tripped an assertion. Move checks
2278 inside locks, and discard old precreate requests.
2281 Frequency : always, if extents are used on OSTs
2283 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2284 Details : OSTs using the extents format would not zero the high 16 bits of
2285 the index physical block number. This is not a problem for any
2286 OST filesystems smaller than 16TB, and no kernels support ext3
2287 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2288 new/modified files) and can be fixed for existing filesystems
2289 with e2fsprogs-1.39-cfs1.
2294 Description: import connection selection may be incorrect if timer wraps
2295 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2296 import connections to be ignored if the 32-bit jiffies counter
2297 wraps. Use a 64-bit jiffies counter.
2300 Frequency : very large clusters immediately after boot
2302 Description: LNET request buffers exhausted under heavy short-term load
2303 Details : If a large number of client requests are generated on a service
2304 that has previously never seen so many requests it is possible
2305 that the request buffer growth cannot keep up with the spike in
2306 demand. Instead of dropping incoming requests, they are held in
2307 the LND until the RPC service can accept more requests.
2310 Frequency : Sometimes during replay
2312 Description: Assertion failure in ll_local_open after replay.
2313 Details : If replay happened on an open request reply before we were able
2314 to set replay handler, reply will become not swabbed tripping the
2315 assertion in ll_local_open. Now we set the handler right after
2316 recognising of open request
2319 Frequency : very rare
2321 Description: kernel reports "badness in vsnprintf"
2322 Details : Reading from the "recovery_status" /proc file in small chunks
2323 may cause a negative length in lprocfs_obd_rd_recovery_status()
2324 call to vsnprintf() (which is otherwise harmless). Exit early
2325 if there is no more space in the output buffer.
2327 Severity : enhancement
2329 Description: clear OBD RPC statistics by writing to them
2330 Details : It is now possible to clear the OBD RPC statistics by writing
2331 to the "stats" file.
2336 Description: Client mtime is not the same on different clients after utimes
2337 Details : In some cases, the client was using the utimes() syscall on
2338 a file cached on another node. The clients now validate the
2339 ctime from the MDS + OSTs to determine which one is right.
2344 Description: Inability to activate failout mode
2345 Details : lconf script incorrectly assumed that in pythong string's numeric
2346 value is used in comparisons.
2349 Frequency : always with multiple stripes per file
2351 Description: Inefficient object allocation for mutli-stripe files
2352 Details : When selecting which OSTs to stripe files over, for files with
2353 a stripe count that divides evenly into the number of OSTs,
2354 the MDS is always picking the same starting OST for each file.
2355 Return the OST selection heuristic to the original design.
2360 Description: mount failures may take full timeout to return an error
2361 Details : Under some heavy load conditions it is possible that a
2362 failed mount can wait for the full obd_timeout interval,
2363 possibly several minutes, before reporting an error.
2364 Instead return an error as soon as the status is known.
2366 Frequency : quota enabled and large files being deleted
2368 Description: releasing more than 4GB of quota at once hangs OST
2369 Details : If a user deletes more than 4GB of files on a single OST it
2370 will cause the OST to spin in an infinite loop. Release
2371 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2376 Description: statfs data retrieved from /proc may be stale or zero
2377 Details : When reading per-device statfs data from /proc, in the
2378 {kbytes,files}_{total,free,avail} files, it may appear
2379 as zero or be out of date.
2382 Frequency : systems with MD RAID1 external journal devices
2384 Description: lconf's call to blkid is confused by RAID1 journal devices
2385 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2386 of returning all block devices that match the journal UUID.
2389 Frequency : always, for aggregate stripe size over 4GB
2391 Description: assertion fails when trying to use 4GB stripe size
2392 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2393 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2396 Frequency : always on ppc64
2398 Description: the first write on an ext3 filesystem with mballoc got stuck
2399 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2400 perform endianness conversion.
2402 ------------------------------------------------------------------------------
2404 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2406 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2407 INTEROPERATE with older versions automatically. Please read the
2408 user documentation before upgrading any part of a live system.
2409 * WARNING: Lustre networking configuration changes are required with
2410 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2413 * Support for kernels:
2414 2.6.9-22.0.2.EL (RHEL 4)
2415 2.6.5-7.244 (SLES 9)
2416 2.6.12.6 vanilla (kernel.org)
2419 Severity : enhancement
2420 Bugzilla : 7981/8208
2421 Description: Introduced Lustre Networking (LNET)
2422 Details : LNET is new networking infrastructure for Lustre, it includes
2423 a reorganized network configuration mode (see the user
2424 documentation for full details) as well as support for routing
2425 between different network fabrics. Lustre Networking Devices
2426 (LNDs) for the supported network fabrics have also been
2427 created for this new infrastructure.
2429 Severity : enhancement
2430 Description: Introduced Access control lists
2431 Details : clients can set ACLs on files and directories in order to have
2432 more fine-grained permissions than the standard Unix UGO+RWX.
2433 The MDS must be started with the "-o acl" mount option.
2435 Severity : enhancement
2436 Description: Introduced filesystem quotas
2437 Details : Administrators may now establish per-user quotas on the
2440 Severity : enhancement
2442 Description: Configuration change for the XT3
2443 The PTLLND is now used to run Lustre over Portals on the XT3
2444 The configure option(s) --with-cray-portals are no longer used.
2445 Rather --with-portals=<path-to-portals-includes> is used to
2446 enable building on the XT3. In addition to enable XT3 specific
2447 features the option --enable-cray-xt3 must be used.
2452 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2453 Details : A race between allocating a new llcd and re-getting the llcd_lock
2454 allowed another thread to grab newly-allocated llcd.
2456 Severity : enhancement
2458 Description: 2.6 OST async journal commit and locking fix to improve performance
2459 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2460 2.6 kernels are now async as they already were in 2.4 kernels so
2461 that they can commit concurrently with the network bulk transfer.
2462 For block-allocated files the filter allocation semaphore is held
2463 to avoid filesystem fragmentation during allocation. BKL lock
2464 removed for 2.6 xattr operations where it is no longer needed.
2469 Description: lconf incorrectly determined whether two IP networks could talk
2470 Details : In some more complicated routing and multiple-network
2471 configurations, lconf will avoid trying to make a network
2472 connection to a disjoint part of the IP space. It was doing the
2473 math incorrectly for one set of cases.
2478 Description: Fix for potential infinite loop processing records in an llog.
2479 Details : If an llog record is corrupted/zeroed, it is possible to loop
2480 forever in llog_process(). Validate the llog record length
2481 and skip the remainder of the block on error.
2484 Frequency : occasional (liblustre only)
2486 Description: liblustre could not open files whose last component is a symlink
2487 Details : sysio_path_walk() would incorrectly pass the open intent to
2488 intermediate path components.
2491 Frequency : rare (liblustre only with non-standard tuning)
2492 Bugzilla : 7201 (7350)
2493 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2494 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2495 after it has been released, only do this for referenced locks
2497 Severity : enhancement
2499 Description: specifying an (invalid) directory default stripe_size of -1
2500 would reset the directory default striping
2501 Details : stripe_size -1 was used internally to signal directory stripe
2502 removal, now use "all default" to signal dir stripe removal
2503 as a directory striping of "all default" is not useful
2506 Frequency : common for large clusters running liblustre clients
2508 Description: doing an ls when liblustre clients are running is slow
2509 Details : sending a glimpse AST to a liblustre client waits for every AST
2510 to time out, as liblustre clients will not respond. Since they
2511 cannot cache data we refresh the OST lock LVB from disk instead.
2513 Severity : enhancement
2515 Description: doing an ls at the same time as file IO can be slow
2516 Details : enqueue and other "small" requests can be blocked behind many
2517 large IO requests. Create a new OST IO portal for non-IO
2518 requests so they can be processed faster.
2521 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2523 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2524 created file with mode 000
2525 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2526 allow an ftruncate() as a non-root user to a file with mode 000.
2527 Lustre now respects this flag to disable mode checks when
2528 truncating a file owned by the user
2531 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2534 Description: Revoking locks from clients that went dead or catatonic might take
2536 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2537 cancellation of such locks instant on servers without waiting for
2538 any reply from clients. Clients drops these locks when cancel
2539 notification from server is received without replying.
2542 Frequency : liblustre-only, when liblustre client dies or becomes busy
2544 Description: Doing ls on Linux clients can take a long time with active
2546 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2547 granting such locks to it in the first place if possible. Locks
2548 are taken by proxy on the OST during the read or write and
2549 dropped immediately afterward. Add connect flags handling, do
2550 not grant locks to liblustre clients for glimpse ASTs.
2552 Severity : enhancement
2554 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2555 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2556 sequential reads and behaves badly with applications doing random
2557 reads. Improve it by reading ahead at least read region, and
2558 avoiding excessive large RPC for small reads.
2560 Severity : enhancement
2562 Description: Creating more than 1000 files for a single job may cause a load
2563 imbalance on the OSTs if there are also a large number of OSTs.
2564 Details : qos_prep_create() uses an OST index reseed value that is an
2565 even multiple of the number of available OSTs so that if the
2566 reseed happens in the middle of the object allocation it will
2567 still utilize the OSTs as uniformly as possible.
2572 Description: OST or MDS may oops in ping_evictor_main()
2573 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2574 but doesn't restart at beginning of obd_exports_timed list
2577 Severity : enhancement
2579 Description: improve by-nid export eviction on the MDS and OST
2580 Details : allow multiple exports with the same NID to be evicted at one
2581 time without re-searching the exports list.
2584 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2586 Description: MDS may oops in groups_free()
2587 Details : in rare race conditions a newly allocated group_info struct is
2588 freed again, and this can be NULL. The 2.4 compatibility code
2589 for groups_free() checked for a NULL pointer, but 2.6 did not.
