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 : occasional
18 Description: Do not fail import if osc_interpret_create gets -EAGAIN
19 Details : If osc_interpret_create got -EAGAIN it immediately exits and
20 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
21 oscc_has_objects and see OSC has no objests and call
22 oscc_internal_create to resend create request.
24 Severity : enhancement
26 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
28 Severity : enhancement
30 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
35 Description: Fix replay if there is an un-replied request and open
36 Details : In some cases, older replay request will revert the
37 mcd->mcd_last_xid on MDS which is used to record the client's
40 Severity : enhancement
42 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
44 Severity : enhancement
46 Description: Add rhel5 support to HEAD.
48 Serverity : enhancement
50 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
55 Description: Don't allow skipping OSTs if index has been specified.
56 Details : Don't allow skipping OSTs if index has been specified, make locking
57 in internal create lots better.
61 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
62 Details : access to bitfield in structure is always rounded to long
63 and this produce problem with not atomic change any bit.
67 Description: Lustre make rpms failed.
68 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
69 builds Lustre rpms from tarball.
72 Frequency : rare on shutdown ost
74 Description: If llog cancel was not send before clean_exports phase, this can
75 produce deadlock in llog code.
76 Details : If llog thread has last reference to obd and call class_import_put
77 this produce deadlock because llog_cleanup_commit_master wait when
78 last llog_commit_thread exited, but this never success because was
79 called from llog_commit_thread.
83 Description: allow userland application know is lost one of stripes.
84 Details : fill lvb_blocks with error code on ost and return it to
85 application if error flag found.
89 Description: NULL lov_tgts causing MDS oops
90 Details : more safe checks for NULL lov_tgts for avoid oops.
92 Severity : enhancement
94 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
98 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
99 Details : If an OST is down the MDS will hang indefinitely in
100 obd_statfs_rqset() waiting for the statfs data. While for
101 MDS QOS usage of statfs, it should not stuck in waiting.
103 Severity : enhancement
105 Description: remote_acl support
106 Details : Support ACL-based permission check for remote user.
107 Support setfacl/getfacl for remote user with the utils
108 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
109 the system "{s,g}etfacl" utils.
111 Severity : enhancement
113 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
115 Severity : enhancement
117 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
121 Description: make mgs_setparam() handle fsname containing dash
122 Details : fsname containing a dash does not work with lctl conf_param
124 Severity : enhancement
126 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
128 Severity : enhancement
130 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
134 Frequency : rare, at shutdown
135 Description: access already free / zero obd_namespace.
136 Details : if client_disconnect_export was called without force flag set,
137 and exist connect request in flight, this can produce access to
138 NULL pointer (or already free pointer) when connect_interpret
139 store ocd flags in obd_namespace.
143 Frequency : only at startup
144 Description: not alloc memory with spinlock held.
145 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
146 if any spinlock held.
148 Severity : enhancement
150 Description: make lustre randomly fail allocating memory
151 Details : Make lustre randomly failed allocating memory for testing purpose.
153 Severity : enhancement
155 Description: lost problems with lov objid file
156 Details : Fixes some scability and access to not inited memory problems
157 in work with lov objdid file.
162 Description: lfs find does not continue on file error
163 Details : Continue other files processing when a file/dir is absent.
167 Description: Inconsistent usage of lustre_pack_reply()
168 Details : Standardize the usage of lustre_pack_reply() such that it
169 always generate a CERROR on failure.
172 Frequency : occasional
174 Description: MDS hang or stay in waiting lock
175 Details : If client receive lock with CBPENDING flag ldlm need send lock
176 cancel as separate rpc, to avoid situation when cancel request
177 can't processed due all i/o threads stay in wait lock.
181 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
185 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
189 Description: Update to SLES9 kernel-2.6.5-7.287.3.
193 Description: Update to RHEL4 latest kernel.
195 Severity : enhancement
197 Description: Build SLES10 patchless client fails
198 Details : The configure was broken by run ./configure with
199 --with-linux-obj=.... argument for patchless client. When the
200 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
201 can't search header adequately. Use absolute path such as
202 -I($LINUX)/include instead.
206 Description: interrupt oig_wait produce painc on resend.
207 Details : brw_redo_request can be used for resend requests from ptlrpcd and
208 private set, and this produce situation when rq_ptlrpcd_data not
209 copyed to new allocated request and triggered LBUG on assert
210 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
211 ptlrpcd set if request is changed and can be safety changed to use
216 Description: LASSERT_{REQ,REP}SWAB macros are buggy
217 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
218 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
219 these macros and replace them with their difinitions instead.
223 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
224 Details : Update which_patch & target file for SLES10 latest kernel.
227 Description: Remove client patches from SLES 10 kernel.
228 Details : This causes SLES 10 clients to behave as patchless clients
229 even on a Lustre-patched (server) kernel.
231 Severity : enhancement
233 Description: self-adjustable client's lru lists
234 Details : use adaptive algorithm for managing client cached locks lru
235 lists according to current server load, other client's work
236 pattern, memory activities, etc. Both, server and client
237 side namespaces provide number of proc tunables for controlling
240 Severity : enhancement
242 Description: light-weight GSS support
243 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
244 and reduce performance overhead.
246 Severity : enhancement
248 Description: Linux keyring support
249 Details : Support using service of Linux keyring for Lustre GSS internal
250 context refresh/cache mechanism.
254 Description: Fix errors in lfs documentation
255 Details : Fixes man pages
259 Description: don't use GFP_* in generic Lustre code.
260 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
264 Description: obdclass is limited by single OBD_ALLOC(idarray)
265 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
269 Description: warnings with build patchless client with vanila 2.6.19 and up
270 Details : change old ctl_table style and replace ctl_table/ctl_table_header
271 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
275 Frequency : Only for SLES
276 Description: sanity_quota fail test_1
277 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
278 file, which makes TSTID[2] inunique.
282 Frequency : Only for SLES9
283 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
284 Details : Update target/ChangeLog/which_patch .
288 Frequency : Only for test_10 in sanity-quota.sh
289 Description: a bug in quota test code
290 Details : checking a test flag in a wrong place
294 Frequency : Only for test_13 in sanity-quota.sh
295 Description: a bug in quota test script
296 Details : shouldn't check overfull allocated quota
300 Description: jbd statistics
301 Details : Port older jbd statistics patch for sles10
303 Severity : enhancement
305 Description: Add printing inode info into message about error in writepage.
310 Description: reply_lock_interpret crash due to race with it and lock cancel.
311 Details : Do not replay locks that are being cancelled. Do not reference
312 locks by their address during replay, just by their handle.
316 Frequency : When flocks are used.
317 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
318 Details : Flock locks might destroy just granted lock if it could be merged
319 with another existing flock, this is done in completion handler,
320 so teach ldlm_cli_enquque_fini that this is a valid case for
326 Description: Oops in read and write path when failing to allocate lock.
327 Details : Check if lock allocation failed and return error back.
331 Description: lstripe command fails for valid OST index
332 Details : The stripe offset is compared to lov->desc.ld_tgt_count
333 instead of lov->desc.ld_active_tgt_count.
337 Description: sanity.sh failed test 103
338 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
343 Description: df doesn't work properly if diskfs blocksize != 4K
344 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
348 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
349 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
353 Description: obd_health_check_timeout too short
354 Details : set obd_health_check_timeout as 1.5x of obd_timeout
358 Description: llapi_file_create() does not allow some changes
359 Details : add llapi_file_open() that allows specifying the mode and
360 open flags, and also returns an open file handle.
364 Description: merge and cleanup kernel patches.
365 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
369 Description: Add journal checksum support.(Kernel part)
370 Details : The journal checksum feature adds two new flags i.e
371 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
372 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
373 indicates that the commit block contains the checksum for
374 the blocks described by the descriptor blocks. Now commit
375 record can be sent to disk without waiting for descriptor
376 blocks to be written to disk. This behavior is controlled
377 using JBD2_FEATURE_ASYNC_COMMIT flag.
381 Description: OSS needs mutliple precreate threads
382 Details : Add ability to start more than one create thread per OSS.
386 Description: Sanity.sh test_65k failed: setstripe should have succeeded
387 Details : execute lfs setstripe on client
391 Description: mds_obd_create error creating tmp object
392 Details : When the user sets quota on root, llog will be affected and can't
393 create files and write files.
396 Frequency : Always on ia64 patchless client, and possibly others.
398 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
399 Details : This allows the patchless client to be loaded on architectures
404 Description: RedHat Update kernel for RHEL5
405 Details : Add the kernel config file for RHEL5.
409 Description: RedHat Update kernel for RHEL5
410 Details : Modify the kernel config file more closer RHEL5.
414 Description: Build failure against Centos5 (RHEL5)
415 Details : Define PAGE_SIZE when it isn't present.
419 Description: test_7 on sanity-quota.sh can't be used on separate machines
420 Details : add facet to handle it
424 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
425 Details : with NFS, the anon dentry's parent was set to itself in
426 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
427 corresponding dentry other than use rec->ur_name.
429 Severity : enhancement
431 Description: lfs setstripe enhancement
432 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
436 Description: enable data checksumming by default
437 Details : enable checksum by default, allow --disable-checksum
438 configure option and "-o nochecksum" mount option. Checksums
439 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
443 Description: lustre support for RHEL5
444 Details : Add support for RHEL5.
448 Description: Client eviction due to failover config
449 Details : after a connection loss, the lustre client should attempt to
450 reconnect to the last active server first before trying the
451 other potential connections.
455 Description: when mds and osts use different quota unit(32bit and 64bit),
456 quota will be released repeatly.
457 Details : void sending multiple quota reqs to mds, which will keep the
458 status between the reqs.
462 Description: osts not allocated evenly to files
463 Details : change the condition to increase offset_idx
466 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
468 Description: Data corruption for OSTs that are formatted larger than 2TB
470 Details : When generating the bio request for lustre file writes the
471 sector number would overflow a temporary variable before being
472 used for the IO. The data reads correctly from Lustre (which
473 will overflow in a similar manner) but other file data or
474 filesystem metadata may be corrupted in some cases.
478 Description: Tune the kernel for good SCSI performance.
479 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
480 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
485 Description: rewrite ext2-derived code in obdclass/uuid.c
486 Details : rewrite inherited code (uuid parsing code from ext2 utils)
487 from scratch preserving functionality.
491 Description: "lfs find -obd UUID" prints directories
492 Details : "lfs find -obd UUID" will return all directory names instead
493 of just file names. It is incorrect because the directories
494 do not reside on the OSTs.
498 Description: use i_size_read and i_size_write in 2.6 port
499 Details : replace inode->i_size access with i_size_read/write()
502 Frequency : when removing large files
504 Description: scheduling issue during removal of large Lustre files
505 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
506 It causes scheduling issues when removing large files (17TB in the
510 Frequency : only with liblustre clients on XT3
512 Description: evictions taking too long
513 Details : allow llrd to evict clients directly on OSTs
516 Frequency : only on ppc
518 Description: /proc/fs/lustre/devices broken on ppc
519 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
520 We should make sure the type of 'index' is loff_t and then cast
521 explicitly as needed below. Do not assign an explicitly cast
525 Frequency : only for rhel5
527 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
528 Details : Modify the target file & which_kernel.
530 Severity : enhancement
532 Description: omit set fsid for export NFS
533 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
534 as FS_REQUIRES_DEV for avoid problems with generate fsid.
538 Frequency : Always, for kernels after 2.6.16
539 Description: Fix warning idr_remove called for id=.. which is not allocated.
540 Details : Last kernels save old s_dev before kill super and not allow
541 to restore from callback - restore it before call kill_anon_super.
545 Description: buffer overruns could theoretically occur
546 Details : llapi_semantic_traverse() modifies the "path" argument by
547 appending values to the end of the origin string, and a
548 overrun may occur. Adding buffer overrun check in liblustreapi.
552 Description: Fix error on 'ls .' at the top of the Lustre mount.
553 Details : Don't revalidate dentry if it is a root dentry.
557 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
558 Details : Modify vm-tunables-rhel4.patch.
562 Description: Kernel config for 2.6.18-vanilla.
563 Details : Modify targets/2.6-vanilla.target.in.
564 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
565 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
566 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
567 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
571 Description: improve handling recoverable errors
572 Details : if request processig with error which can be recoverable on server
573 request should be resend, otherwise page released from cache and
578 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
579 Details : Modify target file & which_patch.
580 A flaw was found in the IA32 system call emulation provided
581 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
582 value could be stored in the %RAX register, which could trigger an
583 out-of-bounds system call table access. An untrusted local user
584 could exploit this flaw to run code in the kernel
585 (ie a root privilege escalation). (CVE-2007-4573).
589 Description: O_DIRECT bypasses client statistics.
590 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
591 read_ahead_stats were updated. copied the stats section from
592 osc_send_oap_rpc() into async_internal().
596 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
600 Description: change order of libsysio includes
601 Details : '#include sysio.h' should always come before '#include xtio.h'
605 Description: handle "serious error: objid * already exists" more gracefully
606 Details : If LAST_ID value on disk is smaller than the objects existing in
607 the O/0/d* directories, it indicates disk corruption and causes an
608 LBUG(). If the object is 0-length, then we should use the existing
609 object. This will help to avoid a full fsck in most cases.
611 Severity : enhancement
613 Description: adapt the lustre_config script to support the upgrade case
614 Details : Add "-u" option for lustre_config script to support upgrading 1.4
615 server targets to 1.6 in parallel.
619 Description: To avoid grant space > avaible space when the disk is almost
620 full. Without this patch you might see the error "grant XXXX >
621 available" or some LBUG about grant, when the disk is almost
623 Details : In filter_check_grant, for non_grant cache write, we should
624 check the left space by if (*left > ungranted + bytes), instead
625 of (*left > ungranted), because only we are sure the left space
626 is enough for another "bytes", then the ungrant space should be
627 increase. In client, we should update cl_avail_grant only there
628 is OBD_MD_FLGRANT in the reply.
632 Description: Update RHEL 4 kernel to fix local root privilege escalation.
633 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
634 described in CVE-2007-4573. This problem could allow untrusted
635 local users to gain root access.
638 Frequency : when using O_DIRECT and quotas
640 Description: Incorrect file ownership on O_DIRECT output files
641 Details : block usage reported by 'lfs quota' does not take into account
642 files that have been written with O_DIRECT.
647 Description: touch file failed when fs is not full
648 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
649 otherwise we can get ENOSP while fs is not full.
653 Description: parallel lock callbacks
654 Details : Instead of sending blocking and completion callbacks as separated
655 requests, adding them to a set and sending in parallel.
658 Frequency : only for Cray XT3
659 Bugzilla : 12829/13455
660 Description: Changing primary group doesn't change the group lustre assigns to
662 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
663 overridden with the primary group provided by the group upcall,
664 whereas the supplied fsgid can be trusted if it is in the list of
665 supplementary groups returned by the group upcall.
667 Severity : enhancement
669 Description: Allow masking D_WARNING, D_ERROR messages from console
670 Details : Console messages can now be disabled via lnet.printk.
674 Description: User code with malformed file open parameter crashes client node
675 Details : Before packing join_file req, all the related reference should be
676 checked carefully in case some malformed flags cause fake
677 join_file req on client.
681 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
682 Details : ldlm_completion_ast() assumes that a lock is granted when the req
683 mode is equal to the granted mode. However, it should also check
684 that LDLM_FL_CP_REQD is not set.
688 Description: Heavy nfs access might result in deadlocks
689 Details : After ELC code landed, it is now improper to enqueue any mds
690 locks under och_sem, because enqueue might want to decide to
691 cancel open locks for same inode we are holding och_sem for.
695 Description: Client eviction while running blogbench
696 Details : A lot of unlink operations with concurrent I/O can lead to a
697 deadlock causing evictions. To address the problem, the number of
698 oustanding OST_DESTROY requests is now throttled to
699 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
700 callbacks are processed in priority.
704 Description: enable ACLs on MDS by default
705 Details : ACLs must be enabled on MDS by default.
708 Frequency : PPC/PPC64 only
710 Description: conflicts between asm-ppc64/types.h and lustre_types.h
711 Details : fix duplicated definitions between asm-ppc64/types.h and
712 lustre_types.h on PPC.
716 Description: better handle error messages in extents code
718 --------------------------------------------------------------------------------
720 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
722 * Support for kernels:
723 2.6.5-7.283 (SLES 9),
724 2.6.9-55.EL (RHEL 4),
725 2.6.16.46-0.14 (SLES 10),
726 2.6.18.8 vanilla (kernel.org)
727 * Client support for unpatched kernels:
728 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
729 2.6.16 - 2.6.22 vanilla (kernel.org)
730 * Due to recently discovered recovery problems, we do not recommend
731 using patchless RHEL 4 clients with this or any earlier release.
732 * Recommended e2fsprogs version: 1.39.cfs8
733 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
734 * Starting with this release, the ldiskfs backing filesystem required
735 by Lustre is now in its own package, lustre-ldiskfs. This package
736 should be installed. It is versioned separately from Lustre and
737 may be released separately in future.
741 Description: block reactivating mgc import until all deactivates complete
742 Details : Fix race when failing back MDT/MGS to itself (testing)
745 Frequency : at statup only
747 Description: mds_lov_synchronize race leads to various problems
748 Details : simultaneous MDT->OST connections at startup can cause the
749 sync to abort, leaving the OSC in a bad state.
