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 : on start mds
18 Description: Implement get_info(last_id) in obdfilter.
21 Frequency : occasional
23 Description: Correctly check stale fid, not start epoch if ost not support SOM
24 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
25 overwrited with new generated fid, but mds can anwer with one of these
26 two fids and both is not stale. setattr incorectly start epoch and
27 assume will be called done_writeting, but without SOM done_writing
31 Frequency : rare, depends on device drivers and load
33 Description: MDS or OSS nodes crash due to stack overflow
34 Details : Code changes in 1.8.0 increased the stack usage of some functions.
35 In some cases, in conjunction with device drivers that use a lot
36 of stack the MDS (or possibly OSS) service threads could overflow
37 the stack. One change which was identified to consume additional
38 stack has been reworked to avoid the extra stack usage.
41 Frequency : occasional
43 Description: Do not fail import if osc_interpret_create gets -EAGAIN
44 Details : If osc_interpret_create got -EAGAIN it immediately exits and
45 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
46 oscc_has_objects and see OSC has no objests and call
47 oscc_internal_create to resend create request.
49 Severity : enhancement
51 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
53 Severity : enhancement
55 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
60 Description: Fix replay if there is an un-replied request and open
61 Details : In some cases, older replay request will revert the
62 mcd->mcd_last_xid on MDS which is used to record the client's
65 Severity : enhancement
67 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
69 Severity : enhancement
71 Description: Add rhel5 support to HEAD.
73 Serverity : enhancement
75 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
80 Description: Don't allow skipping OSTs if index has been specified.
81 Details : Don't allow skipping OSTs if index has been specified, make locking
82 in internal create lots better.
86 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
87 Details : access to bitfield in structure is always rounded to long
88 and this produce problem with not atomic change any bit.
92 Description: Lustre make rpms failed.
93 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
94 builds Lustre rpms from tarball.
97 Frequency : rare on shutdown ost
99 Description: If llog cancel was not send before clean_exports phase, this can
100 produce deadlock in llog code.
101 Details : If llog thread has last reference to obd and call class_import_put
102 this produce deadlock because llog_cleanup_commit_master wait when
103 last llog_commit_thread exited, but this never success because was
104 called from llog_commit_thread.
108 Description: allow userland application know is lost one of stripes.
109 Details : fill lvb_blocks with error code on ost and return it to
110 application if error flag found.
114 Description: NULL lov_tgts causing MDS oops
115 Details : more safe checks for NULL lov_tgts for avoid oops.
117 Severity : enhancement
119 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
123 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
124 Details : If an OST is down the MDS will hang indefinitely in
125 obd_statfs_rqset() waiting for the statfs data. While for
126 MDS QOS usage of statfs, it should not stuck in waiting.
128 Severity : enhancement
130 Description: remote_acl support
131 Details : Support ACL-based permission check for remote user.
132 Support setfacl/getfacl for remote user with the utils
133 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
134 the system "{s,g}etfacl" utils.
136 Severity : enhancement
138 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
140 Severity : enhancement
142 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
146 Description: make mgs_setparam() handle fsname containing dash
147 Details : fsname containing a dash does not work with lctl conf_param
149 Severity : enhancement
151 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
153 Severity : enhancement
155 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
159 Frequency : rare, at shutdown
160 Description: access already free / zero obd_namespace.
161 Details : if client_disconnect_export was called without force flag set,
162 and exist connect request in flight, this can produce access to
163 NULL pointer (or already free pointer) when connect_interpret
164 store ocd flags in obd_namespace.
168 Frequency : only at startup
169 Description: not alloc memory with spinlock held.
170 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
171 if any spinlock held.
173 Severity : enhancement
175 Description: make lustre randomly fail allocating memory
176 Details : Make lustre randomly failed allocating memory for testing purpose.
178 Severity : enhancement
180 Description: lost problems with lov objid file
181 Details : Fixes some scability and access to not inited memory problems
182 in work with lov objdid file.
187 Description: lfs find does not continue on file error
188 Details : Continue other files processing when a file/dir is absent.
192 Description: Inconsistent usage of lustre_pack_reply()
193 Details : Standardize the usage of lustre_pack_reply() such that it
194 always generate a CERROR on failure.
197 Frequency : occasional
199 Description: MDS hang or stay in waiting lock
200 Details : If client receive lock with CBPENDING flag ldlm need send lock
201 cancel as separate rpc, to avoid situation when cancel request
202 can't processed due all i/o threads stay in wait lock.
206 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
210 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
214 Description: Update to SLES9 kernel-2.6.5-7.287.3.
218 Description: Update to RHEL4 latest kernel.
220 Severity : enhancement
222 Description: Build SLES10 patchless client fails
223 Details : The configure was broken by run ./configure with
224 --with-linux-obj=.... argument for patchless client. When the
225 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
226 can't search header adequately. Use absolute path such as
227 -I($LINUX)/include instead.
231 Description: interrupt oig_wait produce painc on resend.
232 Details : brw_redo_request can be used for resend requests from ptlrpcd and
233 private set, and this produce situation when rq_ptlrpcd_data not
234 copyed to new allocated request and triggered LBUG on assert
235 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
236 ptlrpcd set if request is changed and can be safety changed to use
241 Description: LASSERT_{REQ,REP}SWAB macros are buggy
242 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
243 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
244 these macros and replace them with their difinitions instead.
248 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
249 Details : Update which_patch & target file for SLES10 latest kernel.
252 Description: Remove client patches from SLES 10 kernel.
253 Details : This causes SLES 10 clients to behave as patchless clients
254 even on a Lustre-patched (server) kernel.
256 Severity : enhancement
258 Description: self-adjustable client's lru lists
259 Details : use adaptive algorithm for managing client cached locks lru
260 lists according to current server load, other client's work
261 pattern, memory activities, etc. Both, server and client
262 side namespaces provide number of proc tunables for controlling
265 Severity : enhancement
267 Description: light-weight GSS support
268 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
269 and reduce performance overhead.
271 Severity : enhancement
273 Description: Linux keyring support
274 Details : Support using service of Linux keyring for Lustre GSS internal
275 context refresh/cache mechanism.
279 Description: Fix errors in lfs documentation
280 Details : Fixes man pages
284 Description: don't use GFP_* in generic Lustre code.
285 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
289 Description: obdclass is limited by single OBD_ALLOC(idarray)
290 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
294 Description: warnings with build patchless client with vanila 2.6.19 and up
295 Details : change old ctl_table style and replace ctl_table/ctl_table_header
296 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
300 Frequency : Only for SLES
301 Description: sanity_quota fail test_1
302 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
303 file, which makes TSTID[2] inunique.
307 Frequency : Only for SLES9
308 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
309 Details : Update target/ChangeLog/which_patch .
313 Frequency : Only for test_10 in sanity-quota.sh
314 Description: a bug in quota test code
315 Details : checking a test flag in a wrong place
319 Frequency : Only for test_13 in sanity-quota.sh
320 Description: a bug in quota test script
321 Details : shouldn't check overfull allocated quota
325 Description: jbd statistics
326 Details : Port older jbd statistics patch for sles10
328 Severity : enhancement
330 Description: Add printing inode info into message about error in writepage.
335 Description: reply_lock_interpret crash due to race with it and lock cancel.
336 Details : Do not replay locks that are being cancelled. Do not reference
337 locks by their address during replay, just by their handle.
341 Frequency : When flocks are used.
342 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
343 Details : Flock locks might destroy just granted lock if it could be merged
344 with another existing flock, this is done in completion handler,
345 so teach ldlm_cli_enquque_fini that this is a valid case for
351 Description: Oops in read and write path when failing to allocate lock.
352 Details : Check if lock allocation failed and return error back.
356 Description: lstripe command fails for valid OST index
357 Details : The stripe offset is compared to lov->desc.ld_tgt_count
358 instead of lov->desc.ld_active_tgt_count.
362 Description: sanity.sh failed test 103
363 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
368 Description: df doesn't work properly if diskfs blocksize != 4K
369 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
373 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
374 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
378 Description: obd_health_check_timeout too short
379 Details : set obd_health_check_timeout as 1.5x of obd_timeout
383 Description: llapi_file_create() does not allow some changes
384 Details : add llapi_file_open() that allows specifying the mode and
385 open flags, and also returns an open file handle.
389 Description: merge and cleanup kernel patches.
390 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
394 Description: Add journal checksum support.(Kernel part)
395 Details : The journal checksum feature adds two new flags i.e
396 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
397 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
398 indicates that the commit block contains the checksum for
399 the blocks described by the descriptor blocks. Now commit
400 record can be sent to disk without waiting for descriptor
401 blocks to be written to disk. This behavior is controlled
402 using JBD2_FEATURE_ASYNC_COMMIT flag.
406 Description: OSS needs mutliple precreate threads
407 Details : Add ability to start more than one create thread per OSS.
411 Description: Sanity.sh test_65k failed: setstripe should have succeeded
412 Details : execute lfs setstripe on client
416 Description: mds_obd_create error creating tmp object
417 Details : When the user sets quota on root, llog will be affected and can't
418 create files and write files.
421 Frequency : Always on ia64 patchless client, and possibly others.
423 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
424 Details : This allows the patchless client to be loaded on architectures
429 Description: RedHat Update kernel for RHEL5
430 Details : Add the kernel config file for RHEL5.
434 Description: RedHat Update kernel for RHEL5
435 Details : Modify the kernel config file more closer RHEL5.
439 Description: Build failure against Centos5 (RHEL5)
440 Details : Define PAGE_SIZE when it isn't present.
444 Description: test_7 on sanity-quota.sh can't be used on separate machines
445 Details : add facet to handle it
449 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
450 Details : with NFS, the anon dentry's parent was set to itself in
451 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
452 corresponding dentry other than use rec->ur_name.
454 Severity : enhancement
456 Description: lfs setstripe enhancement
457 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
461 Description: enable data checksumming by default
462 Details : enable checksum by default, allow --disable-checksum
463 configure option and "-o nochecksum" mount option. Checksums
464 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
468 Description: lustre support for RHEL5
469 Details : Add support for RHEL5.
473 Description: Client eviction due to failover config
474 Details : after a connection loss, the lustre client should attempt to
475 reconnect to the last active server first before trying the
476 other potential connections.
480 Description: when mds and osts use different quota unit(32bit and 64bit),
481 quota will be released repeatly.
482 Details : void sending multiple quota reqs to mds, which will keep the
483 status between the reqs.
487 Description: osts not allocated evenly to files
488 Details : change the condition to increase offset_idx
491 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
493 Description: Data corruption for OSTs that are formatted larger than 2TB
495 Details : When generating the bio request for lustre file writes the
496 sector number would overflow a temporary variable before being
497 used for the IO. The data reads correctly from Lustre (which
498 will overflow in a similar manner) but other file data or
499 filesystem metadata may be corrupted in some cases.
503 Description: Tune the kernel for good SCSI performance.
504 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
505 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
510 Description: rewrite ext2-derived code in obdclass/uuid.c
511 Details : rewrite inherited code (uuid parsing code from ext2 utils)
512 from scratch preserving functionality.
516 Description: "lfs find -obd UUID" prints directories
517 Details : "lfs find -obd UUID" will return all directory names instead
518 of just file names. It is incorrect because the directories
519 do not reside on the OSTs.
523 Description: use i_size_read and i_size_write in 2.6 port
524 Details : replace inode->i_size access with i_size_read/write()
527 Frequency : when removing large files
529 Description: scheduling issue during removal of large Lustre files
530 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
531 It causes scheduling issues when removing large files (17TB in the
535 Frequency : only with liblustre clients on XT3
537 Description: evictions taking too long
538 Details : allow llrd to evict clients directly on OSTs
541 Frequency : only on ppc
543 Description: /proc/fs/lustre/devices broken on ppc
544 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
545 We should make sure the type of 'index' is loff_t and then cast
546 explicitly as needed below. Do not assign an explicitly cast
550 Frequency : only for rhel5
552 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
553 Details : Modify the target file & which_kernel.
555 Severity : enhancement
557 Description: omit set fsid for export NFS
558 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
559 as FS_REQUIRES_DEV for avoid problems with generate fsid.
563 Frequency : Always, for kernels after 2.6.16
564 Description: Fix warning idr_remove called for id=.. which is not allocated.
565 Details : Last kernels save old s_dev before kill super and not allow
566 to restore from callback - restore it before call kill_anon_super.
570 Description: buffer overruns could theoretically occur
571 Details : llapi_semantic_traverse() modifies the "path" argument by
572 appending values to the end of the origin string, and a
573 overrun may occur. Adding buffer overrun check in liblustreapi.
577 Description: Fix error on 'ls .' at the top of the Lustre mount.
578 Details : Don't revalidate dentry if it is a root dentry.
582 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
583 Details : Modify vm-tunables-rhel4.patch.
587 Description: Kernel config for 2.6.18-vanilla.
588 Details : Modify targets/2.6-vanilla.target.in.
589 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
590 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
591 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
592 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
596 Description: improve handling recoverable errors
597 Details : if request processig with error which can be recoverable on server
598 request should be resend, otherwise page released from cache and
603 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
604 Details : Modify target file & which_patch.
605 A flaw was found in the IA32 system call emulation provided
606 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
607 value could be stored in the %RAX register, which could trigger an
608 out-of-bounds system call table access. An untrusted local user
609 could exploit this flaw to run code in the kernel
610 (ie a root privilege escalation). (CVE-2007-4573).