2592 Frequency : common for liblustre clients doing little filesystem IO
2593 Bugzilla : 9352, 7313
2594 Description: server may evict liblustre clients accessing contended locks
2595 Details : if a client is granted a lock or receives a completion AST
2596 with a blocking AST already set it would not reply to the AST
2597 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2600 Frequency : lfs setstripe, only systems with more than 160 OSTs
2602 Description: unable to set striping with a starting offset beyond OST 160
2603 Details : llapi_create_file() incorrectly limited the starting stripe
2604 index to the maximum single-file stripe count.
2607 Frequency : LDAP users only
2609 Description: lconf did not handle in-kernel recovery with LDAP properly
2610 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2612 Severity : enhancement
2614 Description: bind OST threads to NUMA nodes to improve performance
2615 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2616 node and do their allocations there to localize memory access
2618 Severity : enhancement
2620 Description: llmount can determine client NID directly from Myrinet (GM)
2621 Details : the client NID code from gmnalnid was moved directly into
2622 llmount, removing the need to use this or specifying the
2623 client NID explicitly when mounting GM clients with zeroconf
2626 Frequency : if client is started with down MDS
2628 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2629 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2630 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2631 ptlrpc_import_delayed_req() because we didn't check if the
2632 request was interrupted, and we also didn't break out of the
2633 event loop if there was no timeout
2638 Description: data loss during non-page-aligned writes to a single file from
2639 both multiple nodes and multiple threads on one node at same time
2640 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2641 inconsistency led to false short-reads, that were cached and later
2642 used by ->prepare_write() to fill in partially written page,
2643 leading to data loss.
2646 Frequency : always, if lconf --abort_recovery used
2648 Description: lconf --abort_recovery fails with 'Operation not supported'
2649 Details : lconf was attempting to abort recovery on the MDT device and not
2652 Severity : enhancement
2654 Description: remove cleanup logs
2655 Details : replace lconf-generated cleanup logs with lustre internal
2656 cleanup routines. Eliminates the need for client-cleanup and
2659 Severity : enhancement
2661 Description: add support for EAs (user and system) on lustre filesystems
2662 Details : it is now possible to store extended attributes in the Lustre
2663 client filesystem, and with the user_xattr mount option it
2664 is possible to allow users to store EAs on their files also
2666 Severity : enhancement
2668 Description: Add possibility (config option) to show minimal available OST free
2670 Details : When compiled with --enable-mindf configure option, statfs(2)
2671 (and so, df) will return least minimal free space available from
2672 all OSTs as amount of free space on FS, instead of summary of
2673 free spaces of all OSTs.
2675 Severity : enhancement
2677 Description: do not expand extent locks acquired on OST-side
2678 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2679 by server: they are not cached anyway.
2682 Frequency : when mmap is used/binaries executed from Lustre
2684 Description: Unmmap pages before throwing them away from read cache.
2685 Details : llap_shrink cache now attempts to unmap pages before discarding
2686 them (if unmapping failed - do not discard). SLES9 kernel has
2687 extra checks that trigger if this unmapping is not done first.
2692 Description: lconf didn't resolve symlinks before checking to see whether a
2693 given mountpoint was already in use
2696 Frequency : when migrating failover services
2697 Bugzilla : 6395, 9514
2698 Description: When migrating a subset of services from a node (e.g. failback
2699 from a failover service node) the remaining services would
2700 time out and evict clients.
2701 Details : lconf --force (implied by --failover) sets the global obd_timeout
2702 to 5 seconds in order to quickly disconnect, but this caused
2703 other RPCs to time out too quickly. Do not change the global
2704 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2706 Severity : enhancement
2707 Frequency : if MDS is started with down OST
2708 Bugzilla : 9439,5706
2709 Description: Allow startup/shutdown of an MDS without depending on the
2710 availability of the OSTs.
2711 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2712 Add appropriate locking and lov-osc refcounts for safe
2713 cleaning. Add osc abort_inflight calls in case the
2714 synchronize never started.
2717 Frequency : occasional (Cray XT3 only)
2719 Description: root not authorized to access files in CRAY_PORTALS environment
2720 Details : The client process capabilities were not honoured on the MDS in
2721 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2722 been accessed by an authorized user then root was able to access
2723 the file on the local client also. The root user capabilities
2724 are now allowed on the MDS, as this environment has secure UID.
2727 Frequency : occasional
2729 Description: ldiskfs "too long searching" message happens too often
2730 Details : A debugging message (otherwise harmless) prints too often on
2731 the OST console. This has been reduced to only happen when
2732 there are fragmentation problems on the filesystem.
2737 Description: Division by zero in statfs when all OSCs are inactive
2738 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2739 lov_statfs divides by value returned by lov_get_stripecnt().
2743 Bugzilla : 9489, 3273
2744 Description: First write from each client to each OST was only 4kB in size,
2745 to initialize client writeback cache, which caused sub-optimal
2746 RPCs and poor layout on disk for the first writen file.
2747 Details : Clients now request an initial cache grant at (re)connect time
2748 and so that they can start streaming writes to the cache right
2749 away and always do full-sized RPCs if there is enough data.
2750 If the OST is rebooted the client also re-establishes its grant
2751 so that client cached writes will be honoured under the grant.
2756 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2757 Details : Now I/O RPCs go to different portal number and (presumably) fast
2758 lock requests (and glimses) and other RPCs get their own service
2759 threads pool that should be able to service those RPCs
2762 Severity : enhancement
2764 Description: Ability to exchange lustre version between client and servers and
2765 issue warnings at client side if client is too old. Also for
2766 liblustre clients there is ability to refuse connection of too old
2768 Details : New 'version' field is added to connect data structure that is
2769 filled with version info. That info is later checked by server and
2773 Frequency : rare, liblustre only.
2774 Bugzilla : 9296, 9581
2775 Description: Two simultaneous writes from liblustre at offset within same page
2776 might proceed at the same time overwriting eachother with stale
2778 Details : I/O lock withing llu_file_prwv was released too early, before data
2779 actually was hitting the wire. Extended lock-holding time until
2780 server acknowledges receiving data.
2783 Frequency : extremely rare. Never observed in practice.
2785 Description: avoid generating lustre_handle cookie of 0.
2786 Details : class_handle_hash() generates handle cookies by incrementing
2787 global counter, and can hit 0 occasionaly (this is unlikely, but
2788 not impossible, because initial value of cookie counter is
2789 selected randonly). Value of 0 is used as a sentinel meaning
2790 "unassigned handle" --- avoid it. Also coalesce two critical
2791 sections in this function into one.
2793 Severity : enhancement
2795 Description: allow liblustre clients to delegate truncate locking to OST
2796 Details : To avoid overhead of locking, liblustre client instructs OST to
2797 take extent lock in ost_punch() on client's behalf. New connection
2798 flag is added to handle backward compatibility.
2800 Severity : enhancement
2801 Bugzilla : 4928, 7341, 9758
2802 Description: allow number of OST service threads to be specified
2803 Details : a module parameter allows the number of OST service threads
2804 to be specified via "options ost ost_num_threads={N}" in the
2805 OSS's /etc/modules.conf or /etc/modprobe.conf.
2809 Bugzilla : 6146, 9635, 9895
2810 Description: servers crash with bad pointer in target_handle_connect()
2811 Details : In rare cases when a client is reconnecting it was possible that
2812 the connection request was the last reference for that export.
2813 We would temporarily drop the export reference and get a new
2814 one, but this may have been the last reference and the export
2815 was just destroyed. Get new reference before dropping old one.
2817 Severity : enhancement
2818 Frequency : if client is started with failover MDS
2820 Description: Allow multiple MDS hostnames in the mount command
2821 Details : Try to read the configuration from all specified MDS
2822 hostnames during a client mount in case the "primary"
2825 Severity : enhancement
2827 Description: Stop sending data to evicted clients as soon as possible.
2828 Details : Check if the client we are about to send or are sending data to
2829 was evicted already. (Check is done every second of waiting,
2830 for which l_wait_event interface was extended to allow checking
2831 of exit condition at specified intervals).
2834 Frequency : rare, normally only when NFS exporting is done from client
2836 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2838 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2841 Frequency : always, if srand() is called before liblustre initialization
2843 Description: Liblustre uses system PRNG disturbing its usage by user application
2844 Details : Introduce internal to lustre fast and high-quality PRNG for
2845 lustre usage and make liblustre and some other places in generic
2846 lustre code to use it.
2848 Severity : enhancement
2849 Bugzilla : 9477, 9557, 9870
2850 Description: Verify that the MDS configuration logs are updated when xml is
2851 Details : Check if the .xml configuration logs are newer than the config
2852 logs stored on the MDS and report an error if this is the case.
2853 Request --write-conf, or allow starting with --old_conf.
2855 Severity : enhancement
2857 Description: Handle symlinks in the path when checking if Lustre is mounted.
2858 Details : Resolve intermediate symlinks when checking if a client has
2859 mounted a filesystem to avoid duplicate client mounts.
2864 Description: lconf can hit an error exception but still return success.
2865 Details : The lconf command catches the Command error exception at the top
2866 level script context and will exit with the associated exit
2867 status, but doesn't ensure that this exit status is non-zero.
2872 Description: failure of ptlrpc thread startup can cause oops
2873 Details : Starting a ptlrpc service thread can fail if there are a large
2874 number of threads or the server memory is very fragmented.
2875 Handle this without oopsing.
2878 Frequency : always, only if liblustre and non-default acceptor port was used
2880 Description: liblustre cannot connect to servers with non-default acceptor port
2881 Details : tcpnal_set_default_params() was not called and was therefore
2882 ignoring the environment varaible TCPNAL_PORT, as well as other
2883 TCPNAL_ environment variables
2888 Description: two objects could be created on the same OST for a single file
2889 Details : If an OST is down, in some cases it was possible to create two
2890 objects on a single OST for a single file. No problems other
2891 than potential performance impact and spurious error messages.