751 Severity : enhancement
753 Description: add optional extra BUILD_VERSION info
754 Details : add a new environment variable (namely LUSTRE_VERS) which allows
755 to override the lustre version.
758 Frequency : 2.6.18 servers only
760 Description: ll_kern_mount() doesn't release the module reference
761 Details : The ldiskfs module reference count never drops down to 0
762 because ll_kern_mount() doesn't release the module reference.
767 Description: server LBUG when using old ost_num_threads parameter
768 Details : Accept the old ost_num_threads parameter but warn that it
769 is deprecated, and fix an off-by-one error that caused an LBUG.
774 Description: Transient SCSI error results in persistent IO issue
775 Details : iobuf->dr_error is not reinitialized to 0 between two
779 Frequency : sometimes when underlying device returns I/O errors
781 Description: OSTs not going read-only during write failures
782 Details : OSTs are not remounted read-only when the journal commit threads
783 get I/O errors because fsfilt_ext3 calls journal_start/stop()
784 instead of the ext3 wrappers.
788 Description: poor connect scaling with increasing client count
789 Details : Don't run filter_grant_sanity_check for more than 100 exports
790 to improve scaling for large numbers of clients.
793 Frequency : SLES10 only
795 Description: sanity-quota.sh quotacheck failed: rc = -22
796 Details : Quotas cannot be enabled on SLES10.
799 Frequency : liblustre clients only
801 Description: getdirentries does not give error when run on compute nodes
802 Details : getdirentries does not fail when the size specified as an argument
803 is too small to contain at least one entry
805 Severity : enhancement
807 Description: Add LNET router traceability for debug purposes
808 Details : If a checksum failure occurs with a router as part of the
809 IO path, the NID of the last router that forwarded the bulk data
810 is printed so it can be identified.
815 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
816 Details : Due to a race condition, liblustre clients were occasionally
819 Severity : enhancement
821 Description: lfs setstripe use optional parameters instead of postional
824 Severity : enhancement
826 Description: Nanosecond timestamp support for ldiskfs
827 Details : The on-disk ldiskfs filesystem has added support for nanosecond
828 resolution timestamps. There is not yet support for this at
829 the Lustre filesystem level.
832 Frequency : during server recovery
834 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
835 Details : request with rq_no_resend flag not awake l_wait_event if they get a
839 Frequency : nfs export on patchless client
841 Description: connectathon hang when test nfs export over patchless client
842 Details : Disconnected dentry cannot be found with lookup, so we do not need
843 to unhash it or make it invalid
846 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
847 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
848 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
853 Description: open req refcounting wrong on reconnect
854 Details : If reconnect happened between getting open reply from server and
855 call to mdc_set_replay_data in ll_file_open, we will schedule
856 replay for unreferenced request that we are about to free.
857 Subsequent close will crash in variety of ways.
858 Check that request is still eligible for replay in
859 mdc_set_replay_data().
864 Description: disable writes to filesystem when reading health_check file
865 Details : the default for reading the health_check proc file has changed
866 to NOT do a journal transaction and write to disk, because this
867 can cause reads of the /proc file to hang and block HA state
868 checking on a healthy but otherwise heavily loaded system. It
869 is possible to return to the previous behaviour during configure
870 with --enable-health-write.
872 Severity : enhancement
874 Description: 64-bit inode version
875 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
876 to the inode. This will be required for version-based recovery.
881 Description: MDS fails to start if a duplicate client export is detected
882 Details : in some rare cases it was possible for a client to connect to
883 an MDS multiple times. Upon recovery the MDS would detect this
884 and fail during startup. Handle this more gracefully.
886 Severity : enhancement
888 Description: Add -o localflock option to simulate old noflock
890 Details : This will achieve local-only flock/fcntl locks
896 Description: log_commit_thread vs filter_destroy race leads to crash
897 Details : Take import reference before releasing llog record semaphore
902 Description: Wrong request locking in request set processing
903 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
904 imp_delayed_list, in this place should be used imp_lock.
907 Frequency : when reconnection
909 Description: Grant Leak when osc reconnect to OST
910 Details : When osc reconnect ost, OST(filter) should check whether it
911 should grant more space to client by comparing fed_grant and
912 cl_avail_grant, and return the granted space to client instead
913 of "new granted" space, because client will call osc_init_grant
914 to update the client grant space info.
917 Frequency : when client reconnect to OST
919 Description: Grant Leak when osc do resend and replay bulk write
920 Details : When osc reconnect to OST, OST(filter)should clear grant info of
921 bulk write request, because the grant info while be sync between
922 OSC and OST when reconnect, and we should ignore the grant info
923 these of resend/replay write req.
928 Description: Grant space more than avaiable left space sometimes.
929 Details : When then OST is about to be full, if two bulk writing from
930 different clients came to OST. Accord the avaliable space of the
931 OST, the first req should be permitted, and the second one
932 should be denied by ENOSPC. But if the seconde arrived before
933 the first one is commited. The OST might wrongly permit second
934 writing, which will cause grant space > avaiable space.
937 Frequency : when client is evicted
939 Description: Grant might be wrongly erased when osc is evicted by OST
940 Details : when the import is evicted by server, it will fork another
941 thread ptlrpc_invalidate_import_thread to invalidate the
942 import, where the grant will be set to 0. While the original
943 thread will update the grant it got when connecting. So if
944 the former happened latter, the grant will be wrongly errased
945 because of this race.
950 Description: Checking Stale with correct fid
951 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
952 to check whether it is stale, because sometimes, we want the
953 enqueue happened anyway, and op_data.fid2 will not be initialized.
955 Severity : enhancement
957 Description: update patchless client
958 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
961 Frequency : only with 2.4 kernel
963 Description: random memory corruption
964 Details : size of struct ll_inode_info is to big for union inode.u and this
965 can be cause of random memory corruption.
970 Description: Memory leak in recovery
971 Details : Lov_mds_md was not free in an error handler in mds_create_object.
972 It should also check obd_fail before fsfilt_start, otherwise if
973 fsfilt_start return -EROFS,(failover mds during mds recovery).
974 then the req will return with repmsg->transno = 0 and rc = EROFS.
975 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
976 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
977 be freed no matter whether fsfilt_commit success or not.
980 Frequency : only with huge count clients
982 Description: Prevents from taking the superblock lock in llap_from_page for
984 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
985 for llap_from_page call in ll_removepage prevents from taking the
986 superblock lock for a soon died page.
991 Description: Not check open intent error before release open handle
992 Details : in some rare cases, the open intent error is not checked before
993 release open handle, which may cause
994 ASSERTION(open_req->rq_transno != 0), because it tries to release
995 the failed open handle.
1000 Description: Set cat log bitmap only after create log success.
1001 Details : in some rare cases, the cat log bitmap is set too early. and it
1002 should be set only after create log success.
1006 Description: Accessing a block bevice can re-enable I/O when Lustre is
1007 tearing down a device.
1008 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1013 Description: service threads may hog cpus when there are a lot of requests
1014 Details : Insert cond_resched to give other threads a chance to use some CPU
1019 Description: the cat log was not initialized in recovery
1020 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1021 unlink log on mds will not be initialized until mds post
1022 recovery. And also in mds post recovery, the unlink log will
1023 initialization will be done asynchronausly, so there will be race
1024 between add unlink log and unlink log initialization.
1028 Description: brw_stats were being printed incorrectly
1029 Details : brw_stats were being printed as log2 but all of them were not
1030 recorded as log2. Also remove some code duplication arising from
1031 filter_tally_{read,write}.
1035 Frequency : rare, only in recovery.
1036 Description: ASSERTION(req->rq_type != LI_POISON) failed
1037 Details : imp_lock should be held while iterating over imp_sending_list for
1038 prevent destroy request after get timeout in ptlrpc_queue_wait.
1042 Description: replay-single.sh test 52 fails
1043 Details : A lock's skiplist need to be cleanup when it being unlinked
1044 from its resource list.
1048 Description: Short directio read returns full requested size rather than
1050 Details : Direct I/O operations should return actual amount of bytes
1051 transferred rather than requested size.
1053 Severity : enhancement
1055 Description: metadata RPC reduction (e.g. for rm performance)
1056 Details : decrease the amount of synchronous RPC between clients and servers
1057 by canceling conflicing lock before the operation on the client
1058 and packing thier handles into the main operation RPC to server.
1060 Severity : enhancement
1062 Description: Async OSC create to avoid the blocking unnecessarily.
1063 Details : If a OST has no remain object, system will block on the creating
1064 when need to create a new object on this OST. Now, ways use
1065 pre-created objects when available, instead of blocking on an
1066 empty osc while others are not empty. If we must block, we block
1067 for the shortest possible period of time.
1069 Severity : enhancement
1071 Description: refine locking for avoid write wrong info into lov_objid file
1072 Details : fix possible races with add new target and write/update data in
1075 --------------------------------------------------------------------------------
1077 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1082 Frequency : on some architectures
1084 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1085 Details : Uninitialized flags sometimes cause configuration commands to
1089 Frequency : patchless clients only
1091 Description: missing __iget() symbol export
1092 Details : The __iget() symbol export is missing. To avoid the need for
1093 this on patchless clients the deathrow inode reaper is turned
1094 off, and we depend on the VM to clean up old inodes. This
1095 dependency was during via the fix for bug 12181.
1097 --------------------------------------------------------------------------------
1099 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1101 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1102 INTEROPERATE with older versions automatically. In many cases a
1103 special upgrade step is needed. Please read the
1104 user documentation before upgrading any part of a live system.
1105 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1106 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1107 * WARNING: Lustre configuration and startup changes are required with
1108 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1114 Severity : enhancement
1116 Description: MountConf
1117 Details : Lustre configuration is now managed via mkfs and mount
1118 commands instead of lmc and lconf. New obd types (MGS, MGC)
1119 are added for dynamic configuration management. See
1120 https://mail.clusterfs.com/wikis/lustre/MountConf for
1123 Severity : enhancement
1125 Description: dynamic OST addition
1126 Details : OSTs can now be added to a live filesystem
1128 Severity : enhancement
1130 Description: startup order invariance
1131 Details : MDTs and OSTs can be started in any order. Clients only
1132 require the MDT to complete startup.
1134 Severity : enhancement
1136 Description: parallel, asynchronous orphan cleanup
1137 Details : orphan cleanup is now performed in separate threads for each
1138 OST, allowing parallel non-blocking operation.
1140 Severity : enhancement
1142 Description: optimized stripe assignment
1143 Details : stripe assignments are now made based on ost space available,
1144 ost previous usage, and OSS previous usage, in order to try
1145 to optimize storage space and networking resources.
1147 Severity : enhancement
1149 Description: Permanently set tunables
1150 Details : All writable /proc/fs/lustre tunables can now be permanently
1151 set on a per-server basis, at mkfs time or on a live system.
1153 Severity : enhancement
1155 Description: Lustre message v2
1156 Details : Add lustre message format v2.
1158 Severity : enhancement
1160 Description: client OST exclusion list
1161 Details : Clients can be started with a list of OSTs that should be
1162 declared "inactive" for known non-responsive OSTs.
1166 Description: SPEC SFS validation failure on NFS v2 over lustre.
1167 Details : Changes the blocksize for regular files to be 2x RPC size,
1168 and not depend on stripe size.
1170 Severity : enhancement
1172 Description: Multiple MD RPCs in flight.
1173 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1174 To avoid overly-overloading MDS, introduce a limit on number of
1175 MDS RPCs in flight for a single client and add /proc controls
1176 to adjust this limit.
1178 Severity : enhancement
1180 Description: client read/write statistics
1181 Details : Add client read/write call usage stats for performance
1182 analysis of user processes.
1183 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1184 file access. extents_stats shows chunk size distribution.
1185 extents_stats_per_process show chunk size distribution per
1188 Severity : enhancement
1190 Description: mds statistics
1191 Details : Add detailed mds operations statistics in
1192 /proc/fs/lustre/mds/*/stats.
1196 Description: Failure of copying files with lustre special EAs.
1197 Details : Client side always return success for setxattr call for lustre
1198 special xattr (currently only "trusted.lov").
1203 Description: Refcount LNET uuids
1204 Details : The global LNET uuid list grew linearly with every startup;
1205 refcount repeated list entries instead of always adding to
1208 Severity : enhancement
1210 Description: Dynamic service threads
1211 Details : Within a small range, start extra service threads
1212 automatically when the request queue builds up.
1215 Frequency : mixed-endian client/server environments
1217 Description: mixed-endian crashes
1218 Details : The new msg_v2 system had some failures in mixed-endian
1221 Severity : enhancement
1223 Description: Easy OST removal
1224 Details : OSTs can be permanently deactivated with e.g. 'lctl
1225 conf_param lustre-OST0001.osc.active=0'
1227 Severity : enhancement
1229 Description: MGS proc entries
1230 Details : Added basic proc entries for the MGS showing what filesystems
1233 Severity : enhancement
1235 Description: provide MGS failover
1236 Details : Added config lock reacquisition after MGS server failover.
1238 Severity : enhancement
1240 Description: add Linux 2.4 support
1241 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1245 Description: a large application tries to do I/O to the same resource and dies
1246 in the middle of it.
1247 Details : Check the req->rq_arrival time after the call to
1248 ost_brw_lock_get(), but before we do anything about
1249 processing it & sending the BULK transfer request. This
1250 should help move old stale pending locks off the queue as
1251 quickly as obd_timeout.
1254 Frequency : when an incorrect nid is specified during startup
1256 Description: ptlrpc connect to non-existant node causes kernel crash
1257 Details : LNET can't be re-entered from an event callback, which
1258 happened when we expire a message after the export has been
1259 cleaned up. Instead, hand the zombie cleanup off to another
1262 Severity : enhancement
1264 Description: plain/inodebits lock performance improvement
1265 Details : Grouping plain/inodebits in granted list by their request modes
1266 and bits policy, thus improving the performance of search through
1270 Frequency : only if OST filesystem is corrupted
1272 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1273 Details : for a short time RPCs with bulk IO are in the replay list,
1274 but replay of bulk IOs is unimplemented. If the OST filesystem
1275 is corrupted due to disk cache incoherency and then replay is
1276 started it is possible to trip an assertion. Avoid putting
1277 committed RPCs into the replay list at all to avoid this issue.
1280 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1282 Description: System hang on startup
1283 Details : This bug allowed the liblustre (e.g. catamount) client to
1284 return to the app before handling all startup RPCs. This
1285 could leave the node unresponsive to lustre network traffic
1286 and manifested as a server ptllnd timeout.
1288 Severity : enhancement
1290 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1291 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1292 boolean to control whether to print peer debug info when a
1293 client's RPC times out.
1296 Frequency : only for kernels with patches from Lustre below 1.4.3
1298 Description: Remove old rdonly API
1299 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1302 Frequency : only for devices with external journals
1304 Description: Set external device read-only also
1305 Details : During a commanded failover stop, we set the disk device
1306 read-only while the server shuts down. We now also set any
1307 external journal device read-only at the same time.
1310 Frequency : when upgrading from 1.4 while trying to change parameters
1312 Description: The wrong (new) MDC name was used when setting parameters for
1313 upgraded MDT's. Also allows changing of OSC (and MDC)
1314 parameters if --writeconf is specified at tunefs upgrade time.
1317 Frequency : when setting specific ost indicies
1319 Description: QOS code breaks on skipped indicies
1320 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1321 created with --index need not be sequential.
1323 Severity : enhancement
1325 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1326 Details : The uninit_groups feature works in conjunction with the kernel
1327 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1328 up the pass1 processing of e2fsck. This is a read-only feature
1329 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1330 mount filesystems that have had this feature enabled.
1332 Severity : enhancement
1334 Description: Improve multi-block allocation algorithm to avoid fragmentation
1335 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1336 allocation locality and avoid filesystem fragmentation.
1338 ------------------------------------------------------------------------------
1340 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1342 * Support for kernels:
1343 2.4.21-47.0.1.EL (RHEL 3)
1344 2.6.5-7.283 (SLES 9)
1345 2.6.9-42.0.10.EL (RHEL 4)
1346 2.6.12.6 vanilla (kernel.org)
1347 2.6.16.27-0.9 (SLES 10)
1348 * Recommended e2fsprogs version: 1.39.cfs5
1350 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1354 Frequency : occasional, depends on client load and configuration
1355 Bugzilla : 12181, 12203
1356 Description: data loss for recently-modified files
1358 Details : In some cases it is possible that recently written or created
1359 files may not be written to disk in a timely manner (this should
1360 normally be within 30s unless client IO load is very high).
1361 The problem appears as zero-length files or files that are a
1362 multiple of 1MB in size after a client crash or client eviction
1363 that are missing data at the end of the file.
1365 This problem is more likely to be hit on clients where files are
1366 repeatedly created and unlinked in the same directory, clients
1367 have a large amount of RAM, have many CPUs, the filesystem has
1368 many OSTs, the clients are rebooted frequently, and/or the files
1369 are not accessed by other nodes after being written.