614 Description: O_DIRECT bypasses client statistics.
615 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
616 read_ahead_stats were updated. copied the stats section from
617 osc_send_oap_rpc() into async_internal().
621 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
625 Description: change order of libsysio includes
626 Details : '#include sysio.h' should always come before '#include xtio.h'
630 Description: handle "serious error: objid * already exists" more gracefully
631 Details : If LAST_ID value on disk is smaller than the objects existing in
632 the O/0/d* directories, it indicates disk corruption and causes an
633 LBUG(). If the object is 0-length, then we should use the existing
634 object. This will help to avoid a full fsck in most cases.
636 Severity : enhancement
638 Description: adapt the lustre_config script to support the upgrade case
639 Details : Add "-u" option for lustre_config script to support upgrading 1.4
640 server targets to 1.6 in parallel.
644 Description: To avoid grant space > avaible space when the disk is almost
645 full. Without this patch you might see the error "grant XXXX >
646 available" or some LBUG about grant, when the disk is almost
648 Details : In filter_check_grant, for non_grant cache write, we should
649 check the left space by if (*left > ungranted + bytes), instead
650 of (*left > ungranted), because only we are sure the left space
651 is enough for another "bytes", then the ungrant space should be
652 increase. In client, we should update cl_avail_grant only there
653 is OBD_MD_FLGRANT in the reply.
657 Description: Update RHEL 4 kernel to fix local root privilege escalation.
658 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
659 described in CVE-2007-4573. This problem could allow untrusted
660 local users to gain root access.
663 Frequency : when using O_DIRECT and quotas
665 Description: Incorrect file ownership on O_DIRECT output files
666 Details : block usage reported by 'lfs quota' does not take into account
667 files that have been written with O_DIRECT.
672 Description: touch file failed when fs is not full
673 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
674 otherwise we can get ENOSP while fs is not full.
678 Description: parallel lock callbacks
679 Details : Instead of sending blocking and completion callbacks as separated
680 requests, adding them to a set and sending in parallel.
683 Frequency : only for Cray XT3
684 Bugzilla : 12829/13455
685 Description: Changing primary group doesn't change the group lustre assigns to
687 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
688 overridden with the primary group provided by the group upcall,
689 whereas the supplied fsgid can be trusted if it is in the list of
690 supplementary groups returned by the group upcall.
692 Severity : enhancement
694 Description: Allow masking D_WARNING, D_ERROR messages from console
695 Details : Console messages can now be disabled via lnet.printk.
699 Description: User code with malformed file open parameter crashes client node
700 Details : Before packing join_file req, all the related reference should be
701 checked carefully in case some malformed flags cause fake
702 join_file req on client.
706 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
707 Details : ldlm_completion_ast() assumes that a lock is granted when the req
708 mode is equal to the granted mode. However, it should also check
709 that LDLM_FL_CP_REQD is not set.
713 Description: Heavy nfs access might result in deadlocks
714 Details : After ELC code landed, it is now improper to enqueue any mds
715 locks under och_sem, because enqueue might want to decide to
716 cancel open locks for same inode we are holding och_sem for.
720 Description: Client eviction while running blogbench
721 Details : A lot of unlink operations with concurrent I/O can lead to a
722 deadlock causing evictions. To address the problem, the number of
723 oustanding OST_DESTROY requests is now throttled to
724 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
725 callbacks are processed in priority.
729 Description: enable ACLs on MDS by default
730 Details : ACLs must be enabled on MDS by default.
733 Frequency : PPC/PPC64 only
735 Description: conflicts between asm-ppc64/types.h and lustre_types.h
736 Details : fix duplicated definitions between asm-ppc64/types.h and
737 lustre_types.h on PPC.
740 Frequency : PPC/PPC64 only
742 Description: asm-ppc/segment.h does not exist
743 Details : fix compile issue on PPC.
747 Description: better handle error messages in extents code
750 Frequency : RHEL4 only
752 Description: mkfs is very slow on IA64/RHEL4
753 Details : A performance regression has been discovered in the MPT Fusion
754 driver between versions 3.02.73rh and 3.02.99.00rh. As a
755 consequence, we have downgraded the MPT Fusion driver in the RHEL4
756 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
758 Severity : enhancement
760 Description: SNMP support enhancement
761 Details : Adding total number of sampled request for an MDS node in snmp
764 Severity : enhancement
766 Description: Optimize ldlm waiting list processing for PR extent locks
767 Details : When processing waiting list for read extent lock and meeting
769 lock that is same or wider to it that is not contended, skip
770 processing rest of the list and immediatelly return current
771 status of conflictness, since we are guaranteed there are no
772 conflicting locks in the rest of the list.
776 Description: Time out and refuse to reconnect
777 Details : When the failover node is the primary node, it is possible
778 to have two identical connections in imp_conn_list. We must
779 compare not conn's pointers but NIDs, otherwise we can defeat
780 connection throttling.
782 --------------------------------------------------------------------------------
784 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
786 * Support for kernels:
787 2.6.5-7.283 (SLES 9),
788 2.6.9-55.EL (RHEL 4),
789 2.6.16.46-0.14 (SLES 10),
790 2.6.18.8 vanilla (kernel.org)
791 * Client support for unpatched kernels:
792 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
793 2.6.16 - 2.6.22 vanilla (kernel.org)
794 * Due to recently discovered recovery problems, we do not recommend
795 using patchless RHEL 4 clients with this or any earlier release.
796 * Recommended e2fsprogs version: 1.39.cfs8
797 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
798 * Starting with this release, the ldiskfs backing filesystem required
799 by Lustre is now in its own package, lustre-ldiskfs. This package
800 should be installed. It is versioned separately from Lustre and
801 may be released separately in future.
805 Description: block reactivating mgc import until all deactivates complete
806 Details : Fix race when failing back MDT/MGS to itself (testing)
809 Frequency : at statup only
811 Description: mds_lov_synchronize race leads to various problems
812 Details : simultaneous MDT->OST connections at startup can cause the
813 sync to abort, leaving the OSC in a bad state.
815 Severity : enhancement
817 Description: add optional extra BUILD_VERSION info
818 Details : add a new environment variable (namely LUSTRE_VERS) which allows
819 to override the lustre version.
822 Frequency : 2.6.18 servers only
824 Description: ll_kern_mount() doesn't release the module reference
825 Details : The ldiskfs module reference count never drops down to 0
826 because ll_kern_mount() doesn't release the module reference.
831 Description: server LBUG when using old ost_num_threads parameter
832 Details : Accept the old ost_num_threads parameter but warn that it
833 is deprecated, and fix an off-by-one error that caused an LBUG.
838 Description: Transient SCSI error results in persistent IO issue
839 Details : iobuf->dr_error is not reinitialized to 0 between two
843 Frequency : sometimes when underlying device returns I/O errors
845 Description: OSTs not going read-only during write failures
846 Details : OSTs are not remounted read-only when the journal commit threads
847 get I/O errors because fsfilt_ext3 calls journal_start/stop()
848 instead of the ext3 wrappers.
852 Description: poor connect scaling with increasing client count
853 Details : Don't run filter_grant_sanity_check for more than 100 exports
854 to improve scaling for large numbers of clients.
857 Frequency : SLES10 only
859 Description: sanity-quota.sh quotacheck failed: rc = -22
860 Details : Quotas cannot be enabled on SLES10.
863 Frequency : liblustre clients only
865 Description: getdirentries does not give error when run on compute nodes
866 Details : getdirentries does not fail when the size specified as an argument
867 is too small to contain at least one entry
869 Severity : enhancement
871 Description: Add LNET router traceability for debug purposes
872 Details : If a checksum failure occurs with a router as part of the
873 IO path, the NID of the last router that forwarded the bulk data
874 is printed so it can be identified.
879 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
880 Details : Due to a race condition, liblustre clients were occasionally
883 Severity : enhancement
885 Description: lfs setstripe use optional parameters instead of postional
888 Severity : enhancement
890 Description: Nanosecond timestamp support for ldiskfs
891 Details : The on-disk ldiskfs filesystem has added support for nanosecond
892 resolution timestamps. There is not yet support for this at
893 the Lustre filesystem level.
896 Frequency : during server recovery
898 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
899 Details : request with rq_no_resend flag not awake l_wait_event if they get a
903 Frequency : nfs export on patchless client
905 Description: connectathon hang when test nfs export over patchless client
906 Details : Disconnected dentry cannot be found with lookup, so we do not need
907 to unhash it or make it invalid
910 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
911 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
912 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
917 Description: open req refcounting wrong on reconnect
918 Details : If reconnect happened between getting open reply from server and
919 call to mdc_set_replay_data in ll_file_open, we will schedule
920 replay for unreferenced request that we are about to free.
921 Subsequent close will crash in variety of ways.
922 Check that request is still eligible for replay in
923 mdc_set_replay_data().
928 Description: disable writes to filesystem when reading health_check file
929 Details : the default for reading the health_check proc file has changed
930 to NOT do a journal transaction and write to disk, because this
931 can cause reads of the /proc file to hang and block HA state
932 checking on a healthy but otherwise heavily loaded system. It
933 is possible to return to the previous behaviour during configure
934 with --enable-health-write.
936 Severity : enhancement
938 Description: 64-bit inode version
939 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
940 to the inode. This will be required for version-based recovery.
945 Description: MDS fails to start if a duplicate client export is detected
946 Details : in some rare cases it was possible for a client to connect to
947 an MDS multiple times. Upon recovery the MDS would detect this
948 and fail during startup. Handle this more gracefully.
950 Severity : enhancement
952 Description: Add -o localflock option to simulate old noflock
954 Details : This will achieve local-only flock/fcntl locks
960 Description: log_commit_thread vs filter_destroy race leads to crash
961 Details : Take import reference before releasing llog record semaphore
966 Description: Wrong request locking in request set processing
967 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
968 imp_delayed_list, in this place should be used imp_lock.
971 Frequency : when reconnection
973 Description: Grant Leak when osc reconnect to OST
974 Details : When osc reconnect ost, OST(filter) should check whether it
975 should grant more space to client by comparing fed_grant and
976 cl_avail_grant, and return the granted space to client instead
977 of "new granted" space, because client will call osc_init_grant
978 to update the client grant space info.
981 Frequency : when client reconnect to OST
983 Description: Grant Leak when osc do resend and replay bulk write
984 Details : When osc reconnect to OST, OST(filter)should clear grant info of
985 bulk write request, because the grant info while be sync between
986 OSC and OST when reconnect, and we should ignore the grant info
987 these of resend/replay write req.
992 Description: Grant space more than avaiable left space sometimes.
993 Details : When then OST is about to be full, if two bulk writing from
994 different clients came to OST. Accord the avaliable space of the
995 OST, the first req should be permitted, and the second one
996 should be denied by ENOSPC. But if the seconde arrived before
997 the first one is commited. The OST might wrongly permit second
998 writing, which will cause grant space > avaiable space.
1001 Frequency : when client is evicted
1003 Description: Grant might be wrongly erased when osc is evicted by OST
1004 Details : when the import is evicted by server, it will fork another
1005 thread ptlrpc_invalidate_import_thread to invalidate the
1006 import, where the grant will be set to 0. While the original
1007 thread will update the grant it got when connecting. So if
1008 the former happened latter, the grant will be wrongly errased
1009 because of this race.
1014 Description: Checking Stale with correct fid
1015 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1016 to check whether it is stale, because sometimes, we want the
1017 enqueue happened anyway, and op_data.fid2 will not be initialized.
1019 Severity : enhancement
1021 Description: update patchless client
1022 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1025 Frequency : only with 2.4 kernel
1027 Description: random memory corruption
1028 Details : size of struct ll_inode_info is to big for union inode.u and this
1029 can be cause of random memory corruption.
1034 Description: Memory leak in recovery
1035 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1036 It should also check obd_fail before fsfilt_start, otherwise if
1037 fsfilt_start return -EROFS,(failover mds during mds recovery).
1038 then the req will return with repmsg->transno = 0 and rc = EROFS.
1039 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1040 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1041 be freed no matter whether fsfilt_commit success or not.
1044 Frequency : only with huge count clients
1046 Description: Prevents from taking the superblock lock in llap_from_page for
1048 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1049 for llap_from_page call in ll_removepage prevents from taking the
1050 superblock lock for a soon died page.
1055 Description: Not check open intent error before release open handle
1056 Details : in some rare cases, the open intent error is not checked before
1057 release open handle, which may cause
1058 ASSERTION(open_req->rq_transno != 0), because it tries to release
1059 the failed open handle.
1064 Description: Set cat log bitmap only after create log success.
1065 Details : in some rare cases, the cat log bitmap is set too early. and it
1066 should be set only after create log success.
1070 Description: Accessing a block bevice can re-enable I/O when Lustre is
1071 tearing down a device.