2895 Bugzilla : 5681, 9562
2896 Description: Client may oops in ll_unhash_aliases
2897 Details : Client dcache may become inconsistent in race condition.
2898 In some cases "getcwd" can fail if the current directory is
2904 Description: Inode refcounting problems in NFS export code
2905 Details : link_raw functions used to call d_instantiate without obtaining
2906 extra inode reference first.
2910 Bugzilla : 9942, 9903
2911 Description: Referencing freed requests leading to crash, memleaks with NFS.
2912 Details : We used to require that call to ll_revalidate_it was always
2913 followed by ll_lookup_it. Also with revalidate_special() it is
2914 possible to call ll_revalidate_it() twice for the same dentry
2915 even if first occurence returned success. This fix changes semantic
2916 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2917 reference on a request referred from the intent.
2918 ll_intent_release() then releases such a request.
2921 Frequency : rare, normally benchmark loads only
2923 Description: unlinked inodes were kept in memory on the client
2924 Details : If a client is repeatedly creating and unlinking files it
2925 can accumulate a lot of stale inodes in the inode slab cache.
2926 If there is no other client load running this can cause the
2927 client node to run out of memory. Instead flush old inodes
2928 from client cache that have the same inode number as a new inode.
2931 Frequency : SLES9 2.6.5 kernel and long filenames only
2932 Bugzilla : 9969, 10379
2933 Description: utime reports stale NFS file handle
2934 Details : SLES9 uses out-of-dentry names in some cases, which confused
2935 the lustre dentry revalidation. Change it to always use the
2939 Frequency : rare, unless heavy write-truncate concurrency is continuous
2940 Bugzilla : 4180, 6984, 7171, 9963, 9331
2941 Description: OST becomes very slow and/or deadlocked during object unlink
2942 Details : filter_destroy() was holding onto the parent directory lock
2943 while truncating+unlinking objects. For very large objects this
2944 may block other threads for a long time and slow overall OST
2945 responsiveness. It may also be possible to get a lock ordering
2946 deadlock in this case, or run out of journal credits because of
2947 the combined truncate+unlink. Solution is to do object truncate
2948 first in one transaction without parent lock, and then do the
2949 final unlink in a new transaction with the parent lock. This
2950 reduces the lock hold time dramatically.
2953 Frequency : rare, 2.4 kernels only
2955 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2956 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2957 before unmounting their filesystem in order to avoid deadlock.
2958 The kernel_locked() function in 2.4 kernels only checks whether
2959 the kernel lock is held, not whether it is this process that is
2960 holding it as 2.6 kernels do.
2965 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2966 Details : The client ptlrpc code may be trying to reconnect to a down
2967 server before a previous connection attempt has timed out.
2968 Increase the reconnect interval to be longer than the connection
2969 timeout interval to avoid sending duplicate connections to
2973 Frequency : echo_client brw_test command
2975 Description: fix echo_client to work with OST preallocated code
2976 Details : OST preallocation code (5137) didn't take echo_client IO path
2977 into account: echo_client calls filter methods outside of any
2978 OST thread and, hence, there is no per-thread preallocated
2979 pages and buffers to use. Solution: hijack pga pages for IO. As
2980 a byproduct, this avoids unnecessary data copying.
2984 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2985 Description: Client can oops in mdc_commit_close() after open replay
2986 Details : It was possible for the MDS to return an open request with no
2987 transaction number in mds_finish_transno() if the client was
2988 evicted, but without actually returning an error. Clients
2989 would later try to replay that open and may trip an assertion
2990 Simplify the client close codepath, and always return an error
2991 from the MDS in case the open is not successful.
2994 Frequency : rare, 2.6 OSTs only
2996 Description: OST may deadlock under high load on fragmented files
2997 Details : If there was a heavy load and highly-fragmented OST filesystems
2998 it was possible to have all the OST threads deadlock waiting on
2999 allocation of biovecs, because the biovecs were not released
3000 until the entire RPC IO was completed. Instead, release biovecs
3001 as soon as they are complete to ensure forward IO progress.
3003 Severity : enhancement
3005 Description: Support for specifying external journal device at mount
3006 Details : If an OST or MDS device is formatted with an external journal
3007 device, this device major/minor is stored in the ext3 superblock
3008 and may not be valid for failover. Allow detecting and
3009 specifying the external journal at mount time.
3014 Description: Mounting an MDS with pending unlinked files may cause oops
3015 Details : target_finish_recovery() calls mds_postrecov() which returned
3016 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3017 considers this an error and immediately begins cleaning up the
3018 lov, just after starting the mds_lov process
3020 Severity : enhancement
3022 Description: Implement 'lfs df' to report actual free space on per-OST basis
3023 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3024 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3025 usage of objects; '-h' to report in human readable format.
3027 ------------------------------------------------------------------------------
3029 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3036 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3037 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3038 at startup during orphan unlinking. Instead, skip buddy update
3039 before setup, as it will be regenerated after recovery is complete.
3044 Description: If an OST is inactive, its locks might reference stale inodes.
3045 Details : lov_change_cbdata() must iterate over all namespaces, even if
3046 they are inactive to clear inode references from the lock.
3048 Severity : enhancement
3049 Frequency : occasional, if non-standard max_dirty_mb used
3051 Description: Client will block write RPCs if not enough grant
3052 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3053 then the client will block writes while waiting for another RPC
3054 to complete instead of consuming its dirty limit. With change
3055 we get improved performance when max_dirty_mb is small.
3057 Severity : enhancement
3058 Bugzilla : 3389, 6253
3059 Description: Add support for supplementary groups on the MDS.
3060 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3061 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3062 lookups for user supplementary groups into a cache.
3066 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3067 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3068 for liblustre O_EXCL is enforced by MDS.
3072 Description: readdir never returns NULL in liblustre.
3073 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3074 dentry in struct dirent.
3078 Frequency : liblustre only, depends on application IO pattern
3079 Description: liblustre clients evicted if not contacting servers
3080 Details : Don't put liblustre clients into the ping_evictor list, so
3081 they will not be evicted by the pinger ever.
3083 Severity : enhancement
3085 Description: Add ability to evict clients by NID from MDS.
3086 Details : By echoing "nid:$NID" string into
3087 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3088 $NID would be instantly evicted from this MDS and from all active
3089 OSTs connected to it.
3093 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3094 Details : lookup_it_finish() used to query file size from OSTs that was not
3099 Description: service threads change working directory to that of init
3100 Details : Starting lustre service threads may pin the working directory
3101 of the parent thread, making that filesystem busy. Threads
3102 now change to the working directory of init to avoid this.
3106 Frequency : during shutdown only
3107 Description: shutdown with a failed MDS or OST can cause unmount to hang
3108 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3113 Frequency : 2.6 only
3114 Description: chmod/chown may include an extra supplementary group
3115 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3116 supplementary group and if none is specified this is used.
3119 Bugzilla : 5479 (6816)
3121 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3122 Details : If lookup is in progress on a file that is unlinked we might try
3123 to revalidate the inode and fail in revalidate after lookup is
3124 complete and ll_file_open() enqueues the open again but
3125 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3128 Frequency : always, if lconf --abort_recovery used
3130 Description: lconf --abort_recovery fails with 'Operation not supported'
3131 Details : lconf was attempting to abort recovery on the MDT device and not
3134 ------------------------------------------------------------------------------
3136 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3141 Frequency : rare (only unsupported configurations with a node running as an
3143 Bugzilla : 6514, 5137
3144 Description: Mounting a Lustre file system on a node running as an OST could
3146 Details : OSTs now preallocates memory needed to write out data at
3147 startup, instead of when needed, to avoid having to
3148 allocate memory in possibly low memory situations.
3149 Specifically, if the file system is mounted on on OST,
3150 memory pressure could force it to try to write out data,
3151 which it needed to allocate memory to do. Due to the low
3152 memory, it would be unable to do so and the node would
3153 become unresponsive.
3155 Severity : enhancement
3157 Description: Addition of lconf --service command line option
3158 Details : lconf now accepts a '--service <arg>' option, which is
3159 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3161 Severity : enhancement
3163 Description: Failover mode is now the default for OSTs.
3164 Details : By default, OSTs will now run in failover mode. To return to
3165 the old behaviour, add '--failout' to the lmc line for OSTs.
3167 Severity : enhancement
3169 Description: Health checks are now provided for MDS and OSTs
3170 Details : Additional detailed health check information on MSD and OSTs
3171 is now provided through the procfs health_check value.
3174 Frequency : occasional, depends on IO load
3176 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3177 after numerous create/delete cycles
3178 Details : The ext3 inode allocation policy would not allocate new inodes
3179 very well on the OSTs because there are no new directories
3180 being created. Instead we look for groups with free space if
3181 the parent directories are nearly full.
3186 Description: Network or server problems during mount may cause partially
3187 mounted clients instead of returning an error.
3188 Details : The config llog parsing code may overwrite the error return
3189 code during mount error handling, returning success instead
3195 Description: MDS can fail to allocate large reply buffers
3196 Details : After long uptimes the MDS can fail to allocate large reply
3197 buffers (e.g. zconf client mount config records) due to memory
3198 fragmentation or consumption by the buffer cache. Preallocate
3199 some large reply buffers so that these replies can be sent even
3200 under memory pressure.
3204 Frequency : rare (liblustre)
3205 Description: fsx running with liblustre complained that using truncate() to
3206 extend the file doesn't work. This patch corrects that issue.
3207 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3208 ATTR_SIZE and lsm use in llu_setattr_raw.
3212 Frequency : rare, only 2.6 kernels
3213 Description: Unusual file access patterns on the MDS may result in inode
3214 data being lost in very rare circumstances.
3215 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3216 code in the 2.6 kernel caused the inode and/or EA data not to
3217 be read from disk, causing single-file corruption.