1371 The presence of the problem can be detected by looking at
1372 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1373 smaller than the second (nr_unused) then dirty files will not
1374 be flushed automatically to disk. "sync; sleep 10" should be
1375 run several times on the node before unmounting it to update
1376 Lustre (this is also safe to run on nodes without this problem).
1378 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1379 that can cause this same problem, so customers using that kernel
1380 also need to update the kernel in addition to Lustre. In order
1381 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1383 It is normal that files written just before a client crash (less
1384 than 30s) may not yet have been flushed to disk, even for local
1388 Frequency : frequent on thin XT3 nodes
1390 Description: UUID collision on thin XT3 Linux nodes
1391 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1392 randomly, since we relied on an insufficiently-seeded PRNG.
1397 Description: OSS hangs after "All ost request buffers busy"
1398 Details : A deadlock between quota and journal operations caused OSS
1399 hangs after printing "All ost request buffers busy."
1402 Frequency : always on liblustre builds
1404 Description: Cleanup compiler warnings on liblustre
1407 Frequency : always on liblustre builds on XT3
1409 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1413 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1414 > 32000 subdirectories
1415 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1416 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1417 subdirectory count crosses 32000. This will aid e2fsck to
1418 correctly handle more than 32000 subdirectories.
1421 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1423 Description: System hang on startup
1424 Details : This bug allowed the liblustre (e.g. catamount) client to
1425 return to the app before handling all startup RPCs. This
1426 could leave the node unresponsive to lustre network traffic
1427 and manifested as a server ptllnd timeout.
1429 Severity : enhancement
1431 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1432 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1433 boolean to control whether to print peer debug info when a
1434 client's RPC times out.
1439 Description: make O_SYNC working on 2.6 kernels
1440 Details : 2.6 kernels use different method for mark pages for write,
1441 so need add a code to lustre for O_SYNC work.
1446 Description: Failure to close file and release space on NFS
1447 Details : Put inode details into lock acquired in ll_intent_file_open.
1448 Use mdc_intent_lock in ll_intent_open to properly
1449 detect all kind of errors unhandled by mdc_enqueue.
1454 Description: proc file read during shutdown sometimes raced obd removal,
1456 Details : Add lock to prevent obd access after proc file removal.
1459 Frequency : Only for files larger than 4GB on 32-bit clients.
1461 Description: improperly doing page alignment of locks
1462 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1463 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1468 Description: RPCs being resent when they shouldn't be
1469 Details : Some RPCs that should not be resent are being resent. This
1470 can cause inconsistencies in the RPC state machine. Do not
1471 resend such requests.
1474 Frequency : rare, only with NFS export
1476 Description: Crash on NFS re-export node
1477 Details : under very unusual load conditions an assertion is hit in
1478 ll_intent_file_open()
1481 Frequency : only if OST filesystem is corrupted
1483 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1484 Details : for a short time RPCs with bulk IO are in the replay list,
1485 but replay of bulk IOs is unimplemented. If the OST filesystem
1486 is corrupted due to disk cache incoherency and then replay is
1487 started it is possible to trip an assertion. Avoid putting
1488 committed RPCs into the replay list at all to avoid this issue.
1493 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1494 Details : Large single O_DIRECT read and write calls can fail to allocate
1495 a sufficiently large buffer to process the request. In case of
1496 allocation failure the allocation is retried with a smaller
1497 buffer and broken into smaller requests.
1499 Severity : enhancement
1501 Description: Add -o localflock option to simulate old noflock behaviour.
1502 Details : This will achieve local-only flock/fcntl locks coherentness.
1507 Description: versioning check is incomplete
1508 Details : Checking the version difference of client vs. server, report
1509 error if the gap is too big.
1514 Description: add support PG_writeback bit
1515 Details : add support for PG_writeback bit for Lustre, for more carefull
1516 work with page cache in 2.6 kernel. This also fix some deadlocks
1517 and remove hack for work O_SYNC with 2.6 kernel.
1519 Severity : enhancement
1521 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1522 Details : The uninit_groups feature works in conjunction with the kernel
1523 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1524 up the pass1 processing of e2fsck. This is a read-only feature
1525 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1526 mount filesystems that have had this feature enabled.
1528 Severity : enhancement
1530 Description: Improve multi-block allocation algorithm to avoid fragmentation
1531 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1532 allocation locality and avoid filesystem fragmentation.
1534 ------------------------------------------------------------------------------
1536 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1538 * Support for kernels:
1539 2.6.9-42.0.3.EL (RHEL 4)
1540 2.6.5-7.276 (SLES 9)
1541 2.4.21-47.0.1.EL (RHEL 3)
1542 2.6.12.6 vanilla (kernel.org)
1543 2.6.16.21-0.8 (SLES10)
1544 * Recommended e2fsprogs version: 1.39.cfs2-0
1546 * The backwards-compatible /proc/sys/portals symlink has been removed
1547 in this release. Before upgrading, please ensure that you change
1548 any configuration scripts or /etc/sysctl.conf files that access
1549 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1550 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1551 in advance of the upgrade on any system running Lustre 1.4.6 or
1552 newer, since /proc/sys/lnet was added in that version.
1553 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1557 Frequency : only when quota is used
1559 Description: avoid scanning export list for quota master
1560 Details : Change the algorithms to avoid scanning export list in order
1561 to improve the efficiency.
1564 Frequency : MDS failover only, very rarely
1566 Description: "went back in time" messages on mds failover
1567 Details : The greatest transno may be lost when the current operation
1568 finishes with an error (transno==0) and the client's last_rcvd
1569 record is over-written. Save the greatest transno in the
1570 mds_last_transno for this case.
1573 Frequency : always for specific kernels and striping counts
1575 Description: client may get "Matching packet too big" without ACL support
1576 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1577 when trying to access files in certain configurations. The
1578 clients should in fact be denied when mounting because they do
1579 not understand ACLs.
1582 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1584 Description: many clients connecting with IO in progress causes connect timeouts
1585 Details : Avoid synchronous journal commits to avoid delays caused by many
1586 clients connecting/disconnecting when bulk IO is in progress.
1587 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1588 OST_IO_PORTAL to avoid delays behind potentially many pending
1592 Frequency : occasionally with multiple writers to a single file
1594 Description: shared writes to file may result in wrong size reported by stat()
1595 Details : Allow growing of kms when extent lock is cancelled
1598 Frequency : always with random mmap IO to multi-striped file
1600 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1601 Details : Only if the hole is at the end of OST object so that kms is too
1602 small. Fix is to increase kms accordingly in ll_nopage.
1605 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1607 Description: writes to a missing object would leak memory on the OST
1608 Details : If there is an inconsistency between the MDS and OST filesystems,
1609 such that the MDS references an object that doesn't exist, writes
1610 to that object will leak memory due to incorrect cleanup in the
1611 error handling path, eventually running out of memory on the OST.
1616 Description: Creating too long symlink causes lustre errors
1617 Details : Check symlink and name lengths before sending requests to MDS.
1620 Frequency : only if flock is enabled (not on by default)
1622 Description: posix locks not released on fd closure on 2.6.9+
1623 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1624 kernels, this caused such locks not to be released on fd close and
1625 then assertions on fs unmount about still used locks.
1628 Frequency : MDS failover only, very rarely
1630 Description: clients may get ASSERTION(granted_lock != NULL)
1631 Details : When request was taking a long time, and a client was resending
1632 a getattr by name lock request. The were multiple lock requests
1633 with the same client lock handle and
1634 mds_getattr_name->fixup_handle_for_resent_request found one of the
1635 lock handles but later failed with ASSERTION(granted_lock != NULL).
1640 Description: handle->h_buffer_credits > 0, assertion failure
1641 Details : h_buffer_credits is zero after truncate, causing assertion
1642 failure. This patch extends the transaction or creates a new
1646 Frequency : NFS re-export or patchless client
1647 Bugzilla : 11179, 10796
1648 Description: Crash on NFS re-export node (__d_move)
1649 Details : We do not want to hash the dentry if we don't have a lock.
1650 But if this dentry is later used in d_move, we'd hit uninitialised
1651 list head d_hash, so we just do this to init d_hash field but
1652 leave dentry unhashed.
1655 Frequency : NFS re-export or patchless client
1657 Description: NFS exports has problem with symbolic link
1658 Details : lustre client didn't properly install dentry when re-exported
1659 to NFS or running patchless client.
1662 Frequency : NFS re-export or patchless client
1664 Description: Various nfs/patchless fixes.
1665 Details : fixes reuse disconected alias for lookup process - this fixes
1666 warning "find_exported_dentry: npd != pd",
1667 fix permission error with open files at nfs.
1668 fix apply umask when do revalidate.
1671 Frequency : occasional
1673 Description: Crash on NFS re-export node
1674 Details : calling clear_page() on the wrong pointer triggered oops in
1675 generic_mapping_read().
1678 Frequency : rarely, using O_DIRECT IO
1680 Description: unaligned directio crashes client with LASSERT
1681 Details : check for unaligned buffers before trying any requests.
1684 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1686 Description: stale data returned from RAID cache
1687 Details : If only a small amount of IO is done to the RAID device before
1688 reading it again it is possible to get stale data from the RAID
1689 cache instead of reading it from disk.
1692 Frequency : always for sles10 kernel
1694 Description: sles10 support
1695 Details : ll_follow_link: compile fixes and using of nd_set_link
1696 under newer kernels.
1699 Frequency : depends on arch, kernel and compiler version, always on sles10
1702 Description: recursive or deep enough symlinks cause stack overflow
1703 Details : getting rid of large stack-allocated variable in
1707 Frequency : depends on hardware
1709 Description: lustre write performance loss in the SLES10 kernel
1710 Details : the performance loss is caused by using of write barriers in the
1711 ext3 code. The SLES10 kernel turns barrier support on by
1712 default. The fix is to undo that change for ldiskfs.
1714 ------------------------------------------------------------------------------
1716 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1718 * Support for kernels:
1719 2.6.9-42.0.3EL (RHEL 4)
1720 2.6.5-7.276 (SLES 9)
1721 2.4.21-47.0.1.EL (RHEL 3)
1722 2.6.12.6 vanilla (kernel.org)
1726 Frequency : quota enabled and large files being deleted
1728 Description: releasing more than 4GB of quota at once hangs OST
1729 Details : If a user deletes more than 4GB of files on a single OST it
1730 will cause the OST to spin in an infinite loop. Release
1731 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1736 Description: statfs data retrieved from /proc may be stale or zero
1737 Details : When reading per-device statfs data from /proc, in the
1738 {kbytes,files}_{total,free,avail} files, it may appear
1739 as zero or be out of date.
1742 Frequency : systems with MD RAID1 external journal devices
1744 Description: lconf's call to blkid is confused by RAID1 journal devices
1745 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1746 of returning all block devices that match the journal UUID.
1749 Frequency : always, for aggregate stripe size over 4GB
1751 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1752 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1753 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1756 Frequency : always if "lfs find" used on a local file/directory
1758 Description: "lfs find" segfaults if used on a local file/directory
1759 Details : The case where a directory component was not specified wasn't
1760 handled correctly. Handle this properly.
1763 Frequency : always on ppc64
1765 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1766 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1767 perform endianness conversion.
1770 Frequency : rarely (truncate to non-zero file size after write under load)
1771 Bugzilla : 10730, 10687
1772 Description: Files padded with zeros to next 4K multiple
1773 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1774 it is possible that files that are truncated to a non-zero size
1775 immediately after being written are filled with zero bytes beyond
1776 the truncated size. No file data is lost.
1778 Severity : enhancement
1780 Description: Allow recovery/failover for liblustre clients.
1781 Details : liblustre clients were unaware of failover configurations until
1784 Severity : enhancement
1786 Description: user file locks should fail when not mounting with flock option
1787 Details : Set up an error-returning stub in ll_file_operations.lock field
1788 to prevent incorrect behaviour when client is mounted without
1789 flock option. Also, set up properly f_op->flock field for
1793 Frequency : always on ia64
1795 Description: "lfs df" loops on printing out MDS statfs information
1796 Details : The obd_ioctl_data was not initialized and in some systems
1797 this caused a failure during the ioctl that did not return
1798 an error. Initialize the struct and return an error on failure.
1801 Frequency : SLES 9 only
1803 Description: Error of copying files with lustre special EAs as root
1804 Details : Client side always return success for setxattr call for lustre
1805 special xattr (currently only "trusted.lov").
1808 Frequency : rarely on clusters with both ia64+i386 clients
1810 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1811 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1812 clients are doing concurrent writes to the same file it is
1813 possible that the ia64 clients may LASSERT because the OST
1814 extent locks are not PAGE_SIZE aligned. Ensure that grown
1815 locks are always aligned on the request boundary.
1818 Frequency : specific use, occasional
1820 Description: Overwriting in use executable truncates on-disk binary image
1821 Details : If one node attempts to overwrite an executable in use by
1822 another node, we now correctly return ETXTBSY instead of
1823 truncating the file.
1825 Severity : enhancement
1827 Description: Async OSC create to avoid the blocking unnecessarily.
1828 Details : If a OST has no remain object, system will block on the creating
1829 when need to create a new object on this OST. Now, ways use
1830 pre-created objects when available, instead of blocking on an
1831 empty osc while others are not empty. If we must block, we block
1832 for the shortest possible period of time.
1837 Description: chmod on Lustre root is propagated to other clients
1838 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1839 invalid by the follow_mount time.
1842 Frequency : liblustre clients only
1844 Description: Race in 'instant cancel' lock handling could lead to such locks
1845 never to be granted in case of SMP MDS
1846 Details : Do not destroy not yet granted but cbpending locks in
1850 Frequency : replay/resend of open
1852 Description: non null lock assetion failure in mds_intent_policy
1853 Details : Trying to replay/resend lockless open requests resulted in
1854 mds_open() returning 0 with no lock. Now it sets a flag if
1855 a lock is going to be returned.
1857 Severity : enhancement
1859 Description: Checksum enhancements
1860 Details : New checksum enhancements allow for resending RPCs that failed
1863 Severity : enhancement
1865 Description: Tunables on number of dirty pages in cacche
1866 Details : Allow to set limit on number of dirty pages cached.
1871 Description: client crash on unmount - lock still has references
1872 Details : In some error handling cases it was possible to leak a lock
1873 reference on a client while accessing a file. This was not
1874 harmful to the client during operation, but would cause the
1875 client to crash when the filesystem is unmounted.
1878 Frequency : specific case, rare
1880 Description: ETXTBSY on mds though file not in use by client
1881 Details : ETXTBSY is no longer incorrectly returned when attempting to
1882 chmod or chown a directory that the user previously tried to
1883 execute or a currently-executing binary.
1886 Frequency : extremely rare except on liblustre-based clients
1888 Description: Lustre space not freed when files are deleted
1889 Details : Clean up open-unlinked files after client eviction. Previously
1890 the unlink was skipped and the files remained as orphans.
1895 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1896 Details : In some cases it was possible to send a blocking callback to
1897 a client doing a glimpse, even though that client didn't get
1898 a lock granted. When the glimpse lock is cancelled on the OST
1899 the freed lock is left on the waiting list and corrupted the list.
1902 Frequency : all core dumps
1904 Description: Broke core dumps to lustre
1905 Details : Negative dentry may be unhashed if parent does not have UPDATE
1906 lock, but some callers, e.g. do_coredump, expect dentry to be
1907 hashed after successful create, hash it in ll_create_it.
1909 ------------------------------------------------------------------------------
1911 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1913 * Support for kernels:
1914 2.6.9-42.0.2.EL (RHEL 4)
1915 2.6.5-7.276 (SLES 9)
1916 2.4.21-40.EL (RHEL 3)
1917 2.6.12.6 vanilla (kernel.org)
1921 Frequency : always on RHEL 3
1923 Description: Number of open files grows over time
1924 Details : The number of open files grows over time, whether or not
1925 Lustre is started. This was due to a filp leak introduced
1926 by one of our kernel patches.
1928 ------------------------------------------------------------------------------
1930 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1932 * Support for kernels:
1933 2.6.9-42.EL (RHEL 4)
1934 2.6.5-7.276 (SLES 9)
1935 2.4.21-40.EL (RHEL 3)
1936 2.6.12.6 vanilla (kernel.org)
1941 Bugzilla : 5719, 9635, 9792, 9684
1942 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1943 Details : If a server is under heavy load and cannot reply to new
1944 connection requests before the client resends the (re)connect,
1945 the connection handling code can behave badly if two service
1946 threads are concurrently handing separate (re)connections from
1947 the same client. Add better locking to the connection handling
1948 code, and ensure that only a single connection will be processed
1949 for a given client UUID, even if the lock is dropped.
1951 Severity : enhancement
1953 Description: add TCP zero-copy support to kernel
1954 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1955 sends if the hardware supports scatter-gather and checksumming.
1956 This allows socklnd to do client-write and server-read more
1957 efficiently and reduce CPU utilization from skbuf copying.
1960 Frequency : only if NFS exporting from client
1962 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1963 Details : If setting attributes on a file created under NFS that had
1964 never been opened it would be possible to oops the client
1965 if the file had no objects.
1968 Frequency : always for liblustre
1970 Description: liblustre client does MDS+OSTs setattr RPC for each write
1971 Details : When doing a write from a liblustre client, the client
1972 incorrectly issued an RPC to the MDS and each OST the file was
1973 striped over in order to update the timestamps. When writing
1974 with small chunks and many clients this could overwhelm the MDS
1975 with RPCs. In all cases it would slow down the write because
1976 these RPCs are unnecessary.