1072 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1077 Description: service threads may hog cpus when there are a lot of requests
1078 Details : Insert cond_resched to give other threads a chance to use some CPU
1083 Description: the cat log was not initialized in recovery
1084 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1085 unlink log on mds will not be initialized until mds post
1086 recovery. And also in mds post recovery, the unlink log will
1087 initialization will be done asynchronausly, so there will be race
1088 between add unlink log and unlink log initialization.
1092 Description: brw_stats were being printed incorrectly
1093 Details : brw_stats were being printed as log2 but all of them were not
1094 recorded as log2. Also remove some code duplication arising from
1095 filter_tally_{read,write}.
1099 Frequency : rare, only in recovery.
1100 Description: ASSERTION(req->rq_type != LI_POISON) failed
1101 Details : imp_lock should be held while iterating over imp_sending_list for
1102 prevent destroy request after get timeout in ptlrpc_queue_wait.
1106 Description: replay-single.sh test 52 fails
1107 Details : A lock's skiplist need to be cleanup when it being unlinked
1108 from its resource list.
1112 Description: Short directio read returns full requested size rather than
1114 Details : Direct I/O operations should return actual amount of bytes
1115 transferred rather than requested size.
1117 Severity : enhancement
1119 Description: metadata RPC reduction (e.g. for rm performance)
1120 Details : decrease the amount of synchronous RPC between clients and servers
1121 by canceling conflicing lock before the operation on the client
1122 and packing thier handles into the main operation RPC to server.
1124 Severity : enhancement
1126 Description: Async OSC create to avoid the blocking unnecessarily.
1127 Details : If a OST has no remain object, system will block on the creating
1128 when need to create a new object on this OST. Now, ways use
1129 pre-created objects when available, instead of blocking on an
1130 empty osc while others are not empty. If we must block, we block
1131 for the shortest possible period of time.
1133 Severity : enhancement
1135 Description: refine locking for avoid write wrong info into lov_objid file
1136 Details : fix possible races with add new target and write/update data in
1139 --------------------------------------------------------------------------------
1141 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1146 Frequency : on some architectures
1148 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1149 Details : Uninitialized flags sometimes cause configuration commands to
1153 Frequency : patchless clients only
1155 Description: missing __iget() symbol export
1156 Details : The __iget() symbol export is missing. To avoid the need for
1157 this on patchless clients the deathrow inode reaper is turned
1158 off, and we depend on the VM to clean up old inodes. This
1159 dependency was during via the fix for bug 12181.
1161 --------------------------------------------------------------------------------
1163 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1165 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1166 INTEROPERATE with older versions automatically. In many cases a
1167 special upgrade step is needed. Please read the
1168 user documentation before upgrading any part of a live system.
1169 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1170 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1171 * WARNING: Lustre configuration and startup changes are required with
1172 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1178 Severity : enhancement
1180 Description: MountConf
1181 Details : Lustre configuration is now managed via mkfs and mount
1182 commands instead of lmc and lconf. New obd types (MGS, MGC)
1183 are added for dynamic configuration management. See
1184 https://mail.clusterfs.com/wikis/lustre/MountConf for
1187 Severity : enhancement
1189 Description: dynamic OST addition
1190 Details : OSTs can now be added to a live filesystem
1192 Severity : enhancement
1194 Description: startup order invariance
1195 Details : MDTs and OSTs can be started in any order. Clients only
1196 require the MDT to complete startup.
1198 Severity : enhancement
1200 Description: parallel, asynchronous orphan cleanup
1201 Details : orphan cleanup is now performed in separate threads for each
1202 OST, allowing parallel non-blocking operation.
1204 Severity : enhancement
1206 Description: optimized stripe assignment
1207 Details : stripe assignments are now made based on ost space available,
1208 ost previous usage, and OSS previous usage, in order to try
1209 to optimize storage space and networking resources.
1211 Severity : enhancement
1213 Description: Permanently set tunables
1214 Details : All writable /proc/fs/lustre tunables can now be permanently
1215 set on a per-server basis, at mkfs time or on a live system.
1217 Severity : enhancement
1219 Description: Lustre message v2
1220 Details : Add lustre message format v2.
1222 Severity : enhancement
1224 Description: client OST exclusion list
1225 Details : Clients can be started with a list of OSTs that should be
1226 declared "inactive" for known non-responsive OSTs.
1230 Description: SPEC SFS validation failure on NFS v2 over lustre.
1231 Details : Changes the blocksize for regular files to be 2x RPC size,
1232 and not depend on stripe size.
1234 Severity : enhancement
1236 Description: Multiple MD RPCs in flight.
1237 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1238 To avoid overly-overloading MDS, introduce a limit on number of
1239 MDS RPCs in flight for a single client and add /proc controls
1240 to adjust this limit.
1242 Severity : enhancement
1244 Description: client read/write statistics
1245 Details : Add client read/write call usage stats for performance
1246 analysis of user processes.
1247 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1248 file access. extents_stats shows chunk size distribution.
1249 extents_stats_per_process show chunk size distribution per
1252 Severity : enhancement
1254 Description: mds statistics
1255 Details : Add detailed mds operations statistics in
1256 /proc/fs/lustre/mds/*/stats.
1260 Description: Failure of copying files with lustre special EAs.
1261 Details : Client side always return success for setxattr call for lustre
1262 special xattr (currently only "trusted.lov").
1267 Description: Refcount LNET uuids
1268 Details : The global LNET uuid list grew linearly with every startup;
1269 refcount repeated list entries instead of always adding to
1272 Severity : enhancement
1274 Description: Dynamic service threads
1275 Details : Within a small range, start extra service threads
1276 automatically when the request queue builds up.
1279 Frequency : mixed-endian client/server environments
1281 Description: mixed-endian crashes
1282 Details : The new msg_v2 system had some failures in mixed-endian
1285 Severity : enhancement
1287 Description: Easy OST removal
1288 Details : OSTs can be permanently deactivated with e.g. 'lctl
1289 conf_param lustre-OST0001.osc.active=0'
1291 Severity : enhancement
1293 Description: MGS proc entries
1294 Details : Added basic proc entries for the MGS showing what filesystems
1297 Severity : enhancement
1299 Description: provide MGS failover
1300 Details : Added config lock reacquisition after MGS server failover.
1302 Severity : enhancement
1304 Description: add Linux 2.4 support
1305 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1309 Description: a large application tries to do I/O to the same resource and dies
1310 in the middle of it.
1311 Details : Check the req->rq_arrival time after the call to
1312 ost_brw_lock_get(), but before we do anything about
1313 processing it & sending the BULK transfer request. This
1314 should help move old stale pending locks off the queue as
1315 quickly as obd_timeout.
1318 Frequency : when an incorrect nid is specified during startup
1320 Description: ptlrpc connect to non-existant node causes kernel crash
1321 Details : LNET can't be re-entered from an event callback, which
1322 happened when we expire a message after the export has been
1323 cleaned up. Instead, hand the zombie cleanup off to another
1326 Severity : enhancement
1328 Description: plain/inodebits lock performance improvement
1329 Details : Grouping plain/inodebits in granted list by their request modes
1330 and bits policy, thus improving the performance of search through
1334 Frequency : only if OST filesystem is corrupted
1336 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1337 Details : for a short time RPCs with bulk IO are in the replay list,
1338 but replay of bulk IOs is unimplemented. If the OST filesystem
1339 is corrupted due to disk cache incoherency and then replay is
1340 started it is possible to trip an assertion. Avoid putting
1341 committed RPCs into the replay list at all to avoid this issue.
1344 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1346 Description: System hang on startup
1347 Details : This bug allowed the liblustre (e.g. catamount) client to
1348 return to the app before handling all startup RPCs. This
1349 could leave the node unresponsive to lustre network traffic
1350 and manifested as a server ptllnd timeout.
1352 Severity : enhancement
1354 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1355 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1356 boolean to control whether to print peer debug info when a
1357 client's RPC times out.
1360 Frequency : only for kernels with patches from Lustre below 1.4.3
1362 Description: Remove old rdonly API
1363 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1366 Frequency : only for devices with external journals
1368 Description: Set external device read-only also
1369 Details : During a commanded failover stop, we set the disk device
1370 read-only while the server shuts down. We now also set any
1371 external journal device read-only at the same time.
1374 Frequency : when upgrading from 1.4 while trying to change parameters
1376 Description: The wrong (new) MDC name was used when setting parameters for
1377 upgraded MDT's. Also allows changing of OSC (and MDC)
1378 parameters if --writeconf is specified at tunefs upgrade time.
1381 Frequency : when setting specific ost indicies
1383 Description: QOS code breaks on skipped indicies
1384 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1385 created with --index need not be sequential.
1387 Severity : enhancement
1389 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1390 Details : The uninit_groups feature works in conjunction with the kernel
1391 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1392 up the pass1 processing of e2fsck. This is a read-only feature
1393 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1394 mount filesystems that have had this feature enabled.
1396 Severity : enhancement
1398 Description: Improve multi-block allocation algorithm to avoid fragmentation
1399 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1400 allocation locality and avoid filesystem fragmentation.
1402 ------------------------------------------------------------------------------
1404 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1406 * Support for kernels:
1407 2.4.21-47.0.1.EL (RHEL 3)
1408 2.6.5-7.283 (SLES 9)
1409 2.6.9-42.0.10.EL (RHEL 4)
1410 2.6.12.6 vanilla (kernel.org)
1411 2.6.16.27-0.9 (SLES 10)
1412 * Recommended e2fsprogs version: 1.39.cfs5
1414 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1418 Frequency : occasional, depends on client load and configuration
1419 Bugzilla : 12181, 12203
1420 Description: data loss for recently-modified files
1422 Details : In some cases it is possible that recently written or created
1423 files may not be written to disk in a timely manner (this should
1424 normally be within 30s unless client IO load is very high).
1425 The problem appears as zero-length files or files that are a
1426 multiple of 1MB in size after a client crash or client eviction
1427 that are missing data at the end of the file.
1429 This problem is more likely to be hit on clients where files are
1430 repeatedly created and unlinked in the same directory, clients
1431 have a large amount of RAM, have many CPUs, the filesystem has
1432 many OSTs, the clients are rebooted frequently, and/or the files
1433 are not accessed by other nodes after being written.
1435 The presence of the problem can be detected by looking at
1436 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1437 smaller than the second (nr_unused) then dirty files will not
1438 be flushed automatically to disk. "sync; sleep 10" should be
1439 run several times on the node before unmounting it to update
1440 Lustre (this is also safe to run on nodes without this problem).
1442 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1443 that can cause this same problem, so customers using that kernel
1444 also need to update the kernel in addition to Lustre. In order
1445 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1447 It is normal that files written just before a client crash (less
1448 than 30s) may not yet have been flushed to disk, even for local
1452 Frequency : frequent on thin XT3 nodes
1454 Description: UUID collision on thin XT3 Linux nodes
1455 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1456 randomly, since we relied on an insufficiently-seeded PRNG.
1461 Description: OSS hangs after "All ost request buffers busy"
1462 Details : A deadlock between quota and journal operations caused OSS
1463 hangs after printing "All ost request buffers busy."
1466 Frequency : always on liblustre builds
1468 Description: Cleanup compiler warnings on liblustre
1471 Frequency : always on liblustre builds on XT3
1473 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1477 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1478 > 32000 subdirectories
1479 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1480 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1481 subdirectory count crosses 32000. This will aid e2fsck to
1482 correctly handle more than 32000 subdirectories.
1485 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1487 Description: System hang on startup
1488 Details : This bug allowed the liblustre (e.g. catamount) client to
1489 return to the app before handling all startup RPCs. This
1490 could leave the node unresponsive to lustre network traffic
1491 and manifested as a server ptllnd timeout.
1493 Severity : enhancement
1495 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1496 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1497 boolean to control whether to print peer debug info when a
1498 client's RPC times out.
1503 Description: make O_SYNC working on 2.6 kernels
1504 Details : 2.6 kernels use different method for mark pages for write,
1505 so need add a code to lustre for O_SYNC work.
1510 Description: Failure to close file and release space on NFS
1511 Details : Put inode details into lock acquired in ll_intent_file_open.
1512 Use mdc_intent_lock in ll_intent_open to properly
1513 detect all kind of errors unhandled by mdc_enqueue.
1518 Description: proc file read during shutdown sometimes raced obd removal,
1520 Details : Add lock to prevent obd access after proc file removal.
1523 Frequency : Only for files larger than 4GB on 32-bit clients.
1525 Description: improperly doing page alignment of locks
1526 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1527 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1532 Description: RPCs being resent when they shouldn't be
1533 Details : Some RPCs that should not be resent are being resent. This
1534 can cause inconsistencies in the RPC state machine. Do not
1535 resend such requests.
1538 Frequency : rare, only with NFS export
1540 Description: Crash on NFS re-export node
1541 Details : under very unusual load conditions an assertion is hit in
1542 ll_intent_file_open()
1545 Frequency : only if OST filesystem is corrupted
1547 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1548 Details : for a short time RPCs with bulk IO are in the replay list,
1549 but replay of bulk IOs is unimplemented. If the OST filesystem
1550 is corrupted due to disk cache incoherency and then replay is
1551 started it is possible to trip an assertion. Avoid putting
1552 committed RPCs into the replay list at all to avoid this issue.