3221 Frequency : rare, only 2.6 filesystems using extents
3222 Description: Heavy concurrent write and delete load may cause data corruption.
3223 Details : It was possible under high-load situations to have an extent
3224 metadata block in the block device cache from a just-unlinked
3225 file overwrite a newly-allocated data block. We now unmap any
3226 metadata buffers that alias just-allocated data blocks.
3230 Frequency : filesystems with default stripe_count larger than 77
3231 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3232 Details : lconf specifies an inode size of 4096 bytes when the default
3233 stripe_count is larger than 77. This conflicts with the default
3234 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3235 this case to avoid pinning too much memory for large EAs.
3237 ------------------------------------------------------------------------------
3239 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3244 Frequency : rare (extremely heavy IO load with hundreds of clients)
3246 Description: Client is evicted, gets IO error writing to file
3247 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3248 caused clients to be evicted for AST timeouts. The fixes in
3249 bug 5192 mean we no longer need to have such short AST timeouts
3250 so ldlm_timeout has been increased.
3253 Frequency : occasional during --force or --failover shutdown under load
3254 Bugzilla : 5949, 4834
3255 Description: Server oops/LBUG if stopped with --force or --failover under load
3256 Details : a collection of import/export refcount and cleanup ordering
3257 issues fixed for safer force cleanup
3260 Frequency : only filesystems larger than 120 OSTs
3261 Bugzilla : 5990, 6223
3262 Description: lfs getstripe would oops on a very large filesystem
3263 Details : lov_getconfig used kfree on vmalloc'd memory
3266 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3267 Bugzilla : 6242, 6243
3268 Description: reading from files that had been truncated to a non-zero size
3269 but never opened returned no data
3270 Details : ll_file_read() reads zeros from no-object files to EOF
3275 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3276 Details : There's little harm in aborting MDS/OSS recovery and letting it
3277 try again, so I removed the LASSERT and return an error instead.
3279 Severity : enhancement
3281 Description: New debugging infrastructure for tracking down data corruption
3282 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3283 (b) cover more of the client-side code path, and (c) try to narrow
3284 down where problems occurred
3288 Bugzilla : 3819, 4364, 4397, 6313
3289 Description: Racing close and eviction MDS could cause assertion in mds_close
3290 Details : It was possible to get multiple mfd references during close and
3291 client eviction, leading to one thread referencing a freed mfd.
3293 Severity: : enhancement
3294 Bugzilla : 3262, 6359
3295 Description: Attempts to reconnect to servers are now more aggressive.
3296 Details : This builds on the enhanced upcall-less recovery that was added
3297 in 1.4.2. When trying to reconnect to servers, clients will
3298 now try each server in the failover group every 10 seconds. By
3299 default, clients would previously try one server every 25 seconds.
3304 Description: After recovery, certain operations trigger a failed
3305 assertion on a client.
3306 Details : Failing over an mds, using lconf -d --failover, while a
3307 client was doing a readdir() call would cause the client to
3308 LBUG after recovery completed and the readdir() was resent.
3310 Severity : enhancement
3312 Description: Default groups are now added by lconf
3313 Details : You can now run lconf --group <servicename> without having to
3314 manually add groups with lmc.
3317 Frequency : occasional
3319 Description: Nodes with an elan id of 0 trigger a failed assertion
3322 Frequency : always when accessing e.g. tty/console device nodes
3324 Description: tty and some other devices nodes cannot be used on lustre
3325 Details : file's private_data field is used by device data and lustre
3326 values in there got lost. New field was added to struct file to
3327 store fs-specific private data.
3330 Frequency : when exporting Lustre via NFS
3332 Description: NFSD failed occasionally when looking up a path component
3333 Details : NFSD is looking up ".." which was broken in ext3 directories
3334 that had grown large enough to become hashed.
3337 Frequency : Clusters with multiple interfaces not on the same subnet
3339 Description: Nodes will repeatedly try to reconnect to an interface which it
3340 cannot reach and report an error to the log.
3341 Details : Extra peer list entries will be created by lconf with some peers
3342 unreachable. lconf now validates the peer before adding it.
3345 Frequency : Only if a default stripe is set on the filesystem root.
3347 Description: Setting a default stripe on the filesystem root prevented the
3348 filesystem from being remounted.
3349 Details : The client was sending extra request flags in the root getattr
3350 request and did not allocate a reply buffer for the dir EA.
3353 Frequency : occasional, higher if lots of files are accessed by one client
3354 Bugzilla : 6159, 6097
3355 Description: Client trips assertion regarding lsm mismatch/magic
3356 Details : While revalidating inodes the VFS looks up inodes with ifind()
3357 and in rare cases can find an inode that is being freed.
3358 The ll_test_inode() code will free the lsm during ifind()
3359 when it finds an existing inode and then the VFS later attaches
3360 this free lsm to a new inode.
3364 Bugzilla : 6422, 7030
3365 Description: MDS deadlock between mkdir and client eviction
3366 Details : Creating a new file via mkdir or mknod (starting a transaction
3367 and getting the ns lock) can deadlock with client eviction
3368 (gets ns lock and trying to finish a synchronous transaction).
3371 Frequency : occasional
3372 Description: While starting a server, the fsfilt_ext3 module could not be
3374 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3375 kernels. Previously, lconf would still use the ext3 name
3376 when trying to load modules. Now, it will correctly use
3377 ext3 on 2.4 and ldiskfs on 2.6.
3379 Severity : enhancement
3380 Description: The default stripe count has been changed to 1
3381 Details : The interpretation of the default stripe count (0, to lfs
3382 or lmc) has been changed to mean striping across a single
3383 OST, rather than all available. For general usage we have
3384 found a stripe count of 1 or 2 works best.
3386 Severity : enhancement
3387 Description: Add support for compiling against Cray portals.
3388 Details : Conditional compiling for some areas that are different
3392 Frequency : occasional
3393 Bugzilla : 6409, 6834
3394 Description: Creating files with an explicit stripe count may lead to
3395 a failed assertion on the MDS
3396 Details : If some OSTs are full or unavailable, creating files may
3397 trigger a failed assertion on the MDS. Now, Lustre will
3398 try to use other servers or return an error to the
3402 Frequency : occasional
3404 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3405 caused lock timeout during readahead (since 1.4.2).
3406 Details : Processes doing readahead might match a lock that hasn't been
3407 granted yet if there are overlapping and conflicting lock
3408 requests. The readahead process waits on ungranted lock
3409 (original lock is CBPENDING), while OST waits for that process
3410 to cancel CBPENDING read lock and eventually evicts client.
3412 Severity : enhancement
3414 Description: Initial enabling of flock support for clients
3415 Details : Implements fcntl advisory locking and file status functions.
3416 This feature is provided as an optional mount flag (default
3417 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3418 locking are implemented yet, and those that are are not guaranteed
3419 to be completely correct in production environments.
3420 mount -t lustre -o [flock|noflock] ...
3423 Frequency : occasional
3425 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3426 assertion in the ext3 JBD (journaling) layer.
3427 Details : The b_committed_data struct is protected by the big kernel lock
3428 in 2.4 kernels, serializing journal_commit_transaction() and
3429 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3430 access to this struct. In 2.6 kernels there is finer grained
3431 locking to improve SMP performance of the JBD layer.
3435 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3440 Description: merge and cleanup kernel patches.
3443 -----------------------------------------------------------------------------
3445 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3447 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3448 versions of Lustre. Please update all servers and clients to
3449 version 1.4.2 or later at the same time. You must also run
3450 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3451 to update the configuration logs.
3453 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3454 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3455 - swab llog records of type '0' so we get proper header size/idx (5861)
3456 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3457 - fix rename of one directory over another leaking an inode (5953)
3458 - avoid SetPageDirty on 2.6 (5981)
3459 - don't re-add just-being-destroyed locks to the waiting list (5653)
3460 - when creating new directories, inherit the parent's custom
3461 striping settings if present parent (3048)
3462 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3463 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3464 - don't hold client locks on temporary worklist from l_lru (5666)
3465 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3466 - automatically evict dead clients (5921)
3467 - Update file size properly in create+truncate+fstat case (6196)
3468 - Do not unhash mountpoint dentries, do not allow removal of
3470 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3471 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3472 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3473 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3474 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3475 - Correctly handle reads of files with no objects (6243)
3476 - lctl recover will also mark a device active if deactivate used (5933)
3478 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3479 - allow --write-conf on an MDS with different nettype than client (5619)
3480 - don't write config llogs to MDS for mounts not from that MDS (5617)
3481 - lconf should create multiple TCP connections from a client (5201)
3482 - init scripts are now turned off by default; run chkconfig --on
3483 lustre and chkconfig --on lustrefs to use them
3484 - upcalls are no longer needed for clients to recover to failover
3486 - add --abort-recovery option to lconf to abort recovery on device
3488 - add support for an arbitrary number of OSTs (3026)
3489 - Quota support protocol changes.