1978 Severity : enhancement
1980 Description: allow number of MDS service threads to be changed at module load
1981 Details : It is now possible to change the number of MDS service threads
1982 running. Adding "options mds mds_num_threads={N}" to the MDS's
1983 /etc/modprobe.conf will set the number of threads for the next
1984 time Lustre is restarted (assuming the "mds" module is also
1985 reloaded at that time). The default number of threads will
1986 stay the same, 32 for most systems.
1991 Description: OST crash if filesystem is unformatted or corrupt
1992 Details : If an OST is started on a device that has never been formatted
1993 or if the filesystem is corrupt and cannot even mount then the
1994 error handling cleanup routines would dereference a NULL pointer.
1999 Description: NULL pointer deref in llap_from_page.
2000 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2001 incorrect) mapping if the page was truncated/reclaimed while it was
2002 searched for. Check for this condition and skip such pages when
2003 doing readahead. Introduce extra check to llap_from_page() to
2004 verify page->mapping->host is non-NULL (so page is not anonymous).
2007 Frequency : Sometimes when using sys_sendfile
2009 Description: "page not covered by a lock" warnings from ll_readpage
2010 Details : sendfile called ll_readpage without right page locks present.
2011 Now we introduced ll_file_sendfile that does necessary locking
2012 around call to generic_file_sendfile() much like we do in
2016 Frequency : with certain MDS communication failures at client mount time
2018 Description: NULL pointer deref after failed client mount
2019 Details : a client connection request may delayed by the network layer
2020 and not be sent until after the PTLRPC layer has timed out the
2021 request. If the client fails the mount immediately it will try
2022 to clean up before the network times out the request. Add a
2023 reference from the request import to the obd device and delay
2024 the cleanup until the network drops the request.
2027 Frequency : occasionally during client (re)connect
2029 Description: assertion failure during client (re)connect
2030 Details : processing a client connection request may be delayed by the
2031 client or server longer than the client connect timeout. This
2032 causes the client to resend the connection request. If the
2033 original connection request is replied in this interval, the
2034 client may trip an assertion failure in ptlrpc_connect_interpret()
2035 which thought it would be the only running connect process.
2038 Frequency : only with obd_echo servers and clients that are rebooted
2040 Description: kernel BUG accessing uninitialized data structure
2041 Details : When running an obd_echo server it did not start the ping_evictor
2042 thread, and when a client was evicted an uninitialized data
2043 structure was accessed. Start the ping_evictor in the RPC
2044 service startup instead of the OBD startup.
2046 Severity : enhancement
2047 Bugzilla : 10193 (patchless)
2048 Description: Remove dependency on various unexported kernel interfaces.
2049 Details : No longer need reparent_to_init, exit_mm, exit_files,
2050 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2053 Frequency : rare (only users of deprecated and unsupported LDAP config)
2055 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2056 Details : LDAP apparently contains 'lustreName' attributes instead of
2057 'name'. A simple remapping of the name is sufficient.
2060 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2062 Description: waiting_locks_callback trips kernel BUG if client is evicted
2063 Details : Running with the dump_on_timeout debug flag turned on makes
2064 it possible that the waiting_locks_callback() can try to dump
2065 the Lustre kernel debug logs from an interrupt handler. Defer
2066 this log dumping to the expired_lock_main() thread.
2068 Severity : enhancement
2070 Description: Support NFS exporting on 2.6 kernels.
2071 Details : Implement non-rawops metadata methods for NFS server to use without
2072 changing NFS server code.
2075 Frequency : very rare (synthetic metadata workload only)
2077 Description: two racing renames might cause an MDS thread to deadlock
2078 Details : Running the "racer" program may cause one MDS thread to rename
2079 a file from being the source of a rename to being the target of
2080 a rename at exactly the same time that another thread is doing
2081 so, and the second thread has already enqueued these locks after
2082 doing a lookup of the target and is trying to relock them in
2083 order. Ensure that we don't try to re-lock the same resource.
2086 Frequency : only very large systems with liblustre clients
2088 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2089 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2090 all OSTs in parallel. This allows the eviction of stale liblustre
2091 clients to proceed much faster than if they were done in series,
2092 and also offers similar improvements for other set_info RPCs.
2097 Description: excessive CPU usage during initial read phase on client
2098 Details : During the initial read phase on a client, it would agressively
2099 retry readahead on the file, consuming too much CPU and impacting
2100 performance (since 1.4.5.8). Improve the readahead algorithm
2101 to avoid this, and also improve some other common cases (read
2102 of small files in particular, where "small" is files smaller than
2103 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2108 Description: MDS crash when receiving packet with unknown intent.
2109 Details : Do not LBUG in unknown intent case, just return -EFAULT
2111 Severity : enhancement
2112 Bugzilla : 9293, 9385
2113 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2114 Details : Do not serialize getattr (non-intent version) and statfs.
2117 Frequency : occasional, when OST network is overloaded/intermittent
2119 Description: client evicted by OST after bulk IO timeout
2120 Details : If a client sends a bulk IO request (read or write) the OST
2121 may evict the client if it is unresposive to its data GET/PUT
2122 request. This is incorrect if the network is overloaded (takes
2123 too long to transfer the RPC data) or dropped the OST GET/PUT
2124 request. There is no need to evict the client at all, since
2125 the pinger and/or lock callbacks will handle this, and the
2126 client can restart the bulk request.
2129 Frequency : Always when mmapping file with no objects
2131 Description: client crashes when mmapping file with no objects
2132 Details : Check that we actually have objects in a file before doing any
2133 operations on objects in ll_vm_open, ll_vm_close and
2139 Description: Request leak when working with deleted CWD
2140 Details : Introduce advanced request refcount tracking for requests
2141 referenced from lustre intent.
2143 Severity : Enhancement
2145 Description: Cache open file handles on client.
2146 Details : MDS now will return special lock along with openhandle, if
2147 requested and client is allowed to hold openhandle, even if unused,
2148 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2149 closing files for every read/write openration.
2151 Severity : Enhancement
2153 Description: Cache open negative dentries on client when possible.
2154 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2155 negative dentries on lock revocation.
2160 Description: Remounting a client read-only wasn't possible with a zconf mount
2161 Details : It wasn't possible to remount a client read-only with llmount.
2163 Severity : enhancement
2164 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2165 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2166 MPICH is included to add Lustre-specific ADIO interfaces.
2167 This is based closely on the UFS ADIO layer and only differs
2168 in file creation, in order to allow the OST striping to be set.
2169 This is user-contributed code and not supported by CFS.
2174 Description: extended inode attributes (immutable, append-only) work improperly
2175 when 2.4 and 2.6 kernels are used on client/server or vice versa
2176 Details : Introduce kernel-independent values for these flags.
2178 Severity : enhancement
2181 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2182 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2183 granularity. Now, Fractional MB granularity is be supported,
2184 this is very useful for low memory system.
2186 Severity : enhancement
2188 Description: Getattr by fid
2189 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2190 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2191 have locks and avoids getattr rpc requests for it on every stat.
2194 Frequency : Always, for filesystems larger than 2TB
2196 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2197 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2198 filesystems larger than 2TB to be mounted with mballoc enabled.
2201 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2203 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2204 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2205 filesystems larger than 2TB is mounted then the kernel will
2206 silently corrupt the start of the filesystem. CONFIG_LBD is
2207 enabled for all CFS-supported kernels, but the possibility of
2208 this happening with a modified kernel config exists.
2210 Severity : enhancement
2212 Description: add client O_DIRECT support for 2.6 kernels
2213 Details : It is now possible to do O_DIRECT reads and writes to files
2214 in the Lustre client mountpoint on 2.6 kernel clients.
2216 Severity : enhancement
2218 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2219 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2220 parallel, not waiting for response from every OST before sending
2221 a rpc to the next OST.
2226 Description: setattr vs write race when updating file timestamps
2227 Details : Client processes that update a file timestamp into the past
2228 right after writing to the file (e.g. tar) it is possible that
2229 the updated file modification time can be reset to the current
2230 time due to a race between processing the setattr and write RPC.
2232 Severity : enhancement
2234 Description: Bring 'lfs find' closer in line with regular Linux find.
2235 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2236 -print0 options and obtains all the needed info through the lustre
2239 Severity : enhancement
2241 Description: support up to 1024 configured devices on one node
2242 Details : change obd_dev array from statically allocated to dynamically
2243 allocated structs as they are first used to reduce memory usage
2248 Description: Flush dirty partially truncated pages during truncate
2249 Details : Immediatelly flush partially truncated pages in filter_setattr,
2250 this way we completely avoid having any pages in page cache on OST
2251 and can retire ugly workarounds during writes to flush such pages.
2256 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2257 Details : protect inode from truncation within vfs_unlink() context
2258 just take a reference before calling vfs_unlink() and release it
2259 when parent's i_sem is free.
2264 Description: last_id value checked outside lock on OST caused LASSERT failure
2265 Details : If there were multiple MDS->OST object precreate requests in
2266 flight, it was possible that the OST's last object id was checked
2267 outside a lock and incorrectly tripped an assertion. Move checks
2268 inside locks, and discard old precreate requests.
2271 Frequency : always, if extents are used on OSTs
2273 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2274 Details : OSTs using the extents format would not zero the high 16 bits of
2275 the index physical block number. This is not a problem for any
2276 OST filesystems smaller than 16TB, and no kernels support ext3
2277 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2278 new/modified files) and can be fixed for existing filesystems
2279 with e2fsprogs-1.39-cfs1.
2284 Description: import connection selection may be incorrect if timer wraps
2285 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2286 import connections to be ignored if the 32-bit jiffies counter
2287 wraps. Use a 64-bit jiffies counter.
2290 Frequency : very large clusters immediately after boot
2292 Description: LNET request buffers exhausted under heavy short-term load
2293 Details : If a large number of client requests are generated on a service
2294 that has previously never seen so many requests it is possible
2295 that the request buffer growth cannot keep up with the spike in
2296 demand. Instead of dropping incoming requests, they are held in
2297 the LND until the RPC service can accept more requests.
2300 Frequency : Sometimes during replay
2302 Description: Assertion failure in ll_local_open after replay.
2303 Details : If replay happened on an open request reply before we were able
2304 to set replay handler, reply will become not swabbed tripping the
2305 assertion in ll_local_open. Now we set the handler right after
2306 recognising of open request
2309 Frequency : very rare
2311 Description: kernel reports "badness in vsnprintf"
2312 Details : Reading from the "recovery_status" /proc file in small chunks
2313 may cause a negative length in lprocfs_obd_rd_recovery_status()
2314 call to vsnprintf() (which is otherwise harmless). Exit early
2315 if there is no more space in the output buffer.
2317 Severity : enhancement
2319 Description: clear OBD RPC statistics by writing to them
2320 Details : It is now possible to clear the OBD RPC statistics by writing
2321 to the "stats" file.
2326 Description: Client mtime is not the same on different clients after utimes
2327 Details : In some cases, the client was using the utimes() syscall on
2328 a file cached on another node. The clients now validate the
2329 ctime from the MDS + OSTs to determine which one is right.
2334 Description: Inability to activate failout mode
2335 Details : lconf script incorrectly assumed that in pythong string's numeric
2336 value is used in comparisons.
2339 Frequency : always with multiple stripes per file
2341 Description: Inefficient object allocation for mutli-stripe files
2342 Details : When selecting which OSTs to stripe files over, for files with
2343 a stripe count that divides evenly into the number of OSTs,
2344 the MDS is always picking the same starting OST for each file.
2345 Return the OST selection heuristic to the original design.
2350 Description: mount failures may take full timeout to return an error
2351 Details : Under some heavy load conditions it is possible that a
2352 failed mount can wait for the full obd_timeout interval,
2353 possibly several minutes, before reporting an error.
2354 Instead return an error as soon as the status is known.
2356 Frequency : quota enabled and large files being deleted
2358 Description: releasing more than 4GB of quota at once hangs OST
2359 Details : If a user deletes more than 4GB of files on a single OST it
2360 will cause the OST to spin in an infinite loop. Release
2361 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2366 Description: statfs data retrieved from /proc may be stale or zero
2367 Details : When reading per-device statfs data from /proc, in the
2368 {kbytes,files}_{total,free,avail} files, it may appear
2369 as zero or be out of date.
2372 Frequency : systems with MD RAID1 external journal devices
2374 Description: lconf's call to blkid is confused by RAID1 journal devices
2375 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2376 of returning all block devices that match the journal UUID.
2379 Frequency : always, for aggregate stripe size over 4GB
2381 Description: assertion fails when trying to use 4GB stripe size
2382 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2383 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2386 Frequency : always on ppc64
2388 Description: the first write on an ext3 filesystem with mballoc got stuck
2389 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2390 perform endianness conversion.
2392 ------------------------------------------------------------------------------
2394 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2396 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2397 INTEROPERATE with older versions automatically. Please read the
2398 user documentation before upgrading any part of a live system.
2399 * WARNING: Lustre networking configuration changes are required with
2400 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2403 * Support for kernels:
2404 2.6.9-22.0.2.EL (RHEL 4)
2405 2.6.5-7.244 (SLES 9)
2406 2.6.12.6 vanilla (kernel.org)
2409 Severity : enhancement
2410 Bugzilla : 7981/8208
2411 Description: Introduced Lustre Networking (LNET)
2412 Details : LNET is new networking infrastructure for Lustre, it includes
2413 a reorganized network configuration mode (see the user
2414 documentation for full details) as well as support for routing
2415 between different network fabrics. Lustre Networking Devices
2416 (LNDs) for the supported network fabrics have also been
2417 created for this new infrastructure.
2419 Severity : enhancement
2420 Description: Introduced Access control lists
2421 Details : clients can set ACLs on files and directories in order to have
2422 more fine-grained permissions than the standard Unix UGO+RWX.
2423 The MDS must be started with the "-o acl" mount option.
2425 Severity : enhancement
2426 Description: Introduced filesystem quotas
2427 Details : Administrators may now establish per-user quotas on the
2430 Severity : enhancement
2432 Description: Configuration change for the XT3
2433 The PTLLND is now used to run Lustre over Portals on the XT3
2434 The configure option(s) --with-cray-portals are no longer used.
2435 Rather --with-portals=<path-to-portals-includes> is used to
2436 enable building on the XT3. In addition to enable XT3 specific
2437 features the option --enable-cray-xt3 must be used.
2442 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2443 Details : A race between allocating a new llcd and re-getting the llcd_lock
2444 allowed another thread to grab newly-allocated llcd.
2446 Severity : enhancement
2448 Description: 2.6 OST async journal commit and locking fix to improve performance
2449 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2450 2.6 kernels are now async as they already were in 2.4 kernels so
2451 that they can commit concurrently with the network bulk transfer.
2452 For block-allocated files the filter allocation semaphore is held
2453 to avoid filesystem fragmentation during allocation. BKL lock
2454 removed for 2.6 xattr operations where it is no longer needed.
2459 Description: lconf incorrectly determined whether two IP networks could talk
2460 Details : In some more complicated routing and multiple-network
2461 configurations, lconf will avoid trying to make a network
2462 connection to a disjoint part of the IP space. It was doing the
2463 math incorrectly for one set of cases.
2468 Description: Fix for potential infinite loop processing records in an llog.
2469 Details : If an llog record is corrupted/zeroed, it is possible to loop
2470 forever in llog_process(). Validate the llog record length
2471 and skip the remainder of the block on error.
2474 Frequency : occasional (liblustre only)
2476 Description: liblustre could not open files whose last component is a symlink
2477 Details : sysio_path_walk() would incorrectly pass the open intent to
2478 intermediate path components.
2481 Frequency : rare (liblustre only with non-standard tuning)
2482 Bugzilla : 7201 (7350)
2483 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2484 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2485 after it has been released, only do this for referenced locks
2487 Severity : enhancement
2489 Description: specifying an (invalid) directory default stripe_size of -1
2490 would reset the directory default striping
2491 Details : stripe_size -1 was used internally to signal directory stripe
2492 removal, now use "all default" to signal dir stripe removal
2493 as a directory striping of "all default" is not useful
2496 Frequency : common for large clusters running liblustre clients
2498 Description: doing an ls when liblustre clients are running is slow
2499 Details : sending a glimpse AST to a liblustre client waits for every AST
2500 to time out, as liblustre clients will not respond. Since they
2501 cannot cache data we refresh the OST lock LVB from disk instead.
2503 Severity : enhancement
2505 Description: doing an ls at the same time as file IO can be slow
2506 Details : enqueue and other "small" requests can be blocked behind many
2507 large IO requests. Create a new OST IO portal for non-IO
2508 requests so they can be processed faster.
2511 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2513 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2514 created file with mode 000
2515 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2516 allow an ftruncate() as a non-root user to a file with mode 000.
2517 Lustre now respects this flag to disable mode checks when
2518 truncating a file owned by the user
2521 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2524 Description: Revoking locks from clients that went dead or catatonic might take
2526 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2527 cancellation of such locks instant on servers without waiting for
2528 any reply from clients. Clients drops these locks when cancel
2529 notification from server is received without replying.
2532 Frequency : liblustre-only, when liblustre client dies or becomes busy
2534 Description: Doing ls on Linux clients can take a long time with active
2536 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2537 granting such locks to it in the first place if possible. Locks
2538 are taken by proxy on the OST during the read or write and
2539 dropped immediately afterward. Add connect flags handling, do
2540 not grant locks to liblustre clients for glimpse ASTs.