1557 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1558 Details : Large single O_DIRECT read and write calls can fail to allocate
1559 a sufficiently large buffer to process the request. In case of
1560 allocation failure the allocation is retried with a smaller
1561 buffer and broken into smaller requests.
1563 Severity : enhancement
1565 Description: Add -o localflock option to simulate old noflock behaviour.
1566 Details : This will achieve local-only flock/fcntl locks coherentness.
1571 Description: versioning check is incomplete
1572 Details : Checking the version difference of client vs. server, report
1573 error if the gap is too big.
1578 Description: add support PG_writeback bit
1579 Details : add support for PG_writeback bit for Lustre, for more carefull
1580 work with page cache in 2.6 kernel. This also fix some deadlocks
1581 and remove hack for work O_SYNC with 2.6 kernel.
1583 Severity : enhancement
1585 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1586 Details : The uninit_groups feature works in conjunction with the kernel
1587 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1588 up the pass1 processing of e2fsck. This is a read-only feature
1589 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1590 mount filesystems that have had this feature enabled.
1592 Severity : enhancement
1594 Description: Improve multi-block allocation algorithm to avoid fragmentation
1595 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1596 allocation locality and avoid filesystem fragmentation.
1598 ------------------------------------------------------------------------------
1600 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1602 * Support for kernels:
1603 2.6.9-42.0.3.EL (RHEL 4)
1604 2.6.5-7.276 (SLES 9)
1605 2.4.21-47.0.1.EL (RHEL 3)
1606 2.6.12.6 vanilla (kernel.org)
1607 2.6.16.21-0.8 (SLES10)
1608 * Recommended e2fsprogs version: 1.39.cfs2-0
1610 * The backwards-compatible /proc/sys/portals symlink has been removed
1611 in this release. Before upgrading, please ensure that you change
1612 any configuration scripts or /etc/sysctl.conf files that access
1613 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1614 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1615 in advance of the upgrade on any system running Lustre 1.4.6 or
1616 newer, since /proc/sys/lnet was added in that version.
1617 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1621 Frequency : only when quota is used
1623 Description: avoid scanning export list for quota master
1624 Details : Change the algorithms to avoid scanning export list in order
1625 to improve the efficiency.
1628 Frequency : MDS failover only, very rarely
1630 Description: "went back in time" messages on mds failover
1631 Details : The greatest transno may be lost when the current operation
1632 finishes with an error (transno==0) and the client's last_rcvd
1633 record is over-written. Save the greatest transno in the
1634 mds_last_transno for this case.
1637 Frequency : always for specific kernels and striping counts
1639 Description: client may get "Matching packet too big" without ACL support
1640 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1641 when trying to access files in certain configurations. The
1642 clients should in fact be denied when mounting because they do
1643 not understand ACLs.
1646 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1648 Description: many clients connecting with IO in progress causes connect timeouts
1649 Details : Avoid synchronous journal commits to avoid delays caused by many
1650 clients connecting/disconnecting when bulk IO is in progress.
1651 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1652 OST_IO_PORTAL to avoid delays behind potentially many pending
1656 Frequency : occasionally with multiple writers to a single file
1658 Description: shared writes to file may result in wrong size reported by stat()
1659 Details : Allow growing of kms when extent lock is cancelled
1662 Frequency : always with random mmap IO to multi-striped file
1664 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1665 Details : Only if the hole is at the end of OST object so that kms is too
1666 small. Fix is to increase kms accordingly in ll_nopage.
1669 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1671 Description: writes to a missing object would leak memory on the OST
1672 Details : If there is an inconsistency between the MDS and OST filesystems,
1673 such that the MDS references an object that doesn't exist, writes
1674 to that object will leak memory due to incorrect cleanup in the
1675 error handling path, eventually running out of memory on the OST.
1680 Description: Creating too long symlink causes lustre errors
1681 Details : Check symlink and name lengths before sending requests to MDS.
1684 Frequency : only if flock is enabled (not on by default)
1686 Description: posix locks not released on fd closure on 2.6.9+
1687 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1688 kernels, this caused such locks not to be released on fd close and
1689 then assertions on fs unmount about still used locks.
1692 Frequency : MDS failover only, very rarely
1694 Description: clients may get ASSERTION(granted_lock != NULL)
1695 Details : When request was taking a long time, and a client was resending
1696 a getattr by name lock request. The were multiple lock requests
1697 with the same client lock handle and
1698 mds_getattr_name->fixup_handle_for_resent_request found one of the
1699 lock handles but later failed with ASSERTION(granted_lock != NULL).
1704 Description: handle->h_buffer_credits > 0, assertion failure
1705 Details : h_buffer_credits is zero after truncate, causing assertion
1706 failure. This patch extends the transaction or creates a new
1710 Frequency : NFS re-export or patchless client
1711 Bugzilla : 11179, 10796
1712 Description: Crash on NFS re-export node (__d_move)
1713 Details : We do not want to hash the dentry if we don't have a lock.
1714 But if this dentry is later used in d_move, we'd hit uninitialised
1715 list head d_hash, so we just do this to init d_hash field but
1716 leave dentry unhashed.
1719 Frequency : NFS re-export or patchless client
1721 Description: NFS exports has problem with symbolic link
1722 Details : lustre client didn't properly install dentry when re-exported
1723 to NFS or running patchless client.
1726 Frequency : NFS re-export or patchless client
1728 Description: Various nfs/patchless fixes.
1729 Details : fixes reuse disconected alias for lookup process - this fixes
1730 warning "find_exported_dentry: npd != pd",
1731 fix permission error with open files at nfs.
1732 fix apply umask when do revalidate.
1735 Frequency : occasional
1737 Description: Crash on NFS re-export node
1738 Details : calling clear_page() on the wrong pointer triggered oops in
1739 generic_mapping_read().
1742 Frequency : rarely, using O_DIRECT IO
1744 Description: unaligned directio crashes client with LASSERT
1745 Details : check for unaligned buffers before trying any requests.
1748 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1750 Description: stale data returned from RAID cache
1751 Details : If only a small amount of IO is done to the RAID device before
1752 reading it again it is possible to get stale data from the RAID
1753 cache instead of reading it from disk.
1756 Frequency : always for sles10 kernel
1758 Description: sles10 support
1759 Details : ll_follow_link: compile fixes and using of nd_set_link
1760 under newer kernels.
1763 Frequency : depends on arch, kernel and compiler version, always on sles10
1766 Description: recursive or deep enough symlinks cause stack overflow
1767 Details : getting rid of large stack-allocated variable in
1771 Frequency : depends on hardware
1773 Description: lustre write performance loss in the SLES10 kernel
1774 Details : the performance loss is caused by using of write barriers in the
1775 ext3 code. The SLES10 kernel turns barrier support on by
1776 default. The fix is to undo that change for ldiskfs.
1778 ------------------------------------------------------------------------------
1780 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1782 * Support for kernels:
1783 2.6.9-42.0.3EL (RHEL 4)
1784 2.6.5-7.276 (SLES 9)
1785 2.4.21-47.0.1.EL (RHEL 3)
1786 2.6.12.6 vanilla (kernel.org)
1790 Frequency : quota enabled and large files being deleted
1792 Description: releasing more than 4GB of quota at once hangs OST
1793 Details : If a user deletes more than 4GB of files on a single OST it
1794 will cause the OST to spin in an infinite loop. Release
1795 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1800 Description: statfs data retrieved from /proc may be stale or zero
1801 Details : When reading per-device statfs data from /proc, in the
1802 {kbytes,files}_{total,free,avail} files, it may appear
1803 as zero or be out of date.
1806 Frequency : systems with MD RAID1 external journal devices
1808 Description: lconf's call to blkid is confused by RAID1 journal devices
1809 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1810 of returning all block devices that match the journal UUID.
1813 Frequency : always, for aggregate stripe size over 4GB
1815 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1816 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1817 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1820 Frequency : always if "lfs find" used on a local file/directory
1822 Description: "lfs find" segfaults if used on a local file/directory
1823 Details : The case where a directory component was not specified wasn't
1824 handled correctly. Handle this properly.
1827 Frequency : always on ppc64
1829 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1830 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1831 perform endianness conversion.
1834 Frequency : rarely (truncate to non-zero file size after write under load)
1835 Bugzilla : 10730, 10687
1836 Description: Files padded with zeros to next 4K multiple
1837 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1838 it is possible that files that are truncated to a non-zero size
1839 immediately after being written are filled with zero bytes beyond
1840 the truncated size. No file data is lost.
1842 Severity : enhancement
1844 Description: Allow recovery/failover for liblustre clients.
1845 Details : liblustre clients were unaware of failover configurations until
1848 Severity : enhancement
1850 Description: user file locks should fail when not mounting with flock option
1851 Details : Set up an error-returning stub in ll_file_operations.lock field
1852 to prevent incorrect behaviour when client is mounted without
1853 flock option. Also, set up properly f_op->flock field for
1857 Frequency : always on ia64
1859 Description: "lfs df" loops on printing out MDS statfs information
1860 Details : The obd_ioctl_data was not initialized and in some systems
1861 this caused a failure during the ioctl that did not return
1862 an error. Initialize the struct and return an error on failure.
1865 Frequency : SLES 9 only
1867 Description: Error of copying files with lustre special EAs as root
1868 Details : Client side always return success for setxattr call for lustre
1869 special xattr (currently only "trusted.lov").
1872 Frequency : rarely on clusters with both ia64+i386 clients
1874 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1875 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1876 clients are doing concurrent writes to the same file it is
1877 possible that the ia64 clients may LASSERT because the OST
1878 extent locks are not PAGE_SIZE aligned. Ensure that grown
1879 locks are always aligned on the request boundary.
1882 Frequency : specific use, occasional
1884 Description: Overwriting in use executable truncates on-disk binary image
1885 Details : If one node attempts to overwrite an executable in use by
1886 another node, we now correctly return ETXTBSY instead of
1887 truncating the file.
1889 Severity : enhancement
1891 Description: Async OSC create to avoid the blocking unnecessarily.
1892 Details : If a OST has no remain object, system will block on the creating
1893 when need to create a new object on this OST. Now, ways use
1894 pre-created objects when available, instead of blocking on an
1895 empty osc while others are not empty. If we must block, we block
1896 for the shortest possible period of time.
1901 Description: chmod on Lustre root is propagated to other clients
1902 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1903 invalid by the follow_mount time.
1906 Frequency : liblustre clients only
1908 Description: Race in 'instant cancel' lock handling could lead to such locks
1909 never to be granted in case of SMP MDS
1910 Details : Do not destroy not yet granted but cbpending locks in
1914 Frequency : replay/resend of open
1916 Description: non null lock assetion failure in mds_intent_policy
1917 Details : Trying to replay/resend lockless open requests resulted in
1918 mds_open() returning 0 with no lock. Now it sets a flag if
1919 a lock is going to be returned.
1921 Severity : enhancement
1923 Description: Checksum enhancements
1924 Details : New checksum enhancements allow for resending RPCs that failed
1927 Severity : enhancement
1929 Description: Tunables on number of dirty pages in cacche
1930 Details : Allow to set limit on number of dirty pages cached.
1935 Description: client crash on unmount - lock still has references
1936 Details : In some error handling cases it was possible to leak a lock
1937 reference on a client while accessing a file. This was not
1938 harmful to the client during operation, but would cause the
1939 client to crash when the filesystem is unmounted.
1942 Frequency : specific case, rare
1944 Description: ETXTBSY on mds though file not in use by client
1945 Details : ETXTBSY is no longer incorrectly returned when attempting to
1946 chmod or chown a directory that the user previously tried to
1947 execute or a currently-executing binary.
1950 Frequency : extremely rare except on liblustre-based clients
1952 Description: Lustre space not freed when files are deleted
1953 Details : Clean up open-unlinked files after client eviction. Previously
1954 the unlink was skipped and the files remained as orphans.
1959 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1960 Details : In some cases it was possible to send a blocking callback to
1961 a client doing a glimpse, even though that client didn't get
1962 a lock granted. When the glimpse lock is cancelled on the OST
1963 the freed lock is left on the waiting list and corrupted the list.
1966 Frequency : all core dumps
1968 Description: Broke core dumps to lustre
1969 Details : Negative dentry may be unhashed if parent does not have UPDATE
1970 lock, but some callers, e.g. do_coredump, expect dentry to be
1971 hashed after successful create, hash it in ll_create_it.
1973 ------------------------------------------------------------------------------
1975 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1977 * Support for kernels:
1978 2.6.9-42.0.2.EL (RHEL 4)
1979 2.6.5-7.276 (SLES 9)
1980 2.4.21-40.EL (RHEL 3)
1981 2.6.12.6 vanilla (kernel.org)
1985 Frequency : always on RHEL 3
1987 Description: Number of open files grows over time
1988 Details : The number of open files grows over time, whether or not
1989 Lustre is started. This was due to a filp leak introduced
1990 by one of our kernel patches.