3490 - forward compatibility changes to wire structs (6007)
3491 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3492 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3493 - improved "lustre status" script
3494 - initialize blocksize for non-regular files (6062)
3495 - added --disable-server and --disable-client configure options (5782)
3496 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3497 - Vanilla 2.4.29 support
3498 - increase maximum number of obd devices to 520 (6242)
3499 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3500 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3501 SLES 9 2.6.5 kernels
3502 - limit stripes per file to 160 (the maximum EA size) (6093)
3504 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3507 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3508 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3509 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3510 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3511 - small ext3 extents cleanups and fixes (5733)
3512 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3513 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3514 - avoid needless client->OST connect, fix handle mismatch (5317)
3515 - fix DLM error path that led to out-of-sync client, long delays (5779)
3516 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3517 - fix several locking issues related to i_size (5492,5624,5654,5672)
3518 - don't move pending lock onto export if it is already evicted (5683)
3519 - fix kernel oops when creating .foo in unlinked directory (5548)
3520 - fix deadlock in obdfilter statistics vs. object create (5811)
3521 - use time_{before,after} to avoid timer jiffies wrap (5882)
3522 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3523 - Do not leak request if server was not able to process it (5154)
3524 - If mds_open unable to find parent dir, make that negative lookup(5154)
3525 - don't create new directories with extent-mapping (5909, 5936)
3527 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3528 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3529 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3530 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3531 - support for RHEL4 kernel on i686 (5773)
3532 - provide error messages when incompatible logs are encountered (5898)
3534 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3535 * version 1.4.0.10 (1.4.1 release candidate 1)
3537 - don't keep a lock reference when lock is not granted (4238)
3538 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3539 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3540 - drop import inflight refcount on signal_completed_replay error (5255)
3541 - unlock page after async write error during send (3677)
3542 - handle missing objects in filter_preprw_read properly (5265)
3543 - no transno return for symlink open, don't save no-trasno open (3440)
3544 - don't try to complete elan receive that already failed (4012)
3545 - free RPC server reply state on error (5406)
3546 - clean up thread from ptlrpc_start_thread() on error (5160)
3547 - readahead could read extra page into cache that wasn't ejected (5388)
3548 - prevent races in class_attach/setup/cleanup/detach (5260)
3549 - don't dereference de->d_inode after l_dput of de (5458)
3550 - use "int" for stripe value returned from lock_to_stripe (5544)
3551 - mballoc allocation and error-checking fixes in 2.6 (5504)
3552 - block device patches to fix I/O request sizes in 2.6 (5482)
3553 - look up hostnames for IB nals (5602)
3554 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3555 - don't start multiple acceptors for the same port (5277)
3556 - fix incorrect LASSERT in mds_getattr_name (5635)
3557 - export a proc file for general "ping" checking (5628)
3558 - fix "lfs check" to not block when the MDS is down (5628)
3560 - service request history (4965)
3561 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3562 - create an "evict_client" /proc entry on OSTs, like the MDS has
3563 - fix mount usage message, return errors per mount(8) (5168)
3564 - change grep [] to grep "[]" in tests so they work in more UMLs
3565 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3566 - remove ancient LOV_MAGIC_V0 EA support (5047)
3567 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3568 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3569 - allow readahead allocations to fail when low on memory (5383)
3570 - mmap locking landed again, after considerable improvement (2828)
3571 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3573 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3576 - send OST transaction number in read/write reply to free req (4966)
3577 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3578 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3579 - the watchdog thread now runs as interruptible (5246)
3580 - flock/lockf fixes (but it's still disabled, pending 5135)
3581 - don't use EXT3 constants in llite code (5094)
3582 - memory shortage at startup could cause assertion (5176)
3584 - reorganization of lov code
3585 - single portals codebase
3587 - add extents/mballoc support (5025)
3588 - direct I/O reads in the obdfilter (4048)
3589 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3591 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3594 - send OST transaction number in read/write reply to free req (4966)
3595 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3596 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3597 - the watchdog thread now runs as interruptible (5246)
3598 - handle missing objects in filter_preprw_read properly (5265)
3599 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3600 - drop import inflight refcount on signal_completed_replay error (5255)
3601 - unlock page after async write error during send (3677)
3602 - return original error code on reconstructed replies (3761)
3603 - no transno return for symlink open, don't save no-trasno open (3440)
3605 - add pid to ldlm debugging output (4922)
3606 - bump the watchdog timeouts -- we can't handle 30sec yet
3607 - extra debugging for orphan dentry/inode bug (5259)
3609 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3612 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3613 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3614 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3615 - fix NULL dereference in /proc/sys/portals/routes (4827)
3616 - allow failed mdc_close() operations to be interrupted (4561)
3617 - stop precreate on OST before MDS would time out on it (4778)
3618 - don't send partial-page writes before EOF from client (4410)
3619 - discard client grant for sub-page writes on large-page clients (4520)
3620 - don't free dentries not owned by NFS code, check generation (4806)
3621 - fix lsm leak if mds_create_objects() fails (4801)
3622 - limit debug_daemon file size, always print CERROR messages (4789)
3623 - use transno after validating reply (3892)
3624 - process timed out requests if import state changes (3754)
3625 - update mtime on OST during writes, return in glimpse (4829)
3626 - add mkfsoptions to LDAP (4679)
3627 - use ->max_readahead method instead of zapping global ra (5039)
3628 - don't interrupt __l_wait_event() during strace
3630 - add software watchdogs to catch hung threads quickly (4941)
3631 - make lustrefs init script start after nfs is mounted
3632 - fix CWARN/ERROR duplication (4930)
3633 - return async write errors to application if possible (2248)
3634 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3635 - print NAL number in %x format (4645)
3636 - update barely-supported suse-2.4.21-171 series (4842)
3637 - support for sles 9 %post scripts
3638 - support for building 2.6 kernel-source packages
3639 - support for sles km_* packages
3641 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3644 - ignore -ENOENT errors in osc_destroy (3639)
3645 - notify osc create thread that OSC is being cleaned up (4600)
3646 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3647 - reconstruct ost_handle() like mds_handle() (4657)
3648 - create a new thread to do import eviction to avoid deadlock (3969)
3649 - let lconf resolve symlinked-to devices (4629)
3650 - don't unlink "objects" from directory with default EA (4554)
3651 - hold socknal file ref over connect in case target is down (4394)
3652 - allow more than 32000 subdirectories in a single directory (3244)
3653 - fix blocks count for O_DIRECT writes (3751)
3654 - OST returns ENOSPC from object create when no space left (4539)
3655 - don't send truncate RPC if file size isn't changing (4410)
3656 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3657 - bind to privileged port in socknal and tcpnal (3689)
3659 - rate limit CERROR/CWARN console message to avoid overload (4519)
3660 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3661 - basic mmap support (3918)
3662 - kernel patch series update from b1_4 (4711)
3664 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3667 - avoid crash during MDS cleanup with OST shut down (2775)
3668 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3669 - don't use bad inodes on the MDS (3744)
3670 - dynamic object preallocation to improve recovery speed (4236)
3671 - don't hold spinlock over lock dumping or change debug flags (4401)
3672 - don't zero obd_dev when it is force cleaned (3651)
3673 - print grants to console if they go negative (4431)
3674 - "lctl deactivate" will stop automatic recovery attempts (3406)
3675 - look for existing locks in ldlm_handle_enqueue() (3764)
3676 - don't resolve lock handle twice in recovery avoiding race (4401)
3677 - revalidate should check working dir is a directory (4134)
3679 - don't always mark "slow" obdfilter messages as errors (4418)
3681 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3684 - don't close LustreDB during write_conf until it is done (3860)
3685 - fix typo in lconf for_each_profile (3821)
3686 - allow dumping logs from multiple threads at one time (3820)
3687 - don't allow multiple threads in OSC recovery (3812)
3688 - fix debug_size parameters (3864)
3689 - fix mds_postrecov to initialize import for llog ctxt (3121)
3690 - replace config semaphore with spinlock (3306)
3691 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3692 - don't allow enqueue to complete on a destroyed export (3822)
3693 - down write_lock before checking llog header bitmap (3825)
3694 - recover from lock replay timeout (3764)
3695 - up llog sem before sending rpc (3652)
3696 - reduce ns lock hold times when setting kms (3267)
3697 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3698 - fix NULL deref and obd_dev leak on setup error (3312)
3699 - replace some LBUG about llog ops with error handling (3841)
3700 - don't match INVALID dentries from d_lookup and spin (3784)
3701 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3702 - fix invalid assertion in ptlrpc_set_wait (3880)
3704 - add libwrap support for the TCP acceptor (3996)
3705 - add /proc/sys/portals/routes for non-root route listing (3994)
3706 - allow setting MDS UUID in .xml (2580)
3707 - print the stack of a process that LBUGs (4228)
3709 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3712 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3713 - make sure to unset replay flag from failed open requests (3440)
3714 - if default stripe count is 0, use OST count for inode size (3636)
3715 - update parent mtime/ctime on client for create/unlink (2611)
3716 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3717 - free recovery state on server during a forced cleanup (3571)
3718 - unregister_reply for resent reqs (3063)
3719 - loop back devices mounting and status check on 2.6 (3563)
3720 - fix resource-creation race that can provoke i_size == 0 (3513)