2542 Severity : enhancement
2544 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2545 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2546 sequential reads and behaves badly with applications doing random
2547 reads. Improve it by reading ahead at least read region, and
2548 avoiding excessive large RPC for small reads.
2550 Severity : enhancement
2552 Description: Creating more than 1000 files for a single job may cause a load
2553 imbalance on the OSTs if there are also a large number of OSTs.
2554 Details : qos_prep_create() uses an OST index reseed value that is an
2555 even multiple of the number of available OSTs so that if the
2556 reseed happens in the middle of the object allocation it will
2557 still utilize the OSTs as uniformly as possible.
2562 Description: OST or MDS may oops in ping_evictor_main()
2563 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2564 but doesn't restart at beginning of obd_exports_timed list
2567 Severity : enhancement
2569 Description: improve by-nid export eviction on the MDS and OST
2570 Details : allow multiple exports with the same NID to be evicted at one
2571 time without re-searching the exports list.
2574 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2576 Description: MDS may oops in groups_free()
2577 Details : in rare race conditions a newly allocated group_info struct is
2578 freed again, and this can be NULL. The 2.4 compatibility code
2579 for groups_free() checked for a NULL pointer, but 2.6 did not.
2582 Frequency : common for liblustre clients doing little filesystem IO
2583 Bugzilla : 9352, 7313
2584 Description: server may evict liblustre clients accessing contended locks
2585 Details : if a client is granted a lock or receives a completion AST
2586 with a blocking AST already set it would not reply to the AST
2587 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2590 Frequency : lfs setstripe, only systems with more than 160 OSTs
2592 Description: unable to set striping with a starting offset beyond OST 160
2593 Details : llapi_create_file() incorrectly limited the starting stripe
2594 index to the maximum single-file stripe count.
2597 Frequency : LDAP users only
2599 Description: lconf did not handle in-kernel recovery with LDAP properly
2600 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2602 Severity : enhancement
2604 Description: bind OST threads to NUMA nodes to improve performance
2605 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2606 node and do their allocations there to localize memory access
2608 Severity : enhancement
2610 Description: llmount can determine client NID directly from Myrinet (GM)
2611 Details : the client NID code from gmnalnid was moved directly into
2612 llmount, removing the need to use this or specifying the
2613 client NID explicitly when mounting GM clients with zeroconf
2616 Frequency : if client is started with down MDS
2618 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2619 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2620 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2621 ptlrpc_import_delayed_req() because we didn't check if the
2622 request was interrupted, and we also didn't break out of the
2623 event loop if there was no timeout
2628 Description: data loss during non-page-aligned writes to a single file from
2629 both multiple nodes and multiple threads on one node at same time
2630 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2631 inconsistency led to false short-reads, that were cached and later
2632 used by ->prepare_write() to fill in partially written page,
2633 leading to data loss.
2636 Frequency : always, if lconf --abort_recovery used
2638 Description: lconf --abort_recovery fails with 'Operation not supported'
2639 Details : lconf was attempting to abort recovery on the MDT device and not
2642 Severity : enhancement
2644 Description: remove cleanup logs
2645 Details : replace lconf-generated cleanup logs with lustre internal
2646 cleanup routines. Eliminates the need for client-cleanup and
2649 Severity : enhancement
2651 Description: add support for EAs (user and system) on lustre filesystems
2652 Details : it is now possible to store extended attributes in the Lustre
2653 client filesystem, and with the user_xattr mount option it
2654 is possible to allow users to store EAs on their files also
2656 Severity : enhancement
2658 Description: Add possibility (config option) to show minimal available OST free
2660 Details : When compiled with --enable-mindf configure option, statfs(2)
2661 (and so, df) will return least minimal free space available from
2662 all OSTs as amount of free space on FS, instead of summary of
2663 free spaces of all OSTs.
2665 Severity : enhancement
2667 Description: do not expand extent locks acquired on OST-side
2668 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2669 by server: they are not cached anyway.
2672 Frequency : when mmap is used/binaries executed from Lustre
2674 Description: Unmmap pages before throwing them away from read cache.
2675 Details : llap_shrink cache now attempts to unmap pages before discarding
2676 them (if unmapping failed - do not discard). SLES9 kernel has
2677 extra checks that trigger if this unmapping is not done first.
2682 Description: lconf didn't resolve symlinks before checking to see whether a
2683 given mountpoint was already in use
2686 Frequency : when migrating failover services
2687 Bugzilla : 6395, 9514
2688 Description: When migrating a subset of services from a node (e.g. failback
2689 from a failover service node) the remaining services would
2690 time out and evict clients.
2691 Details : lconf --force (implied by --failover) sets the global obd_timeout
2692 to 5 seconds in order to quickly disconnect, but this caused
2693 other RPCs to time out too quickly. Do not change the global
2694 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2696 Severity : enhancement
2697 Frequency : if MDS is started with down OST
2698 Bugzilla : 9439,5706
2699 Description: Allow startup/shutdown of an MDS without depending on the
2700 availability of the OSTs.
2701 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2702 Add appropriate locking and lov-osc refcounts for safe
2703 cleaning. Add osc abort_inflight calls in case the
2704 synchronize never started.
2707 Frequency : occasional (Cray XT3 only)
2709 Description: root not authorized to access files in CRAY_PORTALS environment
2710 Details : The client process capabilities were not honoured on the MDS in
2711 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2712 been accessed by an authorized user then root was able to access
2713 the file on the local client also. The root user capabilities
2714 are now allowed on the MDS, as this environment has secure UID.
2717 Frequency : occasional
2719 Description: ldiskfs "too long searching" message happens too often
2720 Details : A debugging message (otherwise harmless) prints too often on
2721 the OST console. This has been reduced to only happen when
2722 there are fragmentation problems on the filesystem.
2727 Description: Division by zero in statfs when all OSCs are inactive
2728 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2729 lov_statfs divides by value returned by lov_get_stripecnt().
2733 Bugzilla : 9489, 3273
2734 Description: First write from each client to each OST was only 4kB in size,
2735 to initialize client writeback cache, which caused sub-optimal
2736 RPCs and poor layout on disk for the first writen file.
2737 Details : Clients now request an initial cache grant at (re)connect time
2738 and so that they can start streaming writes to the cache right
2739 away and always do full-sized RPCs if there is enough data.
2740 If the OST is rebooted the client also re-establishes its grant
2741 so that client cached writes will be honoured under the grant.
2746 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2747 Details : Now I/O RPCs go to different portal number and (presumably) fast
2748 lock requests (and glimses) and other RPCs get their own service
2749 threads pool that should be able to service those RPCs
2752 Severity : enhancement
2754 Description: Ability to exchange lustre version between client and servers and
2755 issue warnings at client side if client is too old. Also for
2756 liblustre clients there is ability to refuse connection of too old
2758 Details : New 'version' field is added to connect data structure that is
2759 filled with version info. That info is later checked by server and
2763 Frequency : rare, liblustre only.
2764 Bugzilla : 9296, 9581
2765 Description: Two simultaneous writes from liblustre at offset within same page
2766 might proceed at the same time overwriting eachother with stale
2768 Details : I/O lock withing llu_file_prwv was released too early, before data
2769 actually was hitting the wire. Extended lock-holding time until
2770 server acknowledges receiving data.
2773 Frequency : extremely rare. Never observed in practice.
2775 Description: avoid generating lustre_handle cookie of 0.
2776 Details : class_handle_hash() generates handle cookies by incrementing
2777 global counter, and can hit 0 occasionaly (this is unlikely, but
2778 not impossible, because initial value of cookie counter is
2779 selected randonly). Value of 0 is used as a sentinel meaning
2780 "unassigned handle" --- avoid it. Also coalesce two critical
2781 sections in this function into one.
2783 Severity : enhancement
2785 Description: allow liblustre clients to delegate truncate locking to OST
2786 Details : To avoid overhead of locking, liblustre client instructs OST to
2787 take extent lock in ost_punch() on client's behalf. New connection
2788 flag is added to handle backward compatibility.
2790 Severity : enhancement
2791 Bugzilla : 4928, 7341, 9758
2792 Description: allow number of OST service threads to be specified
2793 Details : a module parameter allows the number of OST service threads
2794 to be specified via "options ost ost_num_threads={N}" in the
2795 OSS's /etc/modules.conf or /etc/modprobe.conf.
2799 Bugzilla : 6146, 9635, 9895
2800 Description: servers crash with bad pointer in target_handle_connect()
2801 Details : In rare cases when a client is reconnecting it was possible that
2802 the connection request was the last reference for that export.
2803 We would temporarily drop the export reference and get a new
2804 one, but this may have been the last reference and the export
2805 was just destroyed. Get new reference before dropping old one.
2807 Severity : enhancement
2808 Frequency : if client is started with failover MDS
2810 Description: Allow multiple MDS hostnames in the mount command
2811 Details : Try to read the configuration from all specified MDS
2812 hostnames during a client mount in case the "primary"
2815 Severity : enhancement
2817 Description: Stop sending data to evicted clients as soon as possible.
2818 Details : Check if the client we are about to send or are sending data to
2819 was evicted already. (Check is done every second of waiting,
2820 for which l_wait_event interface was extended to allow checking
2821 of exit condition at specified intervals).
2824 Frequency : rare, normally only when NFS exporting is done from client
2826 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2828 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2831 Frequency : always, if srand() is called before liblustre initialization
2833 Description: Liblustre uses system PRNG disturbing its usage by user application
2834 Details : Introduce internal to lustre fast and high-quality PRNG for
2835 lustre usage and make liblustre and some other places in generic
2836 lustre code to use it.
2838 Severity : enhancement
2839 Bugzilla : 9477, 9557, 9870
2840 Description: Verify that the MDS configuration logs are updated when xml is
2841 Details : Check if the .xml configuration logs are newer than the config
2842 logs stored on the MDS and report an error if this is the case.
2843 Request --write-conf, or allow starting with --old_conf.
2845 Severity : enhancement
2847 Description: Handle symlinks in the path when checking if Lustre is mounted.
2848 Details : Resolve intermediate symlinks when checking if a client has
2849 mounted a filesystem to avoid duplicate client mounts.
2854 Description: lconf can hit an error exception but still return success.
2855 Details : The lconf command catches the Command error exception at the top
2856 level script context and will exit with the associated exit
2857 status, but doesn't ensure that this exit status is non-zero.
2862 Description: failure of ptlrpc thread startup can cause oops
2863 Details : Starting a ptlrpc service thread can fail if there are a large
2864 number of threads or the server memory is very fragmented.
2865 Handle this without oopsing.
2868 Frequency : always, only if liblustre and non-default acceptor port was used
2870 Description: liblustre cannot connect to servers with non-default acceptor port
2871 Details : tcpnal_set_default_params() was not called and was therefore
2872 ignoring the environment varaible TCPNAL_PORT, as well as other
2873 TCPNAL_ environment variables
2878 Description: two objects could be created on the same OST for a single file
2879 Details : If an OST is down, in some cases it was possible to create two
2880 objects on a single OST for a single file. No problems other
2881 than potential performance impact and spurious error messages.
2885 Bugzilla : 5681, 9562
2886 Description: Client may oops in ll_unhash_aliases
2887 Details : Client dcache may become inconsistent in race condition.
2888 In some cases "getcwd" can fail if the current directory is
2894 Description: Inode refcounting problems in NFS export code
2895 Details : link_raw functions used to call d_instantiate without obtaining
2896 extra inode reference first.
2900 Bugzilla : 9942, 9903
2901 Description: Referencing freed requests leading to crash, memleaks with NFS.
2902 Details : We used to require that call to ll_revalidate_it was always
2903 followed by ll_lookup_it. Also with revalidate_special() it is
2904 possible to call ll_revalidate_it() twice for the same dentry
2905 even if first occurence returned success. This fix changes semantic
2906 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2907 reference on a request referred from the intent.
2908 ll_intent_release() then releases such a request.
2911 Frequency : rare, normally benchmark loads only
2913 Description: unlinked inodes were kept in memory on the client
2914 Details : If a client is repeatedly creating and unlinking files it
2915 can accumulate a lot of stale inodes in the inode slab cache.
2916 If there is no other client load running this can cause the
2917 client node to run out of memory. Instead flush old inodes
2918 from client cache that have the same inode number as a new inode.
2921 Frequency : SLES9 2.6.5 kernel and long filenames only
2922 Bugzilla : 9969, 10379
2923 Description: utime reports stale NFS file handle
2924 Details : SLES9 uses out-of-dentry names in some cases, which confused
2925 the lustre dentry revalidation. Change it to always use the
2929 Frequency : rare, unless heavy write-truncate concurrency is continuous
2930 Bugzilla : 4180, 6984, 7171, 9963, 9331
2931 Description: OST becomes very slow and/or deadlocked during object unlink
2932 Details : filter_destroy() was holding onto the parent directory lock
2933 while truncating+unlinking objects. For very large objects this
2934 may block other threads for a long time and slow overall OST
2935 responsiveness. It may also be possible to get a lock ordering
2936 deadlock in this case, or run out of journal credits because of
2937 the combined truncate+unlink. Solution is to do object truncate
2938 first in one transaction without parent lock, and then do the
2939 final unlink in a new transaction with the parent lock. This
2940 reduces the lock hold time dramatically.
2943 Frequency : rare, 2.4 kernels only
2945 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2946 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2947 before unmounting their filesystem in order to avoid deadlock.
2948 The kernel_locked() function in 2.4 kernels only checks whether
2949 the kernel lock is held, not whether it is this process that is
2950 holding it as 2.6 kernels do.
2955 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2956 Details : The client ptlrpc code may be trying to reconnect to a down
2957 server before a previous connection attempt has timed out.
2958 Increase the reconnect interval to be longer than the connection
2959 timeout interval to avoid sending duplicate connections to
2963 Frequency : echo_client brw_test command
2965 Description: fix echo_client to work with OST preallocated code
2966 Details : OST preallocation code (5137) didn't take echo_client IO path
2967 into account: echo_client calls filter methods outside of any
2968 OST thread and, hence, there is no per-thread preallocated
2969 pages and buffers to use. Solution: hijack pga pages for IO. As
2970 a byproduct, this avoids unnecessary data copying.
2974 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2975 Description: Client can oops in mdc_commit_close() after open replay
2976 Details : It was possible for the MDS to return an open request with no
2977 transaction number in mds_finish_transno() if the client was
2978 evicted, but without actually returning an error. Clients
2979 would later try to replay that open and may trip an assertion
2980 Simplify the client close codepath, and always return an error
2981 from the MDS in case the open is not successful.
2984 Frequency : rare, 2.6 OSTs only
2986 Description: OST may deadlock under high load on fragmented files
2987 Details : If there was a heavy load and highly-fragmented OST filesystems
2988 it was possible to have all the OST threads deadlock waiting on
2989 allocation of biovecs, because the biovecs were not released
2990 until the entire RPC IO was completed. Instead, release biovecs
2991 as soon as they are complete to ensure forward IO progress.
2993 Severity : enhancement
2995 Description: Support for specifying external journal device at mount
2996 Details : If an OST or MDS device is formatted with an external journal
2997 device, this device major/minor is stored in the ext3 superblock
2998 and may not be valid for failover. Allow detecting and
2999 specifying the external journal at mount time.
3004 Description: Mounting an MDS with pending unlinked files may cause oops
3005 Details : target_finish_recovery() calls mds_postrecov() which returned
3006 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3007 considers this an error and immediately begins cleaning up the
3008 lov, just after starting the mds_lov process
3010 Severity : enhancement
3012 Description: Implement 'lfs df' to report actual free space on per-OST basis
3013 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3014 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3015 usage of objects; '-h' to report in human readable format.
3017 ------------------------------------------------------------------------------
3019 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3026 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3027 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3028 at startup during orphan unlinking. Instead, skip buddy update
3029 before setup, as it will be regenerated after recovery is complete.
3034 Description: If an OST is inactive, its locks might reference stale inodes.
3035 Details : lov_change_cbdata() must iterate over all namespaces, even if
3036 they are inactive to clear inode references from the lock.
3038 Severity : enhancement
3039 Frequency : occasional, if non-standard max_dirty_mb used
3041 Description: Client will block write RPCs if not enough grant
3042 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3043 then the client will block writes while waiting for another RPC
3044 to complete instead of consuming its dirty limit. With change
3045 we get improved performance when max_dirty_mb is small.
3047 Severity : enhancement
3048 Bugzilla : 3389, 6253
3049 Description: Add support for supplementary groups on the MDS.
3050 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3051 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3052 lookups for user supplementary groups into a cache.
3056 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3057 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3058 for liblustre O_EXCL is enforced by MDS.
3062 Description: readdir never returns NULL in liblustre.
3063 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3064 dentry in struct dirent.
3068 Frequency : liblustre only, depends on application IO pattern
3069 Description: liblustre clients evicted if not contacting servers
3070 Details : Don't put liblustre clients into the ping_evictor list, so
3071 they will not be evicted by the pinger ever.
3073 Severity : enhancement
3075 Description: Add ability to evict clients by NID from MDS.
3076 Details : By echoing "nid:$NID" string into
3077 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3078 $NID would be instantly evicted from this MDS and from all active
3079 OSTs connected to it.
3083 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3084 Details : lookup_it_finish() used to query file size from OSTs that was not
3089 Description: service threads change working directory to that of init
3090 Details : Starting lustre service threads may pin the working directory
3091 of the parent thread, making that filesystem busy. Threads
3092 now change to the working directory of init to avoid this.