1992 ------------------------------------------------------------------------------
1994 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1996 * Support for kernels:
1997 2.6.9-42.EL (RHEL 4)
1998 2.6.5-7.276 (SLES 9)
1999 2.4.21-40.EL (RHEL 3)
2000 2.6.12.6 vanilla (kernel.org)
2005 Bugzilla : 5719, 9635, 9792, 9684
2006 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2007 Details : If a server is under heavy load and cannot reply to new
2008 connection requests before the client resends the (re)connect,
2009 the connection handling code can behave badly if two service
2010 threads are concurrently handing separate (re)connections from
2011 the same client. Add better locking to the connection handling
2012 code, and ensure that only a single connection will be processed
2013 for a given client UUID, even if the lock is dropped.
2015 Severity : enhancement
2017 Description: add TCP zero-copy support to kernel
2018 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2019 sends if the hardware supports scatter-gather and checksumming.
2020 This allows socklnd to do client-write and server-read more
2021 efficiently and reduce CPU utilization from skbuf copying.
2024 Frequency : only if NFS exporting from client
2026 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2027 Details : If setting attributes on a file created under NFS that had
2028 never been opened it would be possible to oops the client
2029 if the file had no objects.
2032 Frequency : always for liblustre
2034 Description: liblustre client does MDS+OSTs setattr RPC for each write
2035 Details : When doing a write from a liblustre client, the client
2036 incorrectly issued an RPC to the MDS and each OST the file was
2037 striped over in order to update the timestamps. When writing
2038 with small chunks and many clients this could overwhelm the MDS
2039 with RPCs. In all cases it would slow down the write because
2040 these RPCs are unnecessary.
2042 Severity : enhancement
2044 Description: allow number of MDS service threads to be changed at module load
2045 Details : It is now possible to change the number of MDS service threads
2046 running. Adding "options mds mds_num_threads={N}" to the MDS's
2047 /etc/modprobe.conf will set the number of threads for the next
2048 time Lustre is restarted (assuming the "mds" module is also
2049 reloaded at that time). The default number of threads will
2050 stay the same, 32 for most systems.
2055 Description: OST crash if filesystem is unformatted or corrupt
2056 Details : If an OST is started on a device that has never been formatted
2057 or if the filesystem is corrupt and cannot even mount then the
2058 error handling cleanup routines would dereference a NULL pointer.
2063 Description: NULL pointer deref in llap_from_page.
2064 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2065 incorrect) mapping if the page was truncated/reclaimed while it was
2066 searched for. Check for this condition and skip such pages when
2067 doing readahead. Introduce extra check to llap_from_page() to
2068 verify page->mapping->host is non-NULL (so page is not anonymous).
2071 Frequency : Sometimes when using sys_sendfile
2073 Description: "page not covered by a lock" warnings from ll_readpage
2074 Details : sendfile called ll_readpage without right page locks present.
2075 Now we introduced ll_file_sendfile that does necessary locking
2076 around call to generic_file_sendfile() much like we do in
2080 Frequency : with certain MDS communication failures at client mount time
2082 Description: NULL pointer deref after failed client mount
2083 Details : a client connection request may delayed by the network layer
2084 and not be sent until after the PTLRPC layer has timed out the
2085 request. If the client fails the mount immediately it will try
2086 to clean up before the network times out the request. Add a
2087 reference from the request import to the obd device and delay
2088 the cleanup until the network drops the request.
2091 Frequency : occasionally during client (re)connect
2093 Description: assertion failure during client (re)connect
2094 Details : processing a client connection request may be delayed by the
2095 client or server longer than the client connect timeout. This
2096 causes the client to resend the connection request. If the
2097 original connection request is replied in this interval, the
2098 client may trip an assertion failure in ptlrpc_connect_interpret()
2099 which thought it would be the only running connect process.
2102 Frequency : only with obd_echo servers and clients that are rebooted
2104 Description: kernel BUG accessing uninitialized data structure
2105 Details : When running an obd_echo server it did not start the ping_evictor
2106 thread, and when a client was evicted an uninitialized data
2107 structure was accessed. Start the ping_evictor in the RPC
2108 service startup instead of the OBD startup.
2110 Severity : enhancement
2111 Bugzilla : 10193 (patchless)
2112 Description: Remove dependency on various unexported kernel interfaces.
2113 Details : No longer need reparent_to_init, exit_mm, exit_files,
2114 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2117 Frequency : rare (only users of deprecated and unsupported LDAP config)
2119 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2120 Details : LDAP apparently contains 'lustreName' attributes instead of
2121 'name'. A simple remapping of the name is sufficient.
2124 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2126 Description: waiting_locks_callback trips kernel BUG if client is evicted
2127 Details : Running with the dump_on_timeout debug flag turned on makes
2128 it possible that the waiting_locks_callback() can try to dump
2129 the Lustre kernel debug logs from an interrupt handler. Defer
2130 this log dumping to the expired_lock_main() thread.
2132 Severity : enhancement
2134 Description: Support NFS exporting on 2.6 kernels.
2135 Details : Implement non-rawops metadata methods for NFS server to use without
2136 changing NFS server code.
2139 Frequency : very rare (synthetic metadata workload only)
2141 Description: two racing renames might cause an MDS thread to deadlock
2142 Details : Running the "racer" program may cause one MDS thread to rename
2143 a file from being the source of a rename to being the target of
2144 a rename at exactly the same time that another thread is doing
2145 so, and the second thread has already enqueued these locks after
2146 doing a lookup of the target and is trying to relock them in
2147 order. Ensure that we don't try to re-lock the same resource.
2150 Frequency : only very large systems with liblustre clients
2152 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2153 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2154 all OSTs in parallel. This allows the eviction of stale liblustre
2155 clients to proceed much faster than if they were done in series,
2156 and also offers similar improvements for other set_info RPCs.
2161 Description: excessive CPU usage during initial read phase on client
2162 Details : During the initial read phase on a client, it would agressively
2163 retry readahead on the file, consuming too much CPU and impacting
2164 performance (since 1.4.5.8). Improve the readahead algorithm
2165 to avoid this, and also improve some other common cases (read
2166 of small files in particular, where "small" is files smaller than
2167 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2172 Description: MDS crash when receiving packet with unknown intent.
2173 Details : Do not LBUG in unknown intent case, just return -EFAULT
2175 Severity : enhancement
2176 Bugzilla : 9293, 9385
2177 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2178 Details : Do not serialize getattr (non-intent version) and statfs.
2181 Frequency : occasional, when OST network is overloaded/intermittent
2183 Description: client evicted by OST after bulk IO timeout
2184 Details : If a client sends a bulk IO request (read or write) the OST
2185 may evict the client if it is unresposive to its data GET/PUT
2186 request. This is incorrect if the network is overloaded (takes
2187 too long to transfer the RPC data) or dropped the OST GET/PUT
2188 request. There is no need to evict the client at all, since
2189 the pinger and/or lock callbacks will handle this, and the
2190 client can restart the bulk request.
2193 Frequency : Always when mmapping file with no objects
2195 Description: client crashes when mmapping file with no objects
2196 Details : Check that we actually have objects in a file before doing any
2197 operations on objects in ll_vm_open, ll_vm_close and
2203 Description: Request leak when working with deleted CWD
2204 Details : Introduce advanced request refcount tracking for requests
2205 referenced from lustre intent.
2207 Severity : Enhancement
2209 Description: Cache open file handles on client.
2210 Details : MDS now will return special lock along with openhandle, if
2211 requested and client is allowed to hold openhandle, even if unused,
2212 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2213 closing files for every read/write openration.
2215 Severity : Enhancement
2217 Description: Cache open negative dentries on client when possible.
2218 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2219 negative dentries on lock revocation.
2224 Description: Remounting a client read-only wasn't possible with a zconf mount
2225 Details : It wasn't possible to remount a client read-only with llmount.
2227 Severity : enhancement
2228 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2229 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2230 MPICH is included to add Lustre-specific ADIO interfaces.
2231 This is based closely on the UFS ADIO layer and only differs
2232 in file creation, in order to allow the OST striping to be set.
2233 This is user-contributed code and not supported by CFS.
2238 Description: extended inode attributes (immutable, append-only) work improperly
2239 when 2.4 and 2.6 kernels are used on client/server or vice versa
2240 Details : Introduce kernel-independent values for these flags.
2242 Severity : enhancement
2245 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2246 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2247 granularity. Now, Fractional MB granularity is be supported,
2248 this is very useful for low memory system.
2250 Severity : enhancement
2252 Description: Getattr by fid
2253 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2254 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2255 have locks and avoids getattr rpc requests for it on every stat.
2258 Frequency : Always, for filesystems larger than 2TB
2260 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2261 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2262 filesystems larger than 2TB to be mounted with mballoc enabled.
2265 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2267 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2268 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2269 filesystems larger than 2TB is mounted then the kernel will
2270 silently corrupt the start of the filesystem. CONFIG_LBD is
2271 enabled for all CFS-supported kernels, but the possibility of
2272 this happening with a modified kernel config exists.
2274 Severity : enhancement
2276 Description: add client O_DIRECT support for 2.6 kernels
2277 Details : It is now possible to do O_DIRECT reads and writes to files
2278 in the Lustre client mountpoint on 2.6 kernel clients.
2280 Severity : enhancement
2282 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2283 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2284 parallel, not waiting for response from every OST before sending
2285 a rpc to the next OST.
2290 Description: setattr vs write race when updating file timestamps
2291 Details : Client processes that update a file timestamp into the past
2292 right after writing to the file (e.g. tar) it is possible that
2293 the updated file modification time can be reset to the current
2294 time due to a race between processing the setattr and write RPC.
2296 Severity : enhancement
2298 Description: Bring 'lfs find' closer in line with regular Linux find.
2299 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2300 -print0 options and obtains all the needed info through the lustre
2303 Severity : enhancement
2305 Description: support up to 1024 configured devices on one node
2306 Details : change obd_dev array from statically allocated to dynamically
2307 allocated structs as they are first used to reduce memory usage
2312 Description: Flush dirty partially truncated pages during truncate
2313 Details : Immediatelly flush partially truncated pages in filter_setattr,
2314 this way we completely avoid having any pages in page cache on OST
2315 and can retire ugly workarounds during writes to flush such pages.
2320 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2321 Details : protect inode from truncation within vfs_unlink() context
2322 just take a reference before calling vfs_unlink() and release it
2323 when parent's i_sem is free.
2328 Description: last_id value checked outside lock on OST caused LASSERT failure
2329 Details : If there were multiple MDS->OST object precreate requests in
2330 flight, it was possible that the OST's last object id was checked
2331 outside a lock and incorrectly tripped an assertion. Move checks
2332 inside locks, and discard old precreate requests.
2335 Frequency : always, if extents are used on OSTs
2337 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2338 Details : OSTs using the extents format would not zero the high 16 bits of
2339 the index physical block number. This is not a problem for any
2340 OST filesystems smaller than 16TB, and no kernels support ext3
2341 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2342 new/modified files) and can be fixed for existing filesystems
2343 with e2fsprogs-1.39-cfs1.
2348 Description: import connection selection may be incorrect if timer wraps
2349 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2350 import connections to be ignored if the 32-bit jiffies counter
2351 wraps. Use a 64-bit jiffies counter.
2354 Frequency : very large clusters immediately after boot
2356 Description: LNET request buffers exhausted under heavy short-term load
2357 Details : If a large number of client requests are generated on a service
2358 that has previously never seen so many requests it is possible
2359 that the request buffer growth cannot keep up with the spike in
2360 demand. Instead of dropping incoming requests, they are held in
2361 the LND until the RPC service can accept more requests.
2364 Frequency : Sometimes during replay
2366 Description: Assertion failure in ll_local_open after replay.
2367 Details : If replay happened on an open request reply before we were able
2368 to set replay handler, reply will become not swabbed tripping the
2369 assertion in ll_local_open. Now we set the handler right after
2370 recognising of open request
2373 Frequency : very rare
2375 Description: kernel reports "badness in vsnprintf"
2376 Details : Reading from the "recovery_status" /proc file in small chunks
2377 may cause a negative length in lprocfs_obd_rd_recovery_status()
2378 call to vsnprintf() (which is otherwise harmless). Exit early
2379 if there is no more space in the output buffer.
2381 Severity : enhancement
2383 Description: clear OBD RPC statistics by writing to them
2384 Details : It is now possible to clear the OBD RPC statistics by writing
2385 to the "stats" file.
2390 Description: Client mtime is not the same on different clients after utimes
2391 Details : In some cases, the client was using the utimes() syscall on
2392 a file cached on another node. The clients now validate the
2393 ctime from the MDS + OSTs to determine which one is right.
2398 Description: Inability to activate failout mode
2399 Details : lconf script incorrectly assumed that in pythong string's numeric
2400 value is used in comparisons.
2403 Frequency : always with multiple stripes per file
2405 Description: Inefficient object allocation for mutli-stripe files
2406 Details : When selecting which OSTs to stripe files over, for files with
2407 a stripe count that divides evenly into the number of OSTs,
2408 the MDS is always picking the same starting OST for each file.
2409 Return the OST selection heuristic to the original design.
2414 Description: mount failures may take full timeout to return an error
2415 Details : Under some heavy load conditions it is possible that a
2416 failed mount can wait for the full obd_timeout interval,
2417 possibly several minutes, before reporting an error.