3721 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3722 - more debugging for page-accounting assertion (3746)
3723 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3724 - avoid deadlock after precreation failure (3758)
3725 - fix race and lock order deadlock in orphan handling (3450, 3750)
3726 - add validity checks when grabbing inodes from l_ast_data (3599)
3728 - add /proc/.../recovery_status to obdfilter (3428)
3729 - lightweight CDEBUG infrastructure, debug daemon (3668)
3730 - change default OSC RPC parameters to be better on small clusters
3731 - turn off OST read cache for files smaller than 32MB
3732 - install man pages and include them in rpms (3100)
3733 - add new init script for (un)mounting lustre filesystems (2593)
3734 - run chkconfig in %post for init scripts (3701)
3735 - drop scimac NAL (unmaintained)
3737 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3740 - clean kiobufs before and after use (3485)
3741 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3742 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3743 - change init script to fail more clearly if not run as root (1528)
3744 - allow clients to reconnect during replay (1742)
3745 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3746 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3747 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3748 - fix cond_resched (3554)
3749 - only evict unfinished clients after recovery (3515)
3750 - allow bulk resend, prevent data loss (3570)
3751 - dynamic ptlrpc request buffer allocation (2102)
3752 - don't allow unlinking open directory if it isn't empty (2904)
3753 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3754 - remove extraneous obd dereference causing LASSERT failure (3334)
3755 - don't use get_cycles() when creating temp. files on the mds (3156)
3756 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3757 - handle EEXIST for set-stripe, set proper directory name (3336)
3759 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3760 - fix ksocknal_fmb_callback() error messages (2918)
3762 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3765 - don't copy lvb into (possibly NULL) reply on error (2983)
3766 - don't deref dentry after dput, don't free lvb on error (2922)
3767 - use the kms to determine writeback rpc length (2947)
3768 - increment oti_logcookies when osc is inactive (2948)
3769 - update client's i_blocks count via lvb messages (2543)
3770 - handle intent open/close of special files properly (1557)
3771 - mount MDS with errors=remount-ro, like obdfilter (2009)
3772 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3773 - don't use cancelling-locks' kms values (2947)
3774 - use highest lock extent for kms, not last one (2925)
3775 - don't dereference ERR_PTR() dentry in error handling path (3107)
3776 - fix thread race in portals_debug_dumplog() (3122)
3777 - create lprocfs device entries at setup instead of at attach (1519)
3778 - common AST error handler, don't evict client on completion race (3145)
3779 - zero nameidata in detach_mnt in 2.6 (3118)
3780 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3781 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3782 - initialize RPC timeout timer earlier for 2.6 (3219)
3783 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3784 - print nal/nid for unknown nid (3258)
3785 - additional checks for oscc recovery before doing precreate (3284)
3786 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3787 - don't crash in mdc_close for bad permissions on open (3285)
3788 - zero i_rdev for non-device files (3147)
3789 - clear page->private before handing to FS, better assertion (3119)
3790 - tune the read pipeline (3236)
3791 - fix incorrect decref of invalidated dentry (2350)
3792 - provide read-ahead stats and refine rpc in flight stats (3328)
3793 - don't hold journal transaction open across create RPC (3313)
3794 - update atime on MDS at close time (3265)
3795 - close LDAP connection when recovering to avoid server load (3315)
3796 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3797 - don't leak open file on MDS after open resend (3325)
3798 - serialize filter_precreate and filter_destroy_precreated (3329)
3799 - loop device shouldn't call sync_dev() for nul device (3092)
3800 - clear page cache after eviction (2766)
3801 - resynchronize MDS->OST in background (2824)
3802 - refuse to mount the same filesystem twice on same mountpoint (3394)
3803 - allow llmount to create routes for mounting behind routers (3320)
3804 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3805 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3806 - fix rare problem with rename on htree directories (3417)
3808 - allow default OST striping configuration per directory (1414)
3809 - fix compilation for qswnal for 2.6 kernels (3125)
3810 - increase maximum number of MDS request buffers for large systems
3811 - change liblustreapi to be useful for external progs like lfsck (3098)
3812 - increase local configuration timeout for slow disks (3353)
3813 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3815 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3818 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3819 - don't overwrite extent policy data in reply if lock was blocked (2901)
3820 - drop filter export grants atomically with removal from device (2663)
3821 - del obd_self_export from work_list in class_disconnect_exports (2908)
3822 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3823 - swab reply message in mdc_close, other PPC fixes (2464)
3824 - fix destroying of named logs (2325)
3825 - overwrite old logs when running lconf --write_conf (2264)
3826 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3827 - fix race in target_handle_connect (2898)
3828 - mds_reint_create() should take same inode create lock (2926)
3829 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3830 - don't close files for self_export to avoid uninitialized obd (2936)
3831 - allow MDS with the same name as client node (2939)
3832 - hold dentry reference for closed log files for unlink (2325)
3833 - reserve space for all logs during transactions (2059)
3834 - don't evict page beyond end of stripe extent (2925)
3835 - don't oops on a deleted current working directory (2399)
3836 - handle hard links to targets without a parent properly (2517)
3837 - don't dereference NULL lock when racing during eviction (2867)
3838 - don't grow lock extents when lots of conflicting locks (2919)
3840 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3843 - account for cache space usage on clients to avoid data loss (974)
3844 - lfsck support in lustre kernel code (2349)
3845 - reduce journal credits needed for BRW writes (2370)
3846 - orphan handling to avoid losing space on client/server crashes
3847 - ptlrpcd can be blocked, stopping ALL progress (2477)
3848 - use lock value blocks to assist in proper KMS, faster stat (1021)
3849 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3850 - recovery for initial connections (2355)
3851 - fixes for mds_cleanup_orphans (1934)
3852 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3853 - block all file creations until orphan recovery completes (1901)
3854 - client remove rq_connection from request struct (2423)
3855 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3856 - recovery timer race (2670)
3857 - mdc_close recovey bug (2532)
3858 - ptlrpc cleanup bug (2710)
3859 - mds timeout on local locks (2588)
3860 - namespace lock held during RPCs (2431)
3861 - handle interrupted sync write properly (2503)
3862 - don't try to handle a message that hasn't been replied to (2699)
3863 - client assert failure during cleanup after abort recovery (2701)
3864 - leak mdc device after failed mount (2712)
3865 - ptlrpc_check_set allows timedout requests to complete (2714)
3866 - wait for inflight reqs when ptlrpcd finishes (2710)
3867 - make sure unregistered services are removed from the srv_list
3868 - reset bulk XID's when resending them (caught by 1138 test)
3869 - unregister_bulk after timeout
3870 - fix lconf error (2694)
3871 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3872 - readahead locks pages, leaves pending causing memory pressure (2673)
3873 - increase OST request buffers to 4096 on large machines (2729)
3874 - fix up permission of existing directories in simple_mkdir (2661)
3875 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3876 - don't assign transno to errored transactions (2742)
3877 - don't delete objects on OST if given a bogus objid from MDS (2751)
3878 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3879 - if rq_no_resend, then timeout request after recovery (2432)
3880 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3881 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3882 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3883 - MDS recovery shouldn't delete objects that it has given out (2730)
3884 - if enqueue arrives after completion, don't clobber LVB (2819)
3885 - don't unlock pages twice when trigger_group_io returns error (2814)
3886 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3887 - don't write pages to disk if there was an error (1450)
3888 - don't ping imports that have recovery disabled (2676)
3889 - take buffered bytes into account when balancing socknal conn (2817)
3890 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3891 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3892 - remove little-endian swabbing of llog records (1987)
3893 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3894 - retry reposting request buffers if they fail (1191)
3895 - grow extent at grant time to avoid granting a revoked lock (2809)
3896 - lock revoke doesn't evict page if covered by a second lock (2765)
3897 - disable VM readahead to avoid reading outside lock extents (2805)
3899 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3900 - updated kernel patches for hp-2.4.20 kernel (2681)
3902 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3905 - fix truncated write corruption (2366)
3906 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3908 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3909 - ensure len(uuid) < 37 in lmc (1171)
3910 - fix ia64 OOPS in llog_test (2255)
3911 - zero end of page at obdfilter for partial page writes (2648)
3912 - don't leave stale dentries around after renames (bug 2428)
3913 - fix timeouts when evicting a client with a single lock held (2642)
3914 - set deadline for the initial HELLO message to drain (2634)
3915 - print out dotted-quad IP addresses in the socknal (2302)
3917 - additional debugging for MDS client eviction problem (2443)
3918 - fix mkfsoptions support for osts (2603, 2604)
3920 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3923 - add series for the vanilla 2.6.0 kernel
3924 - add series for the vanilla 2.4.24 kernel
3925 - add series for a cray x86/64 UL kernel drop
3926 - fix xattr patches for the vanilla 2.4.19 series
3928 - generate true UUIDs in lmc (1171)
3929 - have portals stack dumping break in UML (2466)
3930 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3931 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3932 - wake callback waiting threads on client eviction (2460)
3933 - Add --ptldebug and --subsystem to lmc (1719)
3934 - update assertion to allow safe interrupt allocation
3935 - set rq_no_resend for cancel requests (2432)
3936 - recalculate ptlrpcd timeout after resend (2494)
3937 - call vfs_rmdir when removing pending directories (2368)
3938 - fix renaming a file to itself (2429)
3939 - lmc creates a default one-stripe lov (2454)
3940 - expand procfs space to handle large clusters (2326)
3941 - increase UML stack to avoid overflow
3942 - update lconf's list of debug and subsystem masks
3943 - fix lfs find --obd (2510)
3944 - /proc tunable for disabling filter read caching (2591)