3096 Frequency : during shutdown only
3097 Description: shutdown with a failed MDS or OST can cause unmount to hang
3098 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3103 Frequency : 2.6 only
3104 Description: chmod/chown may include an extra supplementary group
3105 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3106 supplementary group and if none is specified this is used.
3109 Bugzilla : 5479 (6816)
3111 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3112 Details : If lookup is in progress on a file that is unlinked we might try
3113 to revalidate the inode and fail in revalidate after lookup is
3114 complete and ll_file_open() enqueues the open again but
3115 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3118 Frequency : always, if lconf --abort_recovery used
3120 Description: lconf --abort_recovery fails with 'Operation not supported'
3121 Details : lconf was attempting to abort recovery on the MDT device and not
3124 ------------------------------------------------------------------------------
3126 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3131 Frequency : rare (only unsupported configurations with a node running as an
3133 Bugzilla : 6514, 5137
3134 Description: Mounting a Lustre file system on a node running as an OST could
3136 Details : OSTs now preallocates memory needed to write out data at
3137 startup, instead of when needed, to avoid having to
3138 allocate memory in possibly low memory situations.
3139 Specifically, if the file system is mounted on on OST,
3140 memory pressure could force it to try to write out data,
3141 which it needed to allocate memory to do. Due to the low
3142 memory, it would be unable to do so and the node would
3143 become unresponsive.
3145 Severity : enhancement
3147 Description: Addition of lconf --service command line option
3148 Details : lconf now accepts a '--service <arg>' option, which is
3149 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3151 Severity : enhancement
3153 Description: Failover mode is now the default for OSTs.
3154 Details : By default, OSTs will now run in failover mode. To return to
3155 the old behaviour, add '--failout' to the lmc line for OSTs.
3157 Severity : enhancement
3159 Description: Health checks are now provided for MDS and OSTs
3160 Details : Additional detailed health check information on MSD and OSTs
3161 is now provided through the procfs health_check value.
3164 Frequency : occasional, depends on IO load
3166 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3167 after numerous create/delete cycles
3168 Details : The ext3 inode allocation policy would not allocate new inodes
3169 very well on the OSTs because there are no new directories
3170 being created. Instead we look for groups with free space if
3171 the parent directories are nearly full.
3176 Description: Network or server problems during mount may cause partially
3177 mounted clients instead of returning an error.
3178 Details : The config llog parsing code may overwrite the error return
3179 code during mount error handling, returning success instead
3185 Description: MDS can fail to allocate large reply buffers
3186 Details : After long uptimes the MDS can fail to allocate large reply
3187 buffers (e.g. zconf client mount config records) due to memory
3188 fragmentation or consumption by the buffer cache. Preallocate
3189 some large reply buffers so that these replies can be sent even
3190 under memory pressure.
3194 Frequency : rare (liblustre)
3195 Description: fsx running with liblustre complained that using truncate() to
3196 extend the file doesn't work. This patch corrects that issue.
3197 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3198 ATTR_SIZE and lsm use in llu_setattr_raw.
3202 Frequency : rare, only 2.6 kernels
3203 Description: Unusual file access patterns on the MDS may result in inode
3204 data being lost in very rare circumstances.
3205 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3206 code in the 2.6 kernel caused the inode and/or EA data not to
3207 be read from disk, causing single-file corruption.
3211 Frequency : rare, only 2.6 filesystems using extents
3212 Description: Heavy concurrent write and delete load may cause data corruption.
3213 Details : It was possible under high-load situations to have an extent
3214 metadata block in the block device cache from a just-unlinked
3215 file overwrite a newly-allocated data block. We now unmap any
3216 metadata buffers that alias just-allocated data blocks.
3220 Frequency : filesystems with default stripe_count larger than 77
3221 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3222 Details : lconf specifies an inode size of 4096 bytes when the default
3223 stripe_count is larger than 77. This conflicts with the default
3224 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3225 this case to avoid pinning too much memory for large EAs.
3227 ------------------------------------------------------------------------------
3229 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3234 Frequency : rare (extremely heavy IO load with hundreds of clients)
3236 Description: Client is evicted, gets IO error writing to file
3237 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3238 caused clients to be evicted for AST timeouts. The fixes in
3239 bug 5192 mean we no longer need to have such short AST timeouts
3240 so ldlm_timeout has been increased.
3243 Frequency : occasional during --force or --failover shutdown under load
3244 Bugzilla : 5949, 4834
3245 Description: Server oops/LBUG if stopped with --force or --failover under load
3246 Details : a collection of import/export refcount and cleanup ordering
3247 issues fixed for safer force cleanup
3250 Frequency : only filesystems larger than 120 OSTs
3251 Bugzilla : 5990, 6223
3252 Description: lfs getstripe would oops on a very large filesystem
3253 Details : lov_getconfig used kfree on vmalloc'd memory
3256 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3257 Bugzilla : 6242, 6243
3258 Description: reading from files that had been truncated to a non-zero size
3259 but never opened returned no data
3260 Details : ll_file_read() reads zeros from no-object files to EOF
3265 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3266 Details : There's little harm in aborting MDS/OSS recovery and letting it
3267 try again, so I removed the LASSERT and return an error instead.
3269 Severity : enhancement
3271 Description: New debugging infrastructure for tracking down data corruption
3272 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3273 (b) cover more of the client-side code path, and (c) try to narrow
3274 down where problems occurred
3278 Bugzilla : 3819, 4364, 4397, 6313
3279 Description: Racing close and eviction MDS could cause assertion in mds_close
3280 Details : It was possible to get multiple mfd references during close and
3281 client eviction, leading to one thread referencing a freed mfd.
3283 Severity: : enhancement
3284 Bugzilla : 3262, 6359
3285 Description: Attempts to reconnect to servers are now more aggressive.
3286 Details : This builds on the enhanced upcall-less recovery that was added
3287 in 1.4.2. When trying to reconnect to servers, clients will
3288 now try each server in the failover group every 10 seconds. By
3289 default, clients would previously try one server every 25 seconds.
3294 Description: After recovery, certain operations trigger a failed
3295 assertion on a client.
3296 Details : Failing over an mds, using lconf -d --failover, while a
3297 client was doing a readdir() call would cause the client to
3298 LBUG after recovery completed and the readdir() was resent.
3300 Severity : enhancement
3302 Description: Default groups are now added by lconf
3303 Details : You can now run lconf --group <servicename> without having to
3304 manually add groups with lmc.
3307 Frequency : occasional
3309 Description: Nodes with an elan id of 0 trigger a failed assertion
3312 Frequency : always when accessing e.g. tty/console device nodes
3314 Description: tty and some other devices nodes cannot be used on lustre
3315 Details : file's private_data field is used by device data and lustre
3316 values in there got lost. New field was added to struct file to
3317 store fs-specific private data.
3320 Frequency : when exporting Lustre via NFS
3322 Description: NFSD failed occasionally when looking up a path component
3323 Details : NFSD is looking up ".." which was broken in ext3 directories
3324 that had grown large enough to become hashed.
3327 Frequency : Clusters with multiple interfaces not on the same subnet
3329 Description: Nodes will repeatedly try to reconnect to an interface which it
3330 cannot reach and report an error to the log.
3331 Details : Extra peer list entries will be created by lconf with some peers
3332 unreachable. lconf now validates the peer before adding it.
3335 Frequency : Only if a default stripe is set on the filesystem root.
3337 Description: Setting a default stripe on the filesystem root prevented the
3338 filesystem from being remounted.
3339 Details : The client was sending extra request flags in the root getattr
3340 request and did not allocate a reply buffer for the dir EA.
3343 Frequency : occasional, higher if lots of files are accessed by one client
3344 Bugzilla : 6159, 6097
3345 Description: Client trips assertion regarding lsm mismatch/magic
3346 Details : While revalidating inodes the VFS looks up inodes with ifind()
3347 and in rare cases can find an inode that is being freed.
3348 The ll_test_inode() code will free the lsm during ifind()
3349 when it finds an existing inode and then the VFS later attaches
3350 this free lsm to a new inode.
3354 Bugzilla : 6422, 7030
3355 Description: MDS deadlock between mkdir and client eviction
3356 Details : Creating a new file via mkdir or mknod (starting a transaction
3357 and getting the ns lock) can deadlock with client eviction
3358 (gets ns lock and trying to finish a synchronous transaction).
3361 Frequency : occasional
3362 Description: While starting a server, the fsfilt_ext3 module could not be
3364 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3365 kernels. Previously, lconf would still use the ext3 name
3366 when trying to load modules. Now, it will correctly use
3367 ext3 on 2.4 and ldiskfs on 2.6.
3369 Severity : enhancement
3370 Description: The default stripe count has been changed to 1
3371 Details : The interpretation of the default stripe count (0, to lfs
3372 or lmc) has been changed to mean striping across a single
3373 OST, rather than all available. For general usage we have
3374 found a stripe count of 1 or 2 works best.
3376 Severity : enhancement
3377 Description: Add support for compiling against Cray portals.
3378 Details : Conditional compiling for some areas that are different
3382 Frequency : occasional
3383 Bugzilla : 6409, 6834
3384 Description: Creating files with an explicit stripe count may lead to
3385 a failed assertion on the MDS
3386 Details : If some OSTs are full or unavailable, creating files may
3387 trigger a failed assertion on the MDS. Now, Lustre will
3388 try to use other servers or return an error to the
3392 Frequency : occasional
3394 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3395 caused lock timeout during readahead (since 1.4.2).
3396 Details : Processes doing readahead might match a lock that hasn't been
3397 granted yet if there are overlapping and conflicting lock
3398 requests. The readahead process waits on ungranted lock
3399 (original lock is CBPENDING), while OST waits for that process
3400 to cancel CBPENDING read lock and eventually evicts client.
3402 Severity : enhancement
3404 Description: Initial enabling of flock support for clients
3405 Details : Implements fcntl advisory locking and file status functions.
3406 This feature is provided as an optional mount flag (default
3407 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3408 locking are implemented yet, and those that are are not guaranteed
3409 to be completely correct in production environments.
3410 mount -t lustre -o [flock|noflock] ...
3413 Frequency : occasional
3415 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3416 assertion in the ext3 JBD (journaling) layer.
3417 Details : The b_committed_data struct is protected by the big kernel lock
3418 in 2.4 kernels, serializing journal_commit_transaction() and
3419 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3420 access to this struct. In 2.6 kernels there is finer grained
3421 locking to improve SMP performance of the JBD layer.
3425 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3430 Description: merge and cleanup kernel patches.
3433 -----------------------------------------------------------------------------
3435 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3437 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3438 versions of Lustre. Please update all servers and clients to
3439 version 1.4.2 or later at the same time. You must also run
3440 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3441 to update the configuration logs.
3443 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3444 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3445 - swab llog records of type '0' so we get proper header size/idx (5861)
3446 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3447 - fix rename of one directory over another leaking an inode (5953)
3448 - avoid SetPageDirty on 2.6 (5981)
3449 - don't re-add just-being-destroyed locks to the waiting list (5653)
3450 - when creating new directories, inherit the parent's custom
3451 striping settings if present parent (3048)
3452 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3453 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3454 - don't hold client locks on temporary worklist from l_lru (5666)
3455 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3456 - automatically evict dead clients (5921)
3457 - Update file size properly in create+truncate+fstat case (6196)
3458 - Do not unhash mountpoint dentries, do not allow removal of
3460 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3461 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3462 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3463 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3464 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3465 - Correctly handle reads of files with no objects (6243)
3466 - lctl recover will also mark a device active if deactivate used (5933)
3468 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3469 - allow --write-conf on an MDS with different nettype than client (5619)
3470 - don't write config llogs to MDS for mounts not from that MDS (5617)
3471 - lconf should create multiple TCP connections from a client (5201)
3472 - init scripts are now turned off by default; run chkconfig --on
3473 lustre and chkconfig --on lustrefs to use them
3474 - upcalls are no longer needed for clients to recover to failover
3476 - add --abort-recovery option to lconf to abort recovery on device
3478 - add support for an arbitrary number of OSTs (3026)
3479 - Quota support protocol changes.