2418 Instead return an error as soon as the status is known.
2420 Frequency : quota enabled and large files being deleted
2422 Description: releasing more than 4GB of quota at once hangs OST
2423 Details : If a user deletes more than 4GB of files on a single OST it
2424 will cause the OST to spin in an infinite loop. Release
2425 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2430 Description: statfs data retrieved from /proc may be stale or zero
2431 Details : When reading per-device statfs data from /proc, in the
2432 {kbytes,files}_{total,free,avail} files, it may appear
2433 as zero or be out of date.
2436 Frequency : systems with MD RAID1 external journal devices
2438 Description: lconf's call to blkid is confused by RAID1 journal devices
2439 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2440 of returning all block devices that match the journal UUID.
2443 Frequency : always, for aggregate stripe size over 4GB
2445 Description: assertion fails when trying to use 4GB stripe size
2446 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2447 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2450 Frequency : always on ppc64
2452 Description: the first write on an ext3 filesystem with mballoc got stuck
2453 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2454 perform endianness conversion.
2456 ------------------------------------------------------------------------------
2458 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2460 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2461 INTEROPERATE with older versions automatically. Please read the
2462 user documentation before upgrading any part of a live system.
2463 * WARNING: Lustre networking configuration changes are required with
2464 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2467 * Support for kernels:
2468 2.6.9-22.0.2.EL (RHEL 4)
2469 2.6.5-7.244 (SLES 9)
2470 2.6.12.6 vanilla (kernel.org)
2473 Severity : enhancement
2474 Bugzilla : 7981/8208
2475 Description: Introduced Lustre Networking (LNET)
2476 Details : LNET is new networking infrastructure for Lustre, it includes
2477 a reorganized network configuration mode (see the user
2478 documentation for full details) as well as support for routing
2479 between different network fabrics. Lustre Networking Devices
2480 (LNDs) for the supported network fabrics have also been
2481 created for this new infrastructure.
2483 Severity : enhancement
2484 Description: Introduced Access control lists
2485 Details : clients can set ACLs on files and directories in order to have
2486 more fine-grained permissions than the standard Unix UGO+RWX.
2487 The MDS must be started with the "-o acl" mount option.
2489 Severity : enhancement
2490 Description: Introduced filesystem quotas
2491 Details : Administrators may now establish per-user quotas on the
2494 Severity : enhancement
2496 Description: Configuration change for the XT3
2497 The PTLLND is now used to run Lustre over Portals on the XT3
2498 The configure option(s) --with-cray-portals are no longer used.
2499 Rather --with-portals=<path-to-portals-includes> is used to
2500 enable building on the XT3. In addition to enable XT3 specific
2501 features the option --enable-cray-xt3 must be used.
2506 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2507 Details : A race between allocating a new llcd and re-getting the llcd_lock
2508 allowed another thread to grab newly-allocated llcd.
2510 Severity : enhancement
2512 Description: 2.6 OST async journal commit and locking fix to improve performance
2513 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2514 2.6 kernels are now async as they already were in 2.4 kernels so
2515 that they can commit concurrently with the network bulk transfer.
2516 For block-allocated files the filter allocation semaphore is held
2517 to avoid filesystem fragmentation during allocation. BKL lock
2518 removed for 2.6 xattr operations where it is no longer needed.
2523 Description: lconf incorrectly determined whether two IP networks could talk
2524 Details : In some more complicated routing and multiple-network
2525 configurations, lconf will avoid trying to make a network
2526 connection to a disjoint part of the IP space. It was doing the
2527 math incorrectly for one set of cases.
2532 Description: Fix for potential infinite loop processing records in an llog.
2533 Details : If an llog record is corrupted/zeroed, it is possible to loop
2534 forever in llog_process(). Validate the llog record length
2535 and skip the remainder of the block on error.
2538 Frequency : occasional (liblustre only)
2540 Description: liblustre could not open files whose last component is a symlink
2541 Details : sysio_path_walk() would incorrectly pass the open intent to
2542 intermediate path components.
2545 Frequency : rare (liblustre only with non-standard tuning)
2546 Bugzilla : 7201 (7350)
2547 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2548 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2549 after it has been released, only do this for referenced locks
2551 Severity : enhancement
2553 Description: specifying an (invalid) directory default stripe_size of -1
2554 would reset the directory default striping
2555 Details : stripe_size -1 was used internally to signal directory stripe
2556 removal, now use "all default" to signal dir stripe removal
2557 as a directory striping of "all default" is not useful
2560 Frequency : common for large clusters running liblustre clients
2562 Description: doing an ls when liblustre clients are running is slow
2563 Details : sending a glimpse AST to a liblustre client waits for every AST
2564 to time out, as liblustre clients will not respond. Since they
2565 cannot cache data we refresh the OST lock LVB from disk instead.
2567 Severity : enhancement
2569 Description: doing an ls at the same time as file IO can be slow
2570 Details : enqueue and other "small" requests can be blocked behind many
2571 large IO requests. Create a new OST IO portal for non-IO
2572 requests so they can be processed faster.
2575 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2577 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2578 created file with mode 000
2579 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2580 allow an ftruncate() as a non-root user to a file with mode 000.
2581 Lustre now respects this flag to disable mode checks when
2582 truncating a file owned by the user
2585 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2588 Description: Revoking locks from clients that went dead or catatonic might take
2590 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2591 cancellation of such locks instant on servers without waiting for
2592 any reply from clients. Clients drops these locks when cancel
2593 notification from server is received without replying.
2596 Frequency : liblustre-only, when liblustre client dies or becomes busy
2598 Description: Doing ls on Linux clients can take a long time with active
2600 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2601 granting such locks to it in the first place if possible. Locks
2602 are taken by proxy on the OST during the read or write and
2603 dropped immediately afterward. Add connect flags handling, do
2604 not grant locks to liblustre clients for glimpse ASTs.
2606 Severity : enhancement
2608 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2609 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2610 sequential reads and behaves badly with applications doing random
2611 reads. Improve it by reading ahead at least read region, and
2612 avoiding excessive large RPC for small reads.
2614 Severity : enhancement
2616 Description: Creating more than 1000 files for a single job may cause a load
2617 imbalance on the OSTs if there are also a large number of OSTs.
2618 Details : qos_prep_create() uses an OST index reseed value that is an
2619 even multiple of the number of available OSTs so that if the
2620 reseed happens in the middle of the object allocation it will
2621 still utilize the OSTs as uniformly as possible.
2626 Description: OST or MDS may oops in ping_evictor_main()
2627 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2628 but doesn't restart at beginning of obd_exports_timed list
2631 Severity : enhancement
2633 Description: improve by-nid export eviction on the MDS and OST
2634 Details : allow multiple exports with the same NID to be evicted at one
2635 time without re-searching the exports list.
2638 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2640 Description: MDS may oops in groups_free()
2641 Details : in rare race conditions a newly allocated group_info struct is
2642 freed again, and this can be NULL. The 2.4 compatibility code
2643 for groups_free() checked for a NULL pointer, but 2.6 did not.
2646 Frequency : common for liblustre clients doing little filesystem IO
2647 Bugzilla : 9352, 7313
2648 Description: server may evict liblustre clients accessing contended locks
2649 Details : if a client is granted a lock or receives a completion AST
2650 with a blocking AST already set it would not reply to the AST
2651 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2654 Frequency : lfs setstripe, only systems with more than 160 OSTs
2656 Description: unable to set striping with a starting offset beyond OST 160
2657 Details : llapi_create_file() incorrectly limited the starting stripe
2658 index to the maximum single-file stripe count.
2661 Frequency : LDAP users only
2663 Description: lconf did not handle in-kernel recovery with LDAP properly
2664 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2666 Severity : enhancement
2668 Description: bind OST threads to NUMA nodes to improve performance
2669 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2670 node and do their allocations there to localize memory access
2672 Severity : enhancement
2674 Description: llmount can determine client NID directly from Myrinet (GM)
2675 Details : the client NID code from gmnalnid was moved directly into
2676 llmount, removing the need to use this or specifying the
2677 client NID explicitly when mounting GM clients with zeroconf
2680 Frequency : if client is started with down MDS
2682 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2683 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2684 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2685 ptlrpc_import_delayed_req() because we didn't check if the
2686 request was interrupted, and we also didn't break out of the
2687 event loop if there was no timeout
2692 Description: data loss during non-page-aligned writes to a single file from
2693 both multiple nodes and multiple threads on one node at same time
2694 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2695 inconsistency led to false short-reads, that were cached and later
2696 used by ->prepare_write() to fill in partially written page,
2697 leading to data loss.
2700 Frequency : always, if lconf --abort_recovery used
2702 Description: lconf --abort_recovery fails with 'Operation not supported'
2703 Details : lconf was attempting to abort recovery on the MDT device and not
2706 Severity : enhancement
2708 Description: remove cleanup logs
2709 Details : replace lconf-generated cleanup logs with lustre internal
2710 cleanup routines. Eliminates the need for client-cleanup and
2713 Severity : enhancement
2715 Description: add support for EAs (user and system) on lustre filesystems
2716 Details : it is now possible to store extended attributes in the Lustre
2717 client filesystem, and with the user_xattr mount option it
2718 is possible to allow users to store EAs on their files also
2720 Severity : enhancement
2722 Description: Add possibility (config option) to show minimal available OST free
2724 Details : When compiled with --enable-mindf configure option, statfs(2)
2725 (and so, df) will return least minimal free space available from
2726 all OSTs as amount of free space on FS, instead of summary of
2727 free spaces of all OSTs.
2729 Severity : enhancement
2731 Description: do not expand extent locks acquired on OST-side
2732 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2733 by server: they are not cached anyway.
2736 Frequency : when mmap is used/binaries executed from Lustre
2738 Description: Unmmap pages before throwing them away from read cache.
2739 Details : llap_shrink cache now attempts to unmap pages before discarding
2740 them (if unmapping failed - do not discard). SLES9 kernel has
2741 extra checks that trigger if this unmapping is not done first.
2746 Description: lconf didn't resolve symlinks before checking to see whether a
2747 given mountpoint was already in use
2750 Frequency : when migrating failover services
2751 Bugzilla : 6395, 9514
2752 Description: When migrating a subset of services from a node (e.g. failback
2753 from a failover service node) the remaining services would
2754 time out and evict clients.
2755 Details : lconf --force (implied by --failover) sets the global obd_timeout
2756 to 5 seconds in order to quickly disconnect, but this caused
2757 other RPCs to time out too quickly. Do not change the global
2758 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2760 Severity : enhancement
2761 Frequency : if MDS is started with down OST
2762 Bugzilla : 9439,5706
2763 Description: Allow startup/shutdown of an MDS without depending on the
2764 availability of the OSTs.
2765 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2766 Add appropriate locking and lov-osc refcounts for safe
2767 cleaning. Add osc abort_inflight calls in case the
2768 synchronize never started.
2771 Frequency : occasional (Cray XT3 only)
2773 Description: root not authorized to access files in CRAY_PORTALS environment
2774 Details : The client process capabilities were not honoured on the MDS in
2775 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2776 been accessed by an authorized user then root was able to access
2777 the file on the local client also. The root user capabilities
2778 are now allowed on the MDS, as this environment has secure UID.
2781 Frequency : occasional
2783 Description: ldiskfs "too long searching" message happens too often
2784 Details : A debugging message (otherwise harmless) prints too often on
2785 the OST console. This has been reduced to only happen when
2786 there are fragmentation problems on the filesystem.
2791 Description: Division by zero in statfs when all OSCs are inactive
2792 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2793 lov_statfs divides by value returned by lov_get_stripecnt().
2797 Bugzilla : 9489, 3273
2798 Description: First write from each client to each OST was only 4kB in size,
2799 to initialize client writeback cache, which caused sub-optimal
2800 RPCs and poor layout on disk for the first writen file.
2801 Details : Clients now request an initial cache grant at (re)connect time
2802 and so that they can start streaming writes to the cache right
2803 away and always do full-sized RPCs if there is enough data.
2804 If the OST is rebooted the client also re-establishes its grant
2805 so that client cached writes will be honoured under the grant.
2810 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2811 Details : Now I/O RPCs go to different portal number and (presumably) fast
2812 lock requests (and glimses) and other RPCs get their own service
2813 threads pool that should be able to service those RPCs
2816 Severity : enhancement
2818 Description: Ability to exchange lustre version between client and servers and
2819 issue warnings at client side if client is too old. Also for
2820 liblustre clients there is ability to refuse connection of too old
2822 Details : New 'version' field is added to connect data structure that is
2823 filled with version info. That info is later checked by server and
2827 Frequency : rare, liblustre only.
2828 Bugzilla : 9296, 9581
2829 Description: Two simultaneous writes from liblustre at offset within same page
2830 might proceed at the same time overwriting eachother with stale
2832 Details : I/O lock withing llu_file_prwv was released too early, before data
2833 actually was hitting the wire. Extended lock-holding time until
2834 server acknowledges receiving data.