3945 - stop rpm packages from altering slapd.conf (2301)
3946 - disable nagle in the socknal under 0conf (2578)
3947 - choose mds inode size based on stripe count (2572)
3948 - fix kernel-source rpm problems (2516)
3950 - add --disable-doc to avoid pdf generation (2421)
3951 - update documentation, tests, type-os, comments
3952 - avoid format warnings on ia64
3953 - remove the TOE NAL
3954 - tiny code cleanups by removing unused fields
3956 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3959 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3960 - remove the most blatant lies from BUILDING file (2371)
3961 - change default debug level to reasonable production setting
3962 - reduce client side cache size to reduce cache flush time
3963 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3964 - make TCP zerocopy and pinger support enabled by default (2476)
3965 - sync writes completed after process exits caused crashes (2319)
3966 - maintain correct mount count on the MDS (2356)
3967 - backout 1557, because 2316 wasn't really fixed
3968 - better file I/O statistics gathering in /proc
3969 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3970 - another kernel patch to fix zero-copy TCP function export
3971 - don't take duplicate lock when processing re-sent getattr (2420)
3972 - lctl uses obd_self_export instead of creating new conn (2353)
3973 - MDS/OST recovery case which requires object creation asserted (2425)
3974 - move lfs from /usr/sbin to /usr/bin in packages
3975 - fix race between mds_client_add and mds_client_free (2417)
3976 - use kmalloc instead of slabs in portals (2430)
3977 - don't create duplicate records when a failover MDS is present (2442)
3978 - remove unnecessary mount age check (2332)
3979 - don't remove directory inodes from locks prematurely (2451)
3980 - don't break if MDS service name is the same as hostname (2103)
3981 - fix races in client write RPC generation when cache full (2482)
3983 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3986 - remove now-unused request->rq_obd (278)
3987 - if an allocation fails, print out how much memory we've used (1933)
3988 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3989 - add the "configurable stack size" patch to most series files (1256)
3990 - ability to write large log records, for 100+ OST configs (2306)
3991 - fix NULL deref when filter_prep fails (2314)
3992 - fix operator precedence error in filter_sync
3993 - dynamic allocation of socknal TX descriptors (2315)
3994 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3995 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3996 - ensure that CWARN messages in Portals always get to the syslog
3997 - __init/__exit are not for prototype decls (ldlm_init/exit)
3998 - x86-64 compile warning fixes
3999 - fix gateway LMC keyword conflict (2318)
4000 - fix MDS lock inversions in getattr/reint paths (1844)
4001 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4002 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4003 - DLM race condition prevented some lock evictions (2328)
4004 - ENOMEM detection and retry on socknal sends (2230)
4005 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4006 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4007 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4008 - better error messages when a client is rejected during recovery (1505)
4009 - avoid cancelling locks which were never granted, after failure (2330)
4010 - fix i_sem/journal inversion in mds_client_add (2333)
4011 - fix truncate/getattr lock cycle deadlock (2334)
4012 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4013 - fix two rare exit paths which could leak an l_lock() ref (2321)
4014 - fencepost error in MDS/OST orphan recovery (2226)
4015 - make log record alignment 8 bytes (1988)
4016 - lstripe now fails when requested offset > ost_count (2237)
4017 - ensure that all kernel series have a complete list.h (1607)
4018 - fix crashes in special-file operations (2316)
4019 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4021 - allow configurable automake binary, for testing new versions
4022 - small update to the lfs documentation
4024 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4026 * fix negative export reference count in fsfilt_sync (2312)
4028 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4029 * release candidate 0.9.1
4031 - orphans are moved into the PENDING directory for possible recovery
4032 - replayed opens now open by fid for orphan/rename safety (1042)
4033 - last close of an orphan inode generates a transno (683)
4034 - chdir() and mount() now pin the directory entry (1020)
4035 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4036 - discard very old requests without processing them (1502)
4037 - remove some common, well-understood CERRORs (1505)
4038 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4039 - clear "grant" flags in OST replies until OST grant code lands (1644)
4040 - fix read performance by not clobbering i_blksize on client (1598)
4041 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4042 - make LRU size a /proc tunable, clears locks when reduced (707)
4043 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4044 - server locks take references on exports now (1558)
4045 - build fixes for 2.4.20-rh trees (1663)
4046 - return an error from lov_create if all OSCs are inactive (1751)
4047 - fix import levels when a reconnect happens without a timeout (1597)
4048 - exit early from mds_open if we get a lookup error (1749)
4049 - partial page read at EOF wouldn't wait for disk before sending (1642)
4050 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4051 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4052 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4053 - write/truncate lock inversion (1639)
4054 - Don't auto-load obdclass, portals modules during cleanup (1495)
4055 - fix timestamps from jumping to "now" (1763)
4056 - extra journal assertions (1648)
4057 - add an extra multiunlink test (1771)
4058 - fix read_record/write_record API (1776)
4059 - fix leak of offset_extent, possible incorrect i_size later (1772)
4060 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4061 - Debugging for the kqswnal_get_idle_tx problems (1820)
4062 - Allow recovery to be attempted multiple times (1536)
4063 - Write out MDS last_rcvd file after it is first created (1600)
4064 - Fix tx_descriptor leak in failed transmit situations (1827)
4065 - ext3 journaling fixes for assertion failure after IO error (1871)
4066 - class_export_put() on freed export after completion AST error (1896)
4067 - Fix revalidate looping in VFS (1322)
4068 - Don't access a freed export during MDS_REINT timeout (1521)
4069 - Add open-unlink recovery support on the MDS (1673,1764)
4070 - Return an error if no MDS data was read from last_rcvd (1946)
4071 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4072 - Don't setup a disk device that doesn't match exported UUID (317)
4073 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4074 - avoid committing NULL handle in force close
4075 - local.sh is now a one-stripe LOV configuration
4076 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4077 - fix ext3 htree duplicate directory entry corruption (1516)
4078 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4079 - update to new LOV EA format (2097)
4080 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4082 2003-06-15 Phil Schwan <phil@clusterfs.com>
4085 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4086 - per-import/export recovery handling (958, 931, 959)
4087 - multiple last-rcvd slots, for serving multiple FSes (949)
4088 - connections are again shared between multiple imp/exports (963, 964)
4089 - "umount -f" would hang if any requests needed to be sent (393, 978)
4090 - avoid pinning large req buffer by copying for queued messages (989)
4091 - add "uuid" to "lctl device" command to help upcalls (991)
4092 - "open" RPCs with transnos would confuse recovery counters (1037)
4093 - do proper endian conversion of all wire messages (288, 340, 891)
4094 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4095 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4096 - fix ptlrpc_connection leak in target_handle_connect (1174)
4097 - fix import refcounting bug in OST and MDS cleanup (1134)
4098 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4099 - fix very unlikely obd_types race condition (501)
4100 - remove osc_open hack for echo_client (1187)
4101 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4102 - a failure in read_inode2 leads to deadlock (1139)
4103 - cancel ack-locks as soon as transaction is committed (1072)
4104 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4105 - make sure to commitrw after any preprw to avoid deadlock (1162)
4106 - failing to execute a file in a lustre FS would lock inode (1203)
4107 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4108 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4109 - obd_uuid2tgt would walk off the end of the list (1255)
4110 - on IA64 the osc would give portals incorrect bulk size (1258)
4111 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4112 - fix lock inversion caused by new llite matching code (1282)
4113 - limit the number of dirty pages on a client to 10MB (1286)
4114 - timed out locks were not being corrected cancelled (1289)
4115 - fix O_DIRECT above 4GB on IA-32 (1292)
4116 * major user-visible changes
4117 - fail out/fail over policy now controlled by the upcall (993)
4119 - add OBD_PING to check server availability and failure (954)
4120 - lustre messages are now sent in sending host order (288, 340, 891)
4121 - add eadatalen to MDS getattr reply (340)
4122 - OST read replies may contain second buffer, with per-page status (593)
4124 2003-03-11 Phil Schwan <phil@clusterfs.com>
4127 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4128 - failed open()s could cause deadlock; fixed (867, 869)
4129 - stop cancelling OST locks when files are closed (481)
4130 - overlapping XID spaces caused network corruption (851, 853)
4131 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4132 - setattr_raw added, to do single-RPC, server-side setattrs
4133 - lmc/lconf syntax change for OST UUIDs
4134 - fix crashy race condition between ptlrpc_free_req and osc_close
4135 - don't use request in mdc_enqueue if we hit a timeout (889)
4136 - don't set the inode i_size for regular files from the MDS (896)
4137 - handle out of order completion AST (842)
4138 - don't LBUG if a lock request times out after receiving AST (913)
4139 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4140 - if a bad lock AST arrives, send an error instead of dropping entirely
4141 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4142 - fix leak in bulk IO when only partially completed (899, 900, 926)
4143 - fix O_DIRECT for ia64 (55)
4144 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4145 - C-z after timeout could hang a process forever; fixed (977)
4147 - client-side I/O cache (678, 924, 929, 941, 970)
4149 - READPAGE and SETATTRs which don't take server-side locks get
4152 2003-02-11 Phil Schwan <phil@clusterfs.com>
4155 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4156 - Fix fsfilt_extN_readpage() to read a full page of directory
4157 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4158 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4159 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4160 - Most intent operations were removed, in favour of a new RPC mode
4161 that does a single RPC to the server and bypasses most of the VFS
4162 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4163 - Aggressively cancel local locks on DLM servers
4164 - mds_reint_unlink sends EA to the client if it's the last nlink.
4165 client uses that EA to unlink OST objects.