3480 - forward compatibility changes to wire structs (6007)
3481 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3482 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3483 - improved "lustre status" script
3484 - initialize blocksize for non-regular files (6062)
3485 - added --disable-server and --disable-client configure options (5782)
3486 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3487 - Vanilla 2.4.29 support
3488 - increase maximum number of obd devices to 520 (6242)
3489 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3490 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3491 SLES 9 2.6.5 kernels
3492 - limit stripes per file to 160 (the maximum EA size) (6093)
3494 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3497 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3498 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3499 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3500 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3501 - small ext3 extents cleanups and fixes (5733)
3502 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3503 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3504 - avoid needless client->OST connect, fix handle mismatch (5317)
3505 - fix DLM error path that led to out-of-sync client, long delays (5779)
3506 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3507 - fix several locking issues related to i_size (5492,5624,5654,5672)
3508 - don't move pending lock onto export if it is already evicted (5683)
3509 - fix kernel oops when creating .foo in unlinked directory (5548)
3510 - fix deadlock in obdfilter statistics vs. object create (5811)
3511 - use time_{before,after} to avoid timer jiffies wrap (5882)
3512 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3513 - Do not leak request if server was not able to process it (5154)
3514 - If mds_open unable to find parent dir, make that negative lookup(5154)
3515 - don't create new directories with extent-mapping (5909, 5936)
3517 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3518 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3519 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3520 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3521 - support for RHEL4 kernel on i686 (5773)
3522 - provide error messages when incompatible logs are encountered (5898)
3524 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3525 * version 1.4.0.10 (1.4.1 release candidate 1)
3527 - don't keep a lock reference when lock is not granted (4238)
3528 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3529 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3530 - drop import inflight refcount on signal_completed_replay error (5255)
3531 - unlock page after async write error during send (3677)
3532 - handle missing objects in filter_preprw_read properly (5265)
3533 - no transno return for symlink open, don't save no-trasno open (3440)
3534 - don't try to complete elan receive that already failed (4012)
3535 - free RPC server reply state on error (5406)
3536 - clean up thread from ptlrpc_start_thread() on error (5160)
3537 - readahead could read extra page into cache that wasn't ejected (5388)
3538 - prevent races in class_attach/setup/cleanup/detach (5260)
3539 - don't dereference de->d_inode after l_dput of de (5458)
3540 - use "int" for stripe value returned from lock_to_stripe (5544)
3541 - mballoc allocation and error-checking fixes in 2.6 (5504)
3542 - block device patches to fix I/O request sizes in 2.6 (5482)
3543 - look up hostnames for IB nals (5602)
3544 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3545 - don't start multiple acceptors for the same port (5277)
3546 - fix incorrect LASSERT in mds_getattr_name (5635)
3547 - export a proc file for general "ping" checking (5628)
3548 - fix "lfs check" to not block when the MDS is down (5628)
3550 - service request history (4965)
3551 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3552 - create an "evict_client" /proc entry on OSTs, like the MDS has
3553 - fix mount usage message, return errors per mount(8) (5168)
3554 - change grep [] to grep "[]" in tests so they work in more UMLs
3555 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3556 - remove ancient LOV_MAGIC_V0 EA support (5047)
3557 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3558 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3559 - allow readahead allocations to fail when low on memory (5383)
3560 - mmap locking landed again, after considerable improvement (2828)
3561 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3563 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3566 - send OST transaction number in read/write reply to free req (4966)
3567 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3568 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3569 - the watchdog thread now runs as interruptible (5246)
3570 - flock/lockf fixes (but it's still disabled, pending 5135)
3571 - don't use EXT3 constants in llite code (5094)
3572 - memory shortage at startup could cause assertion (5176)
3574 - reorganization of lov code
3575 - single portals codebase
3577 - add extents/mballoc support (5025)
3578 - direct I/O reads in the obdfilter (4048)
3579 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3581 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3584 - send OST transaction number in read/write reply to free req (4966)
3585 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3586 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3587 - the watchdog thread now runs as interruptible (5246)
3588 - handle missing objects in filter_preprw_read properly (5265)
3589 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3590 - drop import inflight refcount on signal_completed_replay error (5255)
3591 - unlock page after async write error during send (3677)
3592 - return original error code on reconstructed replies (3761)
3593 - no transno return for symlink open, don't save no-trasno open (3440)
3595 - add pid to ldlm debugging output (4922)
3596 - bump the watchdog timeouts -- we can't handle 30sec yet
3597 - extra debugging for orphan dentry/inode bug (5259)
3599 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3602 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3603 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3604 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3605 - fix NULL dereference in /proc/sys/portals/routes (4827)
3606 - allow failed mdc_close() operations to be interrupted (4561)
3607 - stop precreate on OST before MDS would time out on it (4778)
3608 - don't send partial-page writes before EOF from client (4410)
3609 - discard client grant for sub-page writes on large-page clients (4520)
3610 - don't free dentries not owned by NFS code, check generation (4806)
3611 - fix lsm leak if mds_create_objects() fails (4801)
3612 - limit debug_daemon file size, always print CERROR messages (4789)
3613 - use transno after validating reply (3892)
3614 - process timed out requests if import state changes (3754)
3615 - update mtime on OST during writes, return in glimpse (4829)
3616 - add mkfsoptions to LDAP (4679)
3617 - use ->max_readahead method instead of zapping global ra (5039)
3618 - don't interrupt __l_wait_event() during strace
3620 - add software watchdogs to catch hung threads quickly (4941)
3621 - make lustrefs init script start after nfs is mounted
3622 - fix CWARN/ERROR duplication (4930)
3623 - return async write errors to application if possible (2248)
3624 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3625 - print NAL number in %x format (4645)
3626 - update barely-supported suse-2.4.21-171 series (4842)
3627 - support for sles 9 %post scripts
3628 - support for building 2.6 kernel-source packages
3629 - support for sles km_* packages
3631 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3634 - ignore -ENOENT errors in osc_destroy (3639)
3635 - notify osc create thread that OSC is being cleaned up (4600)
3636 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3637 - reconstruct ost_handle() like mds_handle() (4657)
3638 - create a new thread to do import eviction to avoid deadlock (3969)
3639 - let lconf resolve symlinked-to devices (4629)
3640 - don't unlink "objects" from directory with default EA (4554)
3641 - hold socknal file ref over connect in case target is down (4394)
3642 - allow more than 32000 subdirectories in a single directory (3244)
3643 - fix blocks count for O_DIRECT writes (3751)
3644 - OST returns ENOSPC from object create when no space left (4539)
3645 - don't send truncate RPC if file size isn't changing (4410)
3646 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3647 - bind to privileged port in socknal and tcpnal (3689)
3649 - rate limit CERROR/CWARN console message to avoid overload (4519)
3650 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3651 - basic mmap support (3918)
3652 - kernel patch series update from b1_4 (4711)
3654 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3657 - avoid crash during MDS cleanup with OST shut down (2775)
3658 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3659 - don't use bad inodes on the MDS (3744)
3660 - dynamic object preallocation to improve recovery speed (4236)
3661 - don't hold spinlock over lock dumping or change debug flags (4401)
3662 - don't zero obd_dev when it is force cleaned (3651)
3663 - print grants to console if they go negative (4431)
3664 - "lctl deactivate" will stop automatic recovery attempts (3406)
3665 - look for existing locks in ldlm_handle_enqueue() (3764)
3666 - don't resolve lock handle twice in recovery avoiding race (4401)
3667 - revalidate should check working dir is a directory (4134)
3669 - don't always mark "slow" obdfilter messages as errors (4418)
3671 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3674 - don't close LustreDB during write_conf until it is done (3860)
3675 - fix typo in lconf for_each_profile (3821)
3676 - allow dumping logs from multiple threads at one time (3820)
3677 - don't allow multiple threads in OSC recovery (3812)
3678 - fix debug_size parameters (3864)
3679 - fix mds_postrecov to initialize import for llog ctxt (3121)
3680 - replace config semaphore with spinlock (3306)
3681 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3682 - don't allow enqueue to complete on a destroyed export (3822)
3683 - down write_lock before checking llog header bitmap (3825)
3684 - recover from lock replay timeout (3764)
3685 - up llog sem before sending rpc (3652)
3686 - reduce ns lock hold times when setting kms (3267)
3687 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3688 - fix NULL deref and obd_dev leak on setup error (3312)
3689 - replace some LBUG about llog ops with error handling (3841)
3690 - don't match INVALID dentries from d_lookup and spin (3784)
3691 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3692 - fix invalid assertion in ptlrpc_set_wait (3880)
3694 - add libwrap support for the TCP acceptor (3996)
3695 - add /proc/sys/portals/routes for non-root route listing (3994)
3696 - allow setting MDS UUID in .xml (2580)
3697 - print the stack of a process that LBUGs (4228)
3699 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3702 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3703 - make sure to unset replay flag from failed open requests (3440)
3704 - if default stripe count is 0, use OST count for inode size (3636)
3705 - update parent mtime/ctime on client for create/unlink (2611)
3706 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3707 - free recovery state on server during a forced cleanup (3571)
3708 - unregister_reply for resent reqs (3063)
3709 - loop back devices mounting and status check on 2.6 (3563)
3710 - fix resource-creation race that can provoke i_size == 0 (3513)
3711 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3712 - more debugging for page-accounting assertion (3746)
3713 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3714 - avoid deadlock after precreation failure (3758)
3715 - fix race and lock order deadlock in orphan handling (3450, 3750)
3716 - add validity checks when grabbing inodes from l_ast_data (3599)
3718 - add /proc/.../recovery_status to obdfilter (3428)
3719 - lightweight CDEBUG infrastructure, debug daemon (3668)
3720 - change default OSC RPC parameters to be better on small clusters
3721 - turn off OST read cache for files smaller than 32MB
3722 - install man pages and include them in rpms (3100)
3723 - add new init script for (un)mounting lustre filesystems (2593)
3724 - run chkconfig in %post for init scripts (3701)
3725 - drop scimac NAL (unmaintained)
3727 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3730 - clean kiobufs before and after use (3485)
3731 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3732 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3733 - change init script to fail more clearly if not run as root (1528)
3734 - allow clients to reconnect during replay (1742)
3735 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3736 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3737 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3738 - fix cond_resched (3554)
3739 - only evict unfinished clients after recovery (3515)
3740 - allow bulk resend, prevent data loss (3570)
3741 - dynamic ptlrpc request buffer allocation (2102)
3742 - don't allow unlinking open directory if it isn't empty (2904)
3743 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3744 - remove extraneous obd dereference causing LASSERT failure (3334)
3745 - don't use get_cycles() when creating temp. files on the mds (3156)
3746 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3747 - handle EEXIST for set-stripe, set proper directory name (3336)
3749 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3750 - fix ksocknal_fmb_callback() error messages (2918)
3752 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3755 - don't copy lvb into (possibly NULL) reply on error (2983)
3756 - don't deref dentry after dput, don't free lvb on error (2922)
3757 - use the kms to determine writeback rpc length (2947)
3758 - increment oti_logcookies when osc is inactive (2948)
3759 - update client's i_blocks count via lvb messages (2543)
3760 - handle intent open/close of special files properly (1557)
3761 - mount MDS with errors=remount-ro, like obdfilter (2009)
3762 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3763 - don't use cancelling-locks' kms values (2947)
3764 - use highest lock extent for kms, not last one (2925)
3765 - don't dereference ERR_PTR() dentry in error handling path (3107)
3766 - fix thread race in portals_debug_dumplog() (3122)
3767 - create lprocfs device entries at setup instead of at attach (1519)
3768 - common AST error handler, don't evict client on completion race (3145)
3769 - zero nameidata in detach_mnt in 2.6 (3118)
3770 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3771 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3772 - initialize RPC timeout timer earlier for 2.6 (3219)
3773 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3774 - print nal/nid for unknown nid (3258)
3775 - additional checks for oscc recovery before doing precreate (3284)
3776 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3777 - don't crash in mdc_close for bad permissions on open (3285)
3778 - zero i_rdev for non-device files (3147)
3779 - clear page->private before handing to FS, better assertion (3119)
3780 - tune the read pipeline (3236)
3781 - fix incorrect decref of invalidated dentry (2350)
3782 - provide read-ahead stats and refine rpc in flight stats (3328)
3783 - don't hold journal transaction open across create RPC (3313)
3784 - update atime on MDS at close time (3265)
3785 - close LDAP connection when recovering to avoid server load (3315)
3786 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3787 - don't leak open file on MDS after open resend (3325)
3788 - serialize filter_precreate and filter_destroy_precreated (3329)
3789 - loop device shouldn't call sync_dev() for nul device (3092)
3790 - clear page cache after eviction (2766)
3791 - resynchronize MDS->OST in background (2824)
3792 - refuse to mount the same filesystem twice on same mountpoint (3394)
3793 - allow llmount to create routes for mounting behind routers (3320)
3794 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3795 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3796 - fix rare problem with rename on htree directories (3417)
3798 - allow default OST striping configuration per directory (1414)
3799 - fix compilation for qswnal for 2.6 kernels (3125)
3800 - increase maximum number of MDS request buffers for large systems
3801 - change liblustreapi to be useful for external progs like lfsck (3098)
3802 - increase local configuration timeout for slow disks (3353)
3803 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3805 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3808 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3809 - don't overwrite extent policy data in reply if lock was blocked (2901)
3810 - drop filter export grants atomically with removal from device (2663)
3811 - del obd_self_export from work_list in class_disconnect_exports (2908)
3812 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3813 - swab reply message in mdc_close, other PPC fixes (2464)
3814 - fix destroying of named logs (2325)
3815 - overwrite old logs when running lconf --write_conf (2264)
3816 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3817 - fix race in target_handle_connect (2898)
3818 - mds_reint_create() should take same inode create lock (2926)
3819 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3820 - don't close files for self_export to avoid uninitialized obd (2936)
3821 - allow MDS with the same name as client node (2939)
3822 - hold dentry reference for closed log files for unlink (2325)
3823 - reserve space for all logs during transactions (2059)
3824 - don't evict page beyond end of stripe extent (2925)
3825 - don't oops on a deleted current working directory (2399)
3826 - handle hard links to targets without a parent properly (2517)
3827 - don't dereference NULL lock when racing during eviction (2867)
3828 - don't grow lock extents when lots of conflicting locks (2919)
3830 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3833 - account for cache space usage on clients to avoid data loss (974)
3834 - lfsck support in lustre kernel code (2349)
3835 - reduce journal credits needed for BRW writes (2370)
3836 - orphan handling to avoid losing space on client/server crashes
3837 - ptlrpcd can be blocked, stopping ALL progress (2477)
3838 - use lock value blocks to assist in proper KMS, faster stat (1021)
3839 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3840 - recovery for initial connections (2355)
3841 - fixes for mds_cleanup_orphans (1934)
3842 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3843 - block all file creations until orphan recovery completes (1901)
3844 - client remove rq_connection from request struct (2423)
3845 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3846 - recovery timer race (2670)
3847 - mdc_close recovey bug (2532)
3848 - ptlrpc cleanup bug (2710)
3849 - mds timeout on local locks (2588)
3850 - namespace lock held during RPCs (2431)
3851 - handle interrupted sync write properly (2503)
3852 - don't try to handle a message that hasn't been replied to (2699)
3853 - client assert failure during cleanup after abort recovery (2701)
3854 - leak mdc device after failed mount (2712)
3855 - ptlrpc_check_set allows timedout requests to complete (2714)
3856 - wait for inflight reqs when ptlrpcd finishes (2710)
3857 - make sure unregistered services are removed from the srv_list
3858 - reset bulk XID's when resending them (caught by 1138 test)
3859 - unregister_bulk after timeout
3860 - fix lconf error (2694)
3861 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3862 - readahead locks pages, leaves pending causing memory pressure (2673)
3863 - increase OST request buffers to 4096 on large machines (2729)
3864 - fix up permission of existing directories in simple_mkdir (2661)
3865 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3866 - don't assign transno to errored transactions (2742)
3867 - don't delete objects on OST if given a bogus objid from MDS (2751)
3868 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3869 - if rq_no_resend, then timeout request after recovery (2432)
3870 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3871 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3872 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3873 - MDS recovery shouldn't delete objects that it has given out (2730)
3874 - if enqueue arrives after completion, don't clobber LVB (2819)
3875 - don't unlock pages twice when trigger_group_io returns error (2814)
3876 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3877 - don't write pages to disk if there was an error (1450)
3878 - don't ping imports that have recovery disabled (2676)
3879 - take buffered bytes into account when balancing socknal conn (2817)
3880 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3881 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3882 - remove little-endian swabbing of llog records (1987)
3883 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3884 - retry reposting request buffers if they fail (1191)
3885 - grow extent at grant time to avoid granting a revoked lock (2809)
3886 - lock revoke doesn't evict page if covered by a second lock (2765)
3887 - disable VM readahead to avoid reading outside lock extents (2805)
3889 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3890 - updated kernel patches for hp-2.4.20 kernel (2681)
3892 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3895 - fix truncated write corruption (2366)
3896 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3898 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3899 - ensure len(uuid) < 37 in lmc (1171)
3900 - fix ia64 OOPS in llog_test (2255)
3901 - zero end of page at obdfilter for partial page writes (2648)
3902 - don't leave stale dentries around after renames (bug 2428)
3903 - fix timeouts when evicting a client with a single lock held (2642)
3904 - set deadline for the initial HELLO message to drain (2634)
3905 - print out dotted-quad IP addresses in the socknal (2302)
3907 - additional debugging for MDS client eviction problem (2443)
3908 - fix mkfsoptions support for osts (2603, 2604)
3910 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3913 - add series for the vanilla 2.6.0 kernel
3914 - add series for the vanilla 2.4.24 kernel
3915 - add series for a cray x86/64 UL kernel drop
3916 - fix xattr patches for the vanilla 2.4.19 series
3918 - generate true UUIDs in lmc (1171)
3919 - have portals stack dumping break in UML (2466)
3920 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3921 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3922 - wake callback waiting threads on client eviction (2460)
3923 - Add --ptldebug and --subsystem to lmc (1719)
3924 - update assertion to allow safe interrupt allocation
3925 - set rq_no_resend for cancel requests (2432)
3926 - recalculate ptlrpcd timeout after resend (2494)
3927 - call vfs_rmdir when removing pending directories (2368)
3928 - fix renaming a file to itself (2429)
3929 - lmc creates a default one-stripe lov (2454)
3930 - expand procfs space to handle large clusters (2326)
3931 - increase UML stack to avoid overflow
3932 - update lconf's list of debug and subsystem masks
3933 - fix lfs find --obd (2510)
3934 - /proc tunable for disabling filter read caching (2591)
3935 - stop rpm packages from altering slapd.conf (2301)
3936 - disable nagle in the socknal under 0conf (2578)
3937 - choose mds inode size based on stripe count (2572)
3938 - fix kernel-source rpm problems (2516)
3940 - add --disable-doc to avoid pdf generation (2421)
3941 - update documentation, tests, type-os, comments
3942 - avoid format warnings on ia64
3943 - remove the TOE NAL
3944 - tiny code cleanups by removing unused fields
3946 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3949 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3950 - remove the most blatant lies from BUILDING file (2371)
3951 - change default debug level to reasonable production setting
3952 - reduce client side cache size to reduce cache flush time
3953 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3954 - make TCP zerocopy and pinger support enabled by default (2476)