2837 Frequency : extremely rare. Never observed in practice.
2839 Description: avoid generating lustre_handle cookie of 0.
2840 Details : class_handle_hash() generates handle cookies by incrementing
2841 global counter, and can hit 0 occasionaly (this is unlikely, but
2842 not impossible, because initial value of cookie counter is
2843 selected randonly). Value of 0 is used as a sentinel meaning
2844 "unassigned handle" --- avoid it. Also coalesce two critical
2845 sections in this function into one.
2847 Severity : enhancement
2849 Description: allow liblustre clients to delegate truncate locking to OST
2850 Details : To avoid overhead of locking, liblustre client instructs OST to
2851 take extent lock in ost_punch() on client's behalf. New connection
2852 flag is added to handle backward compatibility.
2854 Severity : enhancement
2855 Bugzilla : 4928, 7341, 9758
2856 Description: allow number of OST service threads to be specified
2857 Details : a module parameter allows the number of OST service threads
2858 to be specified via "options ost ost_num_threads={N}" in the
2859 OSS's /etc/modules.conf or /etc/modprobe.conf.
2863 Bugzilla : 6146, 9635, 9895
2864 Description: servers crash with bad pointer in target_handle_connect()
2865 Details : In rare cases when a client is reconnecting it was possible that
2866 the connection request was the last reference for that export.
2867 We would temporarily drop the export reference and get a new
2868 one, but this may have been the last reference and the export
2869 was just destroyed. Get new reference before dropping old one.
2871 Severity : enhancement
2872 Frequency : if client is started with failover MDS
2874 Description: Allow multiple MDS hostnames in the mount command
2875 Details : Try to read the configuration from all specified MDS
2876 hostnames during a client mount in case the "primary"
2879 Severity : enhancement
2881 Description: Stop sending data to evicted clients as soon as possible.
2882 Details : Check if the client we are about to send or are sending data to
2883 was evicted already. (Check is done every second of waiting,
2884 for which l_wait_event interface was extended to allow checking
2885 of exit condition at specified intervals).
2888 Frequency : rare, normally only when NFS exporting is done from client
2890 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2892 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2895 Frequency : always, if srand() is called before liblustre initialization
2897 Description: Liblustre uses system PRNG disturbing its usage by user application
2898 Details : Introduce internal to lustre fast and high-quality PRNG for
2899 lustre usage and make liblustre and some other places in generic
2900 lustre code to use it.
2902 Severity : enhancement
2903 Bugzilla : 9477, 9557, 9870
2904 Description: Verify that the MDS configuration logs are updated when xml is
2905 Details : Check if the .xml configuration logs are newer than the config
2906 logs stored on the MDS and report an error if this is the case.
2907 Request --write-conf, or allow starting with --old_conf.
2909 Severity : enhancement
2911 Description: Handle symlinks in the path when checking if Lustre is mounted.
2912 Details : Resolve intermediate symlinks when checking if a client has
2913 mounted a filesystem to avoid duplicate client mounts.
2918 Description: lconf can hit an error exception but still return success.
2919 Details : The lconf command catches the Command error exception at the top
2920 level script context and will exit with the associated exit
2921 status, but doesn't ensure that this exit status is non-zero.
2926 Description: failure of ptlrpc thread startup can cause oops
2927 Details : Starting a ptlrpc service thread can fail if there are a large
2928 number of threads or the server memory is very fragmented.
2929 Handle this without oopsing.
2932 Frequency : always, only if liblustre and non-default acceptor port was used
2934 Description: liblustre cannot connect to servers with non-default acceptor port
2935 Details : tcpnal_set_default_params() was not called and was therefore
2936 ignoring the environment varaible TCPNAL_PORT, as well as other
2937 TCPNAL_ environment variables
2942 Description: two objects could be created on the same OST for a single file
2943 Details : If an OST is down, in some cases it was possible to create two
2944 objects on a single OST for a single file. No problems other
2945 than potential performance impact and spurious error messages.
2949 Bugzilla : 5681, 9562
2950 Description: Client may oops in ll_unhash_aliases
2951 Details : Client dcache may become inconsistent in race condition.
2952 In some cases "getcwd" can fail if the current directory is
2958 Description: Inode refcounting problems in NFS export code
2959 Details : link_raw functions used to call d_instantiate without obtaining
2960 extra inode reference first.
2964 Bugzilla : 9942, 9903
2965 Description: Referencing freed requests leading to crash, memleaks with NFS.
2966 Details : We used to require that call to ll_revalidate_it was always
2967 followed by ll_lookup_it. Also with revalidate_special() it is
2968 possible to call ll_revalidate_it() twice for the same dentry
2969 even if first occurence returned success. This fix changes semantic
2970 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2971 reference on a request referred from the intent.
2972 ll_intent_release() then releases such a request.
2975 Frequency : rare, normally benchmark loads only
2977 Description: unlinked inodes were kept in memory on the client
2978 Details : If a client is repeatedly creating and unlinking files it
2979 can accumulate a lot of stale inodes in the inode slab cache.
2980 If there is no other client load running this can cause the
2981 client node to run out of memory. Instead flush old inodes
2982 from client cache that have the same inode number as a new inode.
2985 Frequency : SLES9 2.6.5 kernel and long filenames only
2986 Bugzilla : 9969, 10379
2987 Description: utime reports stale NFS file handle
2988 Details : SLES9 uses out-of-dentry names in some cases, which confused
2989 the lustre dentry revalidation. Change it to always use the
2993 Frequency : rare, unless heavy write-truncate concurrency is continuous
2994 Bugzilla : 4180, 6984, 7171, 9963, 9331
2995 Description: OST becomes very slow and/or deadlocked during object unlink
2996 Details : filter_destroy() was holding onto the parent directory lock
2997 while truncating+unlinking objects. For very large objects this
2998 may block other threads for a long time and slow overall OST
2999 responsiveness. It may also be possible to get a lock ordering
3000 deadlock in this case, or run out of journal credits because of
3001 the combined truncate+unlink. Solution is to do object truncate
3002 first in one transaction without parent lock, and then do the
3003 final unlink in a new transaction with the parent lock. This
3004 reduces the lock hold time dramatically.
3007 Frequency : rare, 2.4 kernels only
3009 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3010 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3011 before unmounting their filesystem in order to avoid deadlock.
3012 The kernel_locked() function in 2.4 kernels only checks whether
3013 the kernel lock is held, not whether it is this process that is
3014 holding it as 2.6 kernels do.
3019 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3020 Details : The client ptlrpc code may be trying to reconnect to a down
3021 server before a previous connection attempt has timed out.
3022 Increase the reconnect interval to be longer than the connection
3023 timeout interval to avoid sending duplicate connections to
3027 Frequency : echo_client brw_test command
3029 Description: fix echo_client to work with OST preallocated code
3030 Details : OST preallocation code (5137) didn't take echo_client IO path
3031 into account: echo_client calls filter methods outside of any
3032 OST thread and, hence, there is no per-thread preallocated
3033 pages and buffers to use. Solution: hijack pga pages for IO. As
3034 a byproduct, this avoids unnecessary data copying.
3038 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3039 Description: Client can oops in mdc_commit_close() after open replay
3040 Details : It was possible for the MDS to return an open request with no
3041 transaction number in mds_finish_transno() if the client was
3042 evicted, but without actually returning an error. Clients
3043 would later try to replay that open and may trip an assertion
3044 Simplify the client close codepath, and always return an error
3045 from the MDS in case the open is not successful.
3048 Frequency : rare, 2.6 OSTs only
3050 Description: OST may deadlock under high load on fragmented files
3051 Details : If there was a heavy load and highly-fragmented OST filesystems
3052 it was possible to have all the OST threads deadlock waiting on
3053 allocation of biovecs, because the biovecs were not released
3054 until the entire RPC IO was completed. Instead, release biovecs
3055 as soon as they are complete to ensure forward IO progress.
3057 Severity : enhancement
3059 Description: Support for specifying external journal device at mount
3060 Details : If an OST or MDS device is formatted with an external journal
3061 device, this device major/minor is stored in the ext3 superblock
3062 and may not be valid for failover. Allow detecting and
3063 specifying the external journal at mount time.
3068 Description: Mounting an MDS with pending unlinked files may cause oops
3069 Details : target_finish_recovery() calls mds_postrecov() which returned
3070 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3071 considers this an error and immediately begins cleaning up the
3072 lov, just after starting the mds_lov process
3074 Severity : enhancement
3076 Description: Implement 'lfs df' to report actual free space on per-OST basis
3077 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3078 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3079 usage of objects; '-h' to report in human readable format.
3081 ------------------------------------------------------------------------------
3083 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3090 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3091 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3092 at startup during orphan unlinking. Instead, skip buddy update
3093 before setup, as it will be regenerated after recovery is complete.
3098 Description: If an OST is inactive, its locks might reference stale inodes.
3099 Details : lov_change_cbdata() must iterate over all namespaces, even if
3100 they are inactive to clear inode references from the lock.
3102 Severity : enhancement
3103 Frequency : occasional, if non-standard max_dirty_mb used
3105 Description: Client will block write RPCs if not enough grant
3106 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3107 then the client will block writes while waiting for another RPC
3108 to complete instead of consuming its dirty limit. With change
3109 we get improved performance when max_dirty_mb is small.
3111 Severity : enhancement
3112 Bugzilla : 3389, 6253
3113 Description: Add support for supplementary groups on the MDS.
3114 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3115 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3116 lookups for user supplementary groups into a cache.
3120 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3121 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3122 for liblustre O_EXCL is enforced by MDS.
3126 Description: readdir never returns NULL in liblustre.
3127 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3128 dentry in struct dirent.
3132 Frequency : liblustre only, depends on application IO pattern
3133 Description: liblustre clients evicted if not contacting servers
3134 Details : Don't put liblustre clients into the ping_evictor list, so
3135 they will not be evicted by the pinger ever.
3137 Severity : enhancement
3139 Description: Add ability to evict clients by NID from MDS.
3140 Details : By echoing "nid:$NID" string into
3141 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3142 $NID would be instantly evicted from this MDS and from all active
3143 OSTs connected to it.
3147 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3148 Details : lookup_it_finish() used to query file size from OSTs that was not
3153 Description: service threads change working directory to that of init
3154 Details : Starting lustre service threads may pin the working directory
3155 of the parent thread, making that filesystem busy. Threads
3156 now change to the working directory of init to avoid this.
3160 Frequency : during shutdown only
3161 Description: shutdown with a failed MDS or OST can cause unmount to hang
3162 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3167 Frequency : 2.6 only
3168 Description: chmod/chown may include an extra supplementary group
3169 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3170 supplementary group and if none is specified this is used.
3173 Bugzilla : 5479 (6816)
3175 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3176 Details : If lookup is in progress on a file that is unlinked we might try
3177 to revalidate the inode and fail in revalidate after lookup is
3178 complete and ll_file_open() enqueues the open again but
3179 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3182 Frequency : always, if lconf --abort_recovery used
3184 Description: lconf --abort_recovery fails with 'Operation not supported'
3185 Details : lconf was attempting to abort recovery on the MDT device and not
3188 ------------------------------------------------------------------------------
3190 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3195 Frequency : rare (only unsupported configurations with a node running as an
3197 Bugzilla : 6514, 5137
3198 Description: Mounting a Lustre file system on a node running as an OST could
3200 Details : OSTs now preallocates memory needed to write out data at
3201 startup, instead of when needed, to avoid having to
3202 allocate memory in possibly low memory situations.
3203 Specifically, if the file system is mounted on on OST,
3204 memory pressure could force it to try to write out data,
3205 which it needed to allocate memory to do. Due to the low
3206 memory, it would be unable to do so and the node would
3207 become unresponsive.
3209 Severity : enhancement
3211 Description: Addition of lconf --service command line option
3212 Details : lconf now accepts a '--service <arg>' option, which is
3213 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3215 Severity : enhancement
3217 Description: Failover mode is now the default for OSTs.
3218 Details : By default, OSTs will now run in failover mode. To return to
3219 the old behaviour, add '--failout' to the lmc line for OSTs.
3221 Severity : enhancement
3223 Description: Health checks are now provided for MDS and OSTs
3224 Details : Additional detailed health check information on MSD and OSTs
3225 is now provided through the procfs health_check value.
3228 Frequency : occasional, depends on IO load
3230 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3231 after numerous create/delete cycles
3232 Details : The ext3 inode allocation policy would not allocate new inodes
3233 very well on the OSTs because there are no new directories
3234 being created. Instead we look for groups with free space if
3235 the parent directories are nearly full.
3240 Description: Network or server problems during mount may cause partially
3241 mounted clients instead of returning an error.
3242 Details : The config llog parsing code may overwrite the error return
3243 code during mount error handling, returning success instead
3249 Description: MDS can fail to allocate large reply buffers
3250 Details : After long uptimes the MDS can fail to allocate large reply
3251 buffers (e.g. zconf client mount config records) due to memory
3252 fragmentation or consumption by the buffer cache. Preallocate
3253 some large reply buffers so that these replies can be sent even
3254 under memory pressure.
3258 Frequency : rare (liblustre)
3259 Description: fsx running with liblustre complained that using truncate() to
3260 extend the file doesn't work. This patch corrects that issue.
3261 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3262 ATTR_SIZE and lsm use in llu_setattr_raw.