4166 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4167 - recursive symlinks were fixed (439)
4168 - fixed NULL deref in DEBUG_REQ
4169 - filter_update_lastobjid no longer calls sync, which annoyed extN
4170 - fixed multi-client small-writes to a single file problem (445)
4171 - fixed mtime updates during file writes (607)
4172 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4173 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4174 - fixed Linux OST size reporting problem (444, 656)
4175 - OST now updates object mtime with writes or setattr (607, 619)
4176 - client verifies file size before zeroing page past EOF (445)
4177 - OST now writes last allocated objid to disk with allocation (108)
4178 - LOV on echo now works (409)
4180 - mds_reint_unlink sends a new buffer, with the EA included. this
4181 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4182 set if a regular file was being unlinked, and it was the last link
4183 - use PtlGet from the target for bulk writes (315)
4184 - OST now updates object mtime with writes or setattr (607, 619)
4185 - LDLM now has a grant-time callback to revalidate locked items, if
4187 - Many MDS operations were reorganized to combat race conditions
4189 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4190 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4192 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4195 - Fully reactivate OST imports after reconnection (512, others)
4196 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4197 - More graceful error handling for truncating on dead OST (515)
4198 - Don't error out unless we're actually accessing dead stripes (474)
4199 - Fix garbage sizes when stripes are missing (410)
4200 - LRU counters were broken, causing constant lock purge (433, 432)
4201 - garbage on read from stripes with failed OSTs (441)
4202 - mark OSCs as active before reconnecting during recovery (438)
4203 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4204 - lfind did not preserve OST order in output (443)
4205 - symlinks cause hung clients, incorrect data (439)
4206 - stop dereferencing request after dropping refcount (457)
4207 - don't LASSERT(spin_is_locked) on non-SMP (455)
4208 - fixes for many rename() bugs
4209 - fstat didn't correctly synchronize attributes (399)
4210 - server must handle lock cancellation during blocking AST prep (487)
4211 - bulk descriptors were free()d too soon (511)
4212 - fix paths in lconf, which would load incorrect modules (451, 507)
4213 - fix confusing lconf 'host not found' error message (386)
4214 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4215 - fix race condition in mdc_blocking_ast() for inode access (526)
4216 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4217 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4218 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4219 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4220 - fix lstripe in several ways (kernel side) (527)
4221 - fix request leak in ldlm_cli_enqueue (262)
4222 - incorrect OSC was marked inactive after OST failure
4223 - call mds_fs_cleanup before unmounting filesystem (524)
4224 - fix races between taking ns_lock and ldlm_lock_change_resource
4225 - fix races updating LOV export open file list
4226 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4227 - fix recovery NULL deref in ldlm_cli_cancel_unused
4228 - fix some DLM races by using new hash table for lock handles (419)
4229 - permit the client to specify desired inodes, at replay
4230 - duplicate requests when we queue them for replay reintegration
4231 - fix last_rcvd offset calculation
4232 - sync after each recovered transaction, so we always make progress
4233 - never, not always, ERESTART requests without transnos
4234 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4236 - skip replay if the MDS says that the client is already connected
4237 - don't check for a recovery-enabled export to match lctl's UUID
4238 - don't INC_USE_COUNT for phantom exports
4239 - don't crash when cleaning up phantom exports (567)
4240 - don't double-finish or set replay data for errored mdc_open requests
4241 - abort requests when they time out, so we don't get old replies
4242 - send/receive replies for AST messages again
4243 - if the client says that it doesn't have the lock, cancel it on the
4245 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4246 mark it as destroyed, it will all work out when decref is called
4247 - fix module use counts (22, 581)
4249 - ASTs now expect a reply (server cancels lock on error reply)
4251 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4254 - fix many simultaneous client startup (392)
4255 - fix dentry->d_it clobbering
4256 - credentials weren't being shipped for readdir/getattr operations
4257 - remove invalid assertions triggered during some concurrent MD
4259 - proper Lustre versions added (336, 389)
4260 - fix memory leak for create error case (398)
4261 - fix LOV locking bug that would get cli/srv out of sync
4262 - fix echo client over LOV (409)
4263 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4264 - fix double-O_EXCL intent crash (424)
4265 - avoid sending multiple lock CANCELs (352)
4267 - MDS can do multi-client recovery (modulo bugs in new code)
4269 - many updates, edits, cleanups
4271 2002-11-18 Phil Schwan <phil@clusterfs.com>
4274 - fix null d_it dereference (346)
4275 - fix full OST/dbench hang (333)
4276 - fix permission problem with file removal (286)
4277 - fix removal of OSCs from LOV when they fail
4278 - fix NULL deref during bulk timeout (214)
4279 - fix problems related to multiple filesystems on one MDS (241)
4280 - fixed serious subtle metadata locking bugs
4281 - free locks on clients when inodes are removed due to memory
4283 - fix inode pointer in lock data (285)
4284 - partial support for multiple MDS on a single host (241)
4285 - data locks weren't cancelled at clear_inode time (290, 311)
4286 - intent locks could lead to unbounded lock growth (205)
4287 - added a maximum lock count, an LRU list, and a flusher
4288 - fix multiple rename (365)
4289 - properly abstracted the echo client
4290 - OSC locked 1 byte too many; fixed
4291 - rewrote brw callback code:
4292 - fixed recovery bugs related to LOVs (306)
4293 - fixed too-many-pages-in-one-write crash (191)
4294 - fixed (again) crash in sync_io_timeout (214)
4295 - probably fixed callback-related race (385)
4297 - Add capability to MDS protocol
4298 - LDLM cancellations and callbacks on different portals
4300 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4303 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4304 - defer open object destruction to close time (601981, 138)
4305 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4306 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4307 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4308 - don't call obd_setattr in ll_file_release for destroyed objects
4309 * protocol change to lustre_msg: move |version| and add |flags|
4310 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4311 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4312 * added replay of create, unlink, link and rename operations during
4313 MDS failover; recovery should be much more robust now
4314 * remove failed OSCs from LOVs (only lov_create uses this so far)
4315 * the lustre-HOWTO was brought (more) up to date (582544)
4317 2002-10-23 Phil Schwan <phil@clusterfs.com>
4320 - in-use dentries weren't being reused properly (617851)
4321 - prevent multiple LDLM setup (599178)
4322 - fix LOV size calculations for truncate (617853)
4323 - fix client handling of MDS intent errors (POSIX)
4324 - fix permission bug in lovstripe.c test (624321)
4325 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4326 - truncate past end of file could corrupt data
4327 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4328 - a race in recovery could return ETIMEDOUT to apps (623947)
4329 - building outside the source directory was fixed
4330 * the lustre-HOWTO was brought (more) up to date (582544)
4331 * major progress was made on recovery functionality
4333 2002-10-10 Phil Schwan <phil@clusterfs.com>
4336 - recovery deadlock fix
4337 - rm -rf causes LBUG fix (617817)
4338 - file open by multiple tasks fix (618962)
4339 - directory permissions bugs (602707 and 620007)
4340 - journal_stop fixed with locking (611313)
4341 - O_APPEND failures resolved (618273, perhaps 614459)
4342 - lconf PATH fix (619770)
4343 - IA64 build fix (621450)
4344 - RPC buffer sizes scale with amount of memory
4346 2002-10-01 Phil Schwan <phil@clusterfs.com>
4349 - locks would be cancelled without throwing away data pages,
4350 resulting in inconsistent data (605627)
4351 - inode attributes were not always being refreshed (605627, 612449)
4352 - lconf now continues to cleanup after lctl reports an error
4353 - MDS now enforces user permissions (602707)
4354 - lprocfs cleanup fixed, but not yet enabled (614157)
4355 - fixed infinite server hang, should a client not respond to an AST
4356 - avoid going into recovery if user calls readlink() with a buffer
4357 that's too small (613941)
4358 - AST RPCs no longer require replies (614867) -- this may be changed
4359 - don't crash server if client sends an IOV that's too big (611336)
4360 - fixed lock conversion deadlock (611892)
4361 - fixed the following of symlinks (614622)
4362 * recovery: the server can remove locks from a client that dies, other
4363 clients can make progress
4364 * more extN patch fixes
4365 * compile-time configurable ptlrpc buffer allocations
4367 - collaborative read cache document
4368 - Lustre Lite Performance CDR document-in-progress
4370 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4373 - fix typo in patch-2.4.18
4375 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4378 - clear ptlrpc request each time in handle_incoming_request()
4379 - unlink of files now destroys the object on the OST
4381 2002-09-19 Peter Braam <braam@clusterfs.com>
4383 * add hard link support
4384 * change obdfile creation method
4385 * kernel patch changed
4387 2002-09-19 Peter Braam <braam@clusterfs.com>
4390 - stack overflow bug in extN fixed
4392 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4394 * documentation updates
4395 - add man pages for config tools
4396 - update tests/README to describe testing with new config tools
4397 - finish metadata API descriptions
4398 * bug fixes and cleanups
4399 - statfs workaround for 16TB limit
4400 - LOV stripe allocation improved, can stripe on subset of OSTs
4401 - LOV file size/IO offset was wrong for files > 4GB in size
4402 - object EA data was being dropped, caused files to be unreadable
4403 - memory overflow with non-LOV OST caused memory corruption
4404 - fixed regression tests to work with new config tools, obdfilter
4405 - fixed bug when directory size became larger than 1 block
4406 - fixed bug (for single client case) when PWD was deleted
4407 - invalidate local directory pages when doing intent-based ops
4408 - avoid LDLM oops when lock callback contained bad data
4410 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4412 * documentation updates
4413 * bug fixes and cleanups
4414 - configuration tools
4417 - 64-bit compile warnings
4418 - 64-bit internal statfs data
4420 * test_brw on persistent OST devices
4422 * lprocfs (disabled)
4424 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4426 * documentation updates
4427 * bug fixes and cleanups
4428 * configuration tools
4430 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4433 * many small fixes to 0_5_4
4434 * io/network handling
4435 * thinkos in MDS operations
4437 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4440 * crucial basic fixes to 0.5.3
4441 * IOR, Iozone work over Elan
4444 2002-08-07 Phil Schwan <phil@clusterfs.com>
4445 * version 0_5_3, our first alpha
4446 * we use the new Portals iovs
4447 * documentation updates
4448 * bug fixes and cleanups
4449 * small changes in the DLM wire protocol
4451 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4452 * version 0_5_1 with some initial stability,
4453 * locking on MD and file I/O.
4454 * documentation updates
4455 * several bug fixes since 0.5.0
4456 * small changes in wire protocol
4458 2002-07-18 Phil Schwan <phil@clusterfs.com>
4460 * delivered as Lustre Light Alpha
4461 * fixed a crash after handling invalid MDS requests
4462 * fixed directory pages for architectures with non-4k pages sizes
4464 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4465 * release version v0_4_4
4466 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4468 * Fixes a number of interruption problems with OST operations.
4469 * Update documentation for portals header changes
4470 * Move all wire protocol structs/defines to lustre_idl.h
4471 * Fixes symlink length bug.
4472 * Add tcpdump to repository.
4474 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4475 * release version v0_4_3
4476 * Fixes statfs for inodes on extN.
4477 * Fixes bug in runtests which would delete /etc/hosts.
4478 * Use 64-bit object IDs wherever possible (not into VFS though)
4479 Remove ost_get_info, which is unused by lustre, and out of date.
4481 2002-07-03 Peter Braam <braam@clusterfs.com>
4482 * release version v0_4_2 Fixes a lookup error (type not passed)
4483 * move forward to head of Portals
4484 * move forward to latest Lustre kernel
4486 2002-06-25 Peter Braam <braam@clusterfs.com>
4487 * release version v0_4_1. Hopefully stable on single node use.