3955 - sync writes completed after process exits caused crashes (2319)
3956 - maintain correct mount count on the MDS (2356)
3957 - backout 1557, because 2316 wasn't really fixed
3958 - better file I/O statistics gathering in /proc
3959 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3960 - another kernel patch to fix zero-copy TCP function export
3961 - don't take duplicate lock when processing re-sent getattr (2420)
3962 - lctl uses obd_self_export instead of creating new conn (2353)
3963 - MDS/OST recovery case which requires object creation asserted (2425)
3964 - move lfs from /usr/sbin to /usr/bin in packages
3965 - fix race between mds_client_add and mds_client_free (2417)
3966 - use kmalloc instead of slabs in portals (2430)
3967 - don't create duplicate records when a failover MDS is present (2442)
3968 - remove unnecessary mount age check (2332)
3969 - don't remove directory inodes from locks prematurely (2451)
3970 - don't break if MDS service name is the same as hostname (2103)
3971 - fix races in client write RPC generation when cache full (2482)
3973 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3976 - remove now-unused request->rq_obd (278)
3977 - if an allocation fails, print out how much memory we've used (1933)
3978 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3979 - add the "configurable stack size" patch to most series files (1256)
3980 - ability to write large log records, for 100+ OST configs (2306)
3981 - fix NULL deref when filter_prep fails (2314)
3982 - fix operator precedence error in filter_sync
3983 - dynamic allocation of socknal TX descriptors (2315)
3984 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3985 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3986 - ensure that CWARN messages in Portals always get to the syslog
3987 - __init/__exit are not for prototype decls (ldlm_init/exit)
3988 - x86-64 compile warning fixes
3989 - fix gateway LMC keyword conflict (2318)
3990 - fix MDS lock inversions in getattr/reint paths (1844)
3991 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3992 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3993 - DLM race condition prevented some lock evictions (2328)
3994 - ENOMEM detection and retry on socknal sends (2230)
3995 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3996 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3997 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3998 - better error messages when a client is rejected during recovery (1505)
3999 - avoid cancelling locks which were never granted, after failure (2330)
4000 - fix i_sem/journal inversion in mds_client_add (2333)
4001 - fix truncate/getattr lock cycle deadlock (2334)
4002 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4003 - fix two rare exit paths which could leak an l_lock() ref (2321)
4004 - fencepost error in MDS/OST orphan recovery (2226)
4005 - make log record alignment 8 bytes (1988)
4006 - lstripe now fails when requested offset > ost_count (2237)
4007 - ensure that all kernel series have a complete list.h (1607)
4008 - fix crashes in special-file operations (2316)
4009 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4011 - allow configurable automake binary, for testing new versions
4012 - small update to the lfs documentation
4014 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4016 * fix negative export reference count in fsfilt_sync (2312)
4018 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4019 * release candidate 0.9.1
4021 - orphans are moved into the PENDING directory for possible recovery
4022 - replayed opens now open by fid for orphan/rename safety (1042)
4023 - last close of an orphan inode generates a transno (683)
4024 - chdir() and mount() now pin the directory entry (1020)
4025 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4026 - discard very old requests without processing them (1502)
4027 - remove some common, well-understood CERRORs (1505)
4028 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4029 - clear "grant" flags in OST replies until OST grant code lands (1644)
4030 - fix read performance by not clobbering i_blksize on client (1598)
4031 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4032 - make LRU size a /proc tunable, clears locks when reduced (707)
4033 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4034 - server locks take references on exports now (1558)
4035 - build fixes for 2.4.20-rh trees (1663)
4036 - return an error from lov_create if all OSCs are inactive (1751)
4037 - fix import levels when a reconnect happens without a timeout (1597)
4038 - exit early from mds_open if we get a lookup error (1749)
4039 - partial page read at EOF wouldn't wait for disk before sending (1642)
4040 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4041 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4042 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4043 - write/truncate lock inversion (1639)
4044 - Don't auto-load obdclass, portals modules during cleanup (1495)
4045 - fix timestamps from jumping to "now" (1763)
4046 - extra journal assertions (1648)
4047 - add an extra multiunlink test (1771)
4048 - fix read_record/write_record API (1776)
4049 - fix leak of offset_extent, possible incorrect i_size later (1772)
4050 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4051 - Debugging for the kqswnal_get_idle_tx problems (1820)
4052 - Allow recovery to be attempted multiple times (1536)
4053 - Write out MDS last_rcvd file after it is first created (1600)
4054 - Fix tx_descriptor leak in failed transmit situations (1827)
4055 - ext3 journaling fixes for assertion failure after IO error (1871)
4056 - class_export_put() on freed export after completion AST error (1896)
4057 - Fix revalidate looping in VFS (1322)
4058 - Don't access a freed export during MDS_REINT timeout (1521)
4059 - Add open-unlink recovery support on the MDS (1673,1764)
4060 - Return an error if no MDS data was read from last_rcvd (1946)
4061 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4062 - Don't setup a disk device that doesn't match exported UUID (317)
4063 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4064 - avoid committing NULL handle in force close
4065 - local.sh is now a one-stripe LOV configuration
4066 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4067 - fix ext3 htree duplicate directory entry corruption (1516)
4068 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4069 - update to new LOV EA format (2097)
4070 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4072 2003-06-15 Phil Schwan <phil@clusterfs.com>
4075 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4076 - per-import/export recovery handling (958, 931, 959)
4077 - multiple last-rcvd slots, for serving multiple FSes (949)
4078 - connections are again shared between multiple imp/exports (963, 964)
4079 - "umount -f" would hang if any requests needed to be sent (393, 978)
4080 - avoid pinning large req buffer by copying for queued messages (989)
4081 - add "uuid" to "lctl device" command to help upcalls (991)
4082 - "open" RPCs with transnos would confuse recovery counters (1037)
4083 - do proper endian conversion of all wire messages (288, 340, 891)
4084 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4085 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4086 - fix ptlrpc_connection leak in target_handle_connect (1174)
4087 - fix import refcounting bug in OST and MDS cleanup (1134)
4088 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4089 - fix very unlikely obd_types race condition (501)
4090 - remove osc_open hack for echo_client (1187)
4091 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4092 - a failure in read_inode2 leads to deadlock (1139)
4093 - cancel ack-locks as soon as transaction is committed (1072)
4094 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4095 - make sure to commitrw after any preprw to avoid deadlock (1162)
4096 - failing to execute a file in a lustre FS would lock inode (1203)
4097 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4098 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4099 - obd_uuid2tgt would walk off the end of the list (1255)
4100 - on IA64 the osc would give portals incorrect bulk size (1258)
4101 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4102 - fix lock inversion caused by new llite matching code (1282)
4103 - limit the number of dirty pages on a client to 10MB (1286)
4104 - timed out locks were not being corrected cancelled (1289)
4105 - fix O_DIRECT above 4GB on IA-32 (1292)
4106 * major user-visible changes
4107 - fail out/fail over policy now controlled by the upcall (993)
4109 - add OBD_PING to check server availability and failure (954)
4110 - lustre messages are now sent in sending host order (288, 340, 891)
4111 - add eadatalen to MDS getattr reply (340)
4112 - OST read replies may contain second buffer, with per-page status (593)
4114 2003-03-11 Phil Schwan <phil@clusterfs.com>
4117 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4118 - failed open()s could cause deadlock; fixed (867, 869)
4119 - stop cancelling OST locks when files are closed (481)
4120 - overlapping XID spaces caused network corruption (851, 853)
4121 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4122 - setattr_raw added, to do single-RPC, server-side setattrs
4123 - lmc/lconf syntax change for OST UUIDs
4124 - fix crashy race condition between ptlrpc_free_req and osc_close
4125 - don't use request in mdc_enqueue if we hit a timeout (889)
4126 - don't set the inode i_size for regular files from the MDS (896)
4127 - handle out of order completion AST (842)
4128 - don't LBUG if a lock request times out after receiving AST (913)
4129 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4130 - if a bad lock AST arrives, send an error instead of dropping entirely
4131 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4132 - fix leak in bulk IO when only partially completed (899, 900, 926)
4133 - fix O_DIRECT for ia64 (55)
4134 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4135 - C-z after timeout could hang a process forever; fixed (977)
4137 - client-side I/O cache (678, 924, 929, 941, 970)
4139 - READPAGE and SETATTRs which don't take server-side locks get
4142 2003-02-11 Phil Schwan <phil@clusterfs.com>
4145 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4146 - Fix fsfilt_extN_readpage() to read a full page of directory
4147 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4148 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4149 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4150 - Most intent operations were removed, in favour of a new RPC mode
4151 that does a single RPC to the server and bypasses most of the VFS
4152 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4153 - Aggressively cancel local locks on DLM servers
4154 - mds_reint_unlink sends EA to the client if it's the last nlink.
4155 client uses that EA to unlink OST objects.
4156 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4157 - recursive symlinks were fixed (439)
4158 - fixed NULL deref in DEBUG_REQ
4159 - filter_update_lastobjid no longer calls sync, which annoyed extN
4160 - fixed multi-client small-writes to a single file problem (445)
4161 - fixed mtime updates during file writes (607)
4162 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4163 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4164 - fixed Linux OST size reporting problem (444, 656)
4165 - OST now updates object mtime with writes or setattr (607, 619)
4166 - client verifies file size before zeroing page past EOF (445)
4167 - OST now writes last allocated objid to disk with allocation (108)
4168 - LOV on echo now works (409)
4170 - mds_reint_unlink sends a new buffer, with the EA included. this
4171 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4172 set if a regular file was being unlinked, and it was the last link
4173 - use PtlGet from the target for bulk writes (315)
4174 - OST now updates object mtime with writes or setattr (607, 619)
4175 - LDLM now has a grant-time callback to revalidate locked items, if
4177 - Many MDS operations were reorganized to combat race conditions
4179 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4180 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4182 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4185 - Fully reactivate OST imports after reconnection (512, others)
4186 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4187 - More graceful error handling for truncating on dead OST (515)
4188 - Don't error out unless we're actually accessing dead stripes (474)
4189 - Fix garbage sizes when stripes are missing (410)
4190 - LRU counters were broken, causing constant lock purge (433, 432)
4191 - garbage on read from stripes with failed OSTs (441)
4192 - mark OSCs as active before reconnecting during recovery (438)
4193 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4194 - lfind did not preserve OST order in output (443)
4195 - symlinks cause hung clients, incorrect data (439)
4196 - stop dereferencing request after dropping refcount (457)
4197 - don't LASSERT(spin_is_locked) on non-SMP (455)
4198 - fixes for many rename() bugs
4199 - fstat didn't correctly synchronize attributes (399)
4200 - server must handle lock cancellation during blocking AST prep (487)
4201 - bulk descriptors were free()d too soon (511)
4202 - fix paths in lconf, which would load incorrect modules (451, 507)
4203 - fix confusing lconf 'host not found' error message (386)
4204 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4205 - fix race condition in mdc_blocking_ast() for inode access (526)
4206 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4207 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4208 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4209 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4210 - fix lstripe in several ways (kernel side) (527)
4211 - fix request leak in ldlm_cli_enqueue (262)
4212 - incorrect OSC was marked inactive after OST failure
4213 - call mds_fs_cleanup before unmounting filesystem (524)
4214 - fix races between taking ns_lock and ldlm_lock_change_resource
4215 - fix races updating LOV export open file list
4216 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4217 - fix recovery NULL deref in ldlm_cli_cancel_unused
4218 - fix some DLM races by using new hash table for lock handles (419)
4219 - permit the client to specify desired inodes, at replay
4220 - duplicate requests when we queue them for replay reintegration
4221 - fix last_rcvd offset calculation
4222 - sync after each recovered transaction, so we always make progress
4223 - never, not always, ERESTART requests without transnos
4224 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4226 - skip replay if the MDS says that the client is already connected
4227 - don't check for a recovery-enabled export to match lctl's UUID
4228 - don't INC_USE_COUNT for phantom exports
4229 - don't crash when cleaning up phantom exports (567)
4230 - don't double-finish or set replay data for errored mdc_open requests
4231 - abort requests when they time out, so we don't get old replies
4232 - send/receive replies for AST messages again
4233 - if the client says that it doesn't have the lock, cancel it on the
4235 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4236 mark it as destroyed, it will all work out when decref is called
4237 - fix module use counts (22, 581)
4239 - ASTs now expect a reply (server cancels lock on error reply)
4241 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4244 - fix many simultaneous client startup (392)
4245 - fix dentry->d_it clobbering
4246 - credentials weren't being shipped for readdir/getattr operations
4247 - remove invalid assertions triggered during some concurrent MD
4249 - proper Lustre versions added (336, 389)
4250 - fix memory leak for create error case (398)
4251 - fix LOV locking bug that would get cli/srv out of sync
4252 - fix echo client over LOV (409)
4253 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4254 - fix double-O_EXCL intent crash (424)
4255 - avoid sending multiple lock CANCELs (352)
4257 - MDS can do multi-client recovery (modulo bugs in new code)
4259 - many updates, edits, cleanups
4261 2002-11-18 Phil Schwan <phil@clusterfs.com>
4264 - fix null d_it dereference (346)
4265 - fix full OST/dbench hang (333)
4266 - fix permission problem with file removal (286)
4267 - fix removal of OSCs from LOV when they fail
4268 - fix NULL deref during bulk timeout (214)
4269 - fix problems related to multiple filesystems on one MDS (241)
4270 - fixed serious subtle metadata locking bugs
4271 - free locks on clients when inodes are removed due to memory
4273 - fix inode pointer in lock data (285)
4274 - partial support for multiple MDS on a single host (241)
4275 - data locks weren't cancelled at clear_inode time (290, 311)
4276 - intent locks could lead to unbounded lock growth (205)
4277 - added a maximum lock count, an LRU list, and a flusher
4278 - fix multiple rename (365)
4279 - properly abstracted the echo client
4280 - OSC locked 1 byte too many; fixed
4281 - rewrote brw callback code:
4282 - fixed recovery bugs related to LOVs (306)
4283 - fixed too-many-pages-in-one-write crash (191)
4284 - fixed (again) crash in sync_io_timeout (214)
4285 - probably fixed callback-related race (385)
4287 - Add capability to MDS protocol
4288 - LDLM cancellations and callbacks on different portals
4290 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4293 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4294 - defer open object destruction to close time (601981, 138)
4295 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4296 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4297 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4298 - don't call obd_setattr in ll_file_release for destroyed objects
4299 * protocol change to lustre_msg: move |version| and add |flags|
4300 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4301 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4302 * added replay of create, unlink, link and rename operations during
4303 MDS failover; recovery should be much more robust now
4304 * remove failed OSCs from LOVs (only lov_create uses this so far)
4305 * the lustre-HOWTO was brought (more) up to date (582544)
4307 2002-10-23 Phil Schwan <phil@clusterfs.com>
4310 - in-use dentries weren't being reused properly (617851)
4311 - prevent multiple LDLM setup (599178)
4312 - fix LOV size calculations for truncate (617853)
4313 - fix client handling of MDS intent errors (POSIX)
4314 - fix permission bug in lovstripe.c test (624321)
4315 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4316 - truncate past end of file could corrupt data
4317 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4318 - a race in recovery could return ETIMEDOUT to apps (623947)
4319 - building outside the source directory was fixed
4320 * the lustre-HOWTO was brought (more) up to date (582544)
4321 * major progress was made on recovery functionality
4323 2002-10-10 Phil Schwan <phil@clusterfs.com>
4326 - recovery deadlock fix
4327 - rm -rf causes LBUG fix (617817)
4328 - file open by multiple tasks fix (618962)
4329 - directory permissions bugs (602707 and 620007)
4330 - journal_stop fixed with locking (611313)
4331 - O_APPEND failures resolved (618273, perhaps 614459)
4332 - lconf PATH fix (619770)
4333 - IA64 build fix (621450)
4334 - RPC buffer sizes scale with amount of memory
4336 2002-10-01 Phil Schwan <phil@clusterfs.com>
4339 - locks would be cancelled without throwing away data pages,
4340 resulting in inconsistent data (605627)
4341 - inode attributes were not always being refreshed (605627, 612449)
4342 - lconf now continues to cleanup after lctl reports an error
4343 - MDS now enforces user permissions (602707)
4344 - lprocfs cleanup fixed, but not yet enabled (614157)
4345 - fixed infinite server hang, should a client not respond to an AST
4346 - avoid going into recovery if user calls readlink() with a buffer
4347 that's too small (613941)
4348 - AST RPCs no longer require replies (614867) -- this may be changed
4349 - don't crash server if client sends an IOV that's too big (611336)
4350 - fixed lock conversion deadlock (611892)
4351 - fixed the following of symlinks (614622)
4352 * recovery: the server can remove locks from a client that dies, other
4353 clients can make progress
4354 * more extN patch fixes
4355 * compile-time configurable ptlrpc buffer allocations
4357 - collaborative read cache document
4358 - Lustre Lite Performance CDR document-in-progress
4360 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4363 - fix typo in patch-2.4.18
4365 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4368 - clear ptlrpc request each time in handle_incoming_request()
4369 - unlink of files now destroys the object on the OST
4371 2002-09-19 Peter Braam <braam@clusterfs.com>
4373 * add hard link support
4374 * change obdfile creation method
4375 * kernel patch changed
4377 2002-09-19 Peter Braam <braam@clusterfs.com>
4380 - stack overflow bug in extN fixed
4382 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4384 * documentation updates
4385 - add man pages for config tools
4386 - update tests/README to describe testing with new config tools
4387 - finish metadata API descriptions
4388 * bug fixes and cleanups
4389 - statfs workaround for 16TB limit
4390 - LOV stripe allocation improved, can stripe on subset of OSTs
4391 - LOV file size/IO offset was wrong for files > 4GB in size
4392 - object EA data was being dropped, caused files to be unreadable
4393 - memory overflow with non-LOV OST caused memory corruption
4394 - fixed regression tests to work with new config tools, obdfilter
4395 - fixed bug when directory size became larger than 1 block
4396 - fixed bug (for single client case) when PWD was deleted
4397 - invalidate local directory pages when doing intent-based ops
4398 - avoid LDLM oops when lock callback contained bad data
4400 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4402 * documentation updates
4403 * bug fixes and cleanups
4404 - configuration tools
4407 - 64-bit compile warnings
4408 - 64-bit internal statfs data
4410 * test_brw on persistent OST devices
4412 * lprocfs (disabled)
4414 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4416 * documentation updates
4417 * bug fixes and cleanups
4418 * configuration tools
4420 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4423 * many small fixes to 0_5_4
4424 * io/network handling
4425 * thinkos in MDS operations
4427 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4430 * crucial basic fixes to 0.5.3
4431 * IOR, Iozone work over Elan
4434 2002-08-07 Phil Schwan <phil@clusterfs.com>
4435 * version 0_5_3, our first alpha
4436 * we use the new Portals iovs
4437 * documentation updates
4438 * bug fixes and cleanups
4439 * small changes in the DLM wire protocol
4441 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4442 * version 0_5_1 with some initial stability,
4443 * locking on MD and file I/O.
4444 * documentation updates
4445 * several bug fixes since 0.5.0
4446 * small changes in wire protocol
4448 2002-07-18 Phil Schwan <phil@clusterfs.com>
4450 * delivered as Lustre Light Alpha
4451 * fixed a crash after handling invalid MDS requests
4452 * fixed directory pages for architectures with non-4k pages sizes
4454 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4455 * release version v0_4_4
4456 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4458 * Fixes a number of interruption problems with OST operations.
4459 * Update documentation for portals header changes
4460 * Move all wire protocol structs/defines to lustre_idl.h
4461 * Fixes symlink length bug.
4462 * Add tcpdump to repository.
4464 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4465 * release version v0_4_3
4466 * Fixes statfs for inodes on extN.
4467 * Fixes bug in runtests which would delete /etc/hosts.
4468 * Use 64-bit object IDs wherever possible (not into VFS though)
4469 Remove ost_get_info, which is unused by lustre, and out of date.
4471 2002-07-03 Peter Braam <braam@clusterfs.com>
4472 * release version v0_4_2 Fixes a lookup error (type not passed)
4473 * move forward to head of Portals
4474 * move forward to latest Lustre kernel
4476 2002-06-25 Peter Braam <braam@clusterfs.com>
4477 * release version v0_4_1. Hopefully stable on single node use.