3266 Frequency : rare, only 2.6 kernels
3267 Description: Unusual file access patterns on the MDS may result in inode
3268 data being lost in very rare circumstances.
3269 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3270 code in the 2.6 kernel caused the inode and/or EA data not to
3271 be read from disk, causing single-file corruption.
3275 Frequency : rare, only 2.6 filesystems using extents
3276 Description: Heavy concurrent write and delete load may cause data corruption.
3277 Details : It was possible under high-load situations to have an extent
3278 metadata block in the block device cache from a just-unlinked
3279 file overwrite a newly-allocated data block. We now unmap any
3280 metadata buffers that alias just-allocated data blocks.
3284 Frequency : filesystems with default stripe_count larger than 77
3285 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3286 Details : lconf specifies an inode size of 4096 bytes when the default
3287 stripe_count is larger than 77. This conflicts with the default
3288 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3289 this case to avoid pinning too much memory for large EAs.
3291 ------------------------------------------------------------------------------
3293 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3298 Frequency : rare (extremely heavy IO load with hundreds of clients)
3300 Description: Client is evicted, gets IO error writing to file
3301 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3302 caused clients to be evicted for AST timeouts. The fixes in
3303 bug 5192 mean we no longer need to have such short AST timeouts
3304 so ldlm_timeout has been increased.
3307 Frequency : occasional during --force or --failover shutdown under load
3308 Bugzilla : 5949, 4834
3309 Description: Server oops/LBUG if stopped with --force or --failover under load
3310 Details : a collection of import/export refcount and cleanup ordering
3311 issues fixed for safer force cleanup
3314 Frequency : only filesystems larger than 120 OSTs
3315 Bugzilla : 5990, 6223
3316 Description: lfs getstripe would oops on a very large filesystem
3317 Details : lov_getconfig used kfree on vmalloc'd memory
3320 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3321 Bugzilla : 6242, 6243
3322 Description: reading from files that had been truncated to a non-zero size
3323 but never opened returned no data
3324 Details : ll_file_read() reads zeros from no-object files to EOF
3329 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3330 Details : There's little harm in aborting MDS/OSS recovery and letting it
3331 try again, so I removed the LASSERT and return an error instead.
3333 Severity : enhancement
3335 Description: New debugging infrastructure for tracking down data corruption
3336 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3337 (b) cover more of the client-side code path, and (c) try to narrow
3338 down where problems occurred
3342 Bugzilla : 3819, 4364, 4397, 6313
3343 Description: Racing close and eviction MDS could cause assertion in mds_close
3344 Details : It was possible to get multiple mfd references during close and
3345 client eviction, leading to one thread referencing a freed mfd.
3347 Severity: : enhancement
3348 Bugzilla : 3262, 6359
3349 Description: Attempts to reconnect to servers are now more aggressive.
3350 Details : This builds on the enhanced upcall-less recovery that was added
3351 in 1.4.2. When trying to reconnect to servers, clients will
3352 now try each server in the failover group every 10 seconds. By
3353 default, clients would previously try one server every 25 seconds.
3358 Description: After recovery, certain operations trigger a failed
3359 assertion on a client.
3360 Details : Failing over an mds, using lconf -d --failover, while a
3361 client was doing a readdir() call would cause the client to
3362 LBUG after recovery completed and the readdir() was resent.
3364 Severity : enhancement
3366 Description: Default groups are now added by lconf
3367 Details : You can now run lconf --group <servicename> without having to
3368 manually add groups with lmc.
3371 Frequency : occasional
3373 Description: Nodes with an elan id of 0 trigger a failed assertion
3376 Frequency : always when accessing e.g. tty/console device nodes
3378 Description: tty and some other devices nodes cannot be used on lustre
3379 Details : file's private_data field is used by device data and lustre
3380 values in there got lost. New field was added to struct file to
3381 store fs-specific private data.
3384 Frequency : when exporting Lustre via NFS
3386 Description: NFSD failed occasionally when looking up a path component
3387 Details : NFSD is looking up ".." which was broken in ext3 directories
3388 that had grown large enough to become hashed.
3391 Frequency : Clusters with multiple interfaces not on the same subnet
3393 Description: Nodes will repeatedly try to reconnect to an interface which it
3394 cannot reach and report an error to the log.
3395 Details : Extra peer list entries will be created by lconf with some peers
3396 unreachable. lconf now validates the peer before adding it.
3399 Frequency : Only if a default stripe is set on the filesystem root.
3401 Description: Setting a default stripe on the filesystem root prevented the
3402 filesystem from being remounted.
3403 Details : The client was sending extra request flags in the root getattr
3404 request and did not allocate a reply buffer for the dir EA.
3407 Frequency : occasional, higher if lots of files are accessed by one client
3408 Bugzilla : 6159, 6097
3409 Description: Client trips assertion regarding lsm mismatch/magic
3410 Details : While revalidating inodes the VFS looks up inodes with ifind()
3411 and in rare cases can find an inode that is being freed.
3412 The ll_test_inode() code will free the lsm during ifind()
3413 when it finds an existing inode and then the VFS later attaches
3414 this free lsm to a new inode.
3418 Bugzilla : 6422, 7030
3419 Description: MDS deadlock between mkdir and client eviction
3420 Details : Creating a new file via mkdir or mknod (starting a transaction
3421 and getting the ns lock) can deadlock with client eviction
3422 (gets ns lock and trying to finish a synchronous transaction).
3425 Frequency : occasional
3426 Description: While starting a server, the fsfilt_ext3 module could not be
3428 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3429 kernels. Previously, lconf would still use the ext3 name
3430 when trying to load modules. Now, it will correctly use
3431 ext3 on 2.4 and ldiskfs on 2.6.
3433 Severity : enhancement
3434 Description: The default stripe count has been changed to 1
3435 Details : The interpretation of the default stripe count (0, to lfs
3436 or lmc) has been changed to mean striping across a single
3437 OST, rather than all available. For general usage we have
3438 found a stripe count of 1 or 2 works best.
3440 Severity : enhancement
3441 Description: Add support for compiling against Cray portals.
3442 Details : Conditional compiling for some areas that are different
3446 Frequency : occasional
3447 Bugzilla : 6409, 6834
3448 Description: Creating files with an explicit stripe count may lead to
3449 a failed assertion on the MDS
3450 Details : If some OSTs are full or unavailable, creating files may
3451 trigger a failed assertion on the MDS. Now, Lustre will
3452 try to use other servers or return an error to the
3456 Frequency : occasional
3458 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3459 caused lock timeout during readahead (since 1.4.2).
3460 Details : Processes doing readahead might match a lock that hasn't been
3461 granted yet if there are overlapping and conflicting lock
3462 requests. The readahead process waits on ungranted lock
3463 (original lock is CBPENDING), while OST waits for that process
3464 to cancel CBPENDING read lock and eventually evicts client.
3466 Severity : enhancement
3468 Description: Initial enabling of flock support for clients
3469 Details : Implements fcntl advisory locking and file status functions.
3470 This feature is provided as an optional mount flag (default
3471 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3472 locking are implemented yet, and those that are are not guaranteed
3473 to be completely correct in production environments.
3474 mount -t lustre -o [flock|noflock] ...
3477 Frequency : occasional
3479 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3480 assertion in the ext3 JBD (journaling) layer.
3481 Details : The b_committed_data struct is protected by the big kernel lock
3482 in 2.4 kernels, serializing journal_commit_transaction() and
3483 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3484 access to this struct. In 2.6 kernels there is finer grained
3485 locking to improve SMP performance of the JBD layer.
3489 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3494 Description: merge and cleanup kernel patches.
3497 -----------------------------------------------------------------------------
3499 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3501 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3502 versions of Lustre. Please update all servers and clients to
3503 version 1.4.2 or later at the same time. You must also run
3504 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3505 to update the configuration logs.
3507 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3508 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3509 - swab llog records of type '0' so we get proper header size/idx (5861)
3510 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3511 - fix rename of one directory over another leaking an inode (5953)
3512 - avoid SetPageDirty on 2.6 (5981)
3513 - don't re-add just-being-destroyed locks to the waiting list (5653)
3514 - when creating new directories, inherit the parent's custom
3515 striping settings if present parent (3048)
3516 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3517 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3518 - don't hold client locks on temporary worklist from l_lru (5666)
3519 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3520 - automatically evict dead clients (5921)
3521 - Update file size properly in create+truncate+fstat case (6196)
3522 - Do not unhash mountpoint dentries, do not allow removal of
3524 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3525 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3526 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3527 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3528 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3529 - Correctly handle reads of files with no objects (6243)
3530 - lctl recover will also mark a device active if deactivate used (5933)
3532 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3533 - allow --write-conf on an MDS with different nettype than client (5619)
3534 - don't write config llogs to MDS for mounts not from that MDS (5617)
3535 - lconf should create multiple TCP connections from a client (5201)
3536 - init scripts are now turned off by default; run chkconfig --on
3537 lustre and chkconfig --on lustrefs to use them
3538 - upcalls are no longer needed for clients to recover to failover
3540 - add --abort-recovery option to lconf to abort recovery on device
3542 - add support for an arbitrary number of OSTs (3026)
3543 - Quota support protocol changes.
3544 - forward compatibility changes to wire structs (6007)
3545 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3546 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3547 - improved "lustre status" script
3548 - initialize blocksize for non-regular files (6062)
3549 - added --disable-server and --disable-client configure options (5782)
3550 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3551 - Vanilla 2.4.29 support
3552 - increase maximum number of obd devices to 520 (6242)
3553 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3554 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3555 SLES 9 2.6.5 kernels
3556 - limit stripes per file to 160 (the maximum EA size) (6093)
3558 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3561 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3562 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3563 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3564 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3565 - small ext3 extents cleanups and fixes (5733)
3566 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3567 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3568 - avoid needless client->OST connect, fix handle mismatch (5317)
3569 - fix DLM error path that led to out-of-sync client, long delays (5779)
3570 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3571 - fix several locking issues related to i_size (5492,5624,5654,5672)
3572 - don't move pending lock onto export if it is already evicted (5683)
3573 - fix kernel oops when creating .foo in unlinked directory (5548)
3574 - fix deadlock in obdfilter statistics vs. object create (5811)
3575 - use time_{before,after} to avoid timer jiffies wrap (5882)
3576 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3577 - Do not leak request if server was not able to process it (5154)
3578 - If mds_open unable to find parent dir, make that negative lookup(5154)
3579 - don't create new directories with extent-mapping (5909, 5936)
3581 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3582 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3583 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3584 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3585 - support for RHEL4 kernel on i686 (5773)
3586 - provide error messages when incompatible logs are encountered (5898)
3588 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3589 * version 1.4.0.10 (1.4.1 release candidate 1)
3591 - don't keep a lock reference when lock is not granted (4238)
3592 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3593 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3594 - drop import inflight refcount on signal_completed_replay error (5255)
3595 - unlock page after async write error during send (3677)
3596 - handle missing objects in filter_preprw_read properly (5265)
3597 - no transno return for symlink open, don't save no-trasno open (3440)
3598 - don't try to complete elan receive that already failed (4012)
3599 - free RPC server reply state on error (5406)
3600 - clean up thread from ptlrpc_start_thread() on error (5160)
3601 - readahead could read extra page into cache that wasn't ejected (5388)
3602 - prevent races in class_attach/setup/cleanup/detach (5260)
3603 - don't dereference de->d_inode after l_dput of de (5458)
3604 - use "int" for stripe value returned from lock_to_stripe (5544)
3605 - mballoc allocation and error-checking fixes in 2.6 (5504)
3606 - block device patches to fix I/O request sizes in 2.6 (5482)
3607 - look up hostnames for IB nals (5602)
3608 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3609 - don't start multiple acceptors for the same port (5277)
3610 - fix incorrect LASSERT in mds_getattr_name (5635)
3611 - export a proc file for general "ping" checking (5628)
3612 - fix "lfs check" to not block when the MDS is down (5628)
3614 - service request history (4965)
3615 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3616 - create an "evict_client" /proc entry on OSTs, like the MDS has
3617 - fix mount usage message, return errors per mount(8) (5168)
3618 - change grep [] to grep "[]" in tests so they work in more UMLs
3619 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3620 - remove ancient LOV_MAGIC_V0 EA support (5047)
3621 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3622 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3623 - allow readahead allocations to fail when low on memory (5383)
3624 - mmap locking landed again, after considerable improvement (2828)
3625 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3627 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3630 - send OST transaction number in read/write reply to free req (4966)
3631 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3632 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3633 - the watchdog thread now runs as interruptible (5246)
3634 - flock/lockf fixes (but it's still disabled, pending 5135)
3635 - don't use EXT3 constants in llite code (5094)
3636 - memory shortage at startup could cause assertion (5176)
3638 - reorganization of lov code
3639 - single portals codebase
3641 - add extents/mballoc support (5025)
3642 - direct I/O reads in the obdfilter (4048)
3643 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3645 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3648 - send OST transaction number in read/write reply to free req (4966)
3649 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3650 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3651 - the watchdog thread now runs as interruptible (5246)
3652 - handle missing objects in filter_preprw_read properly (5265)
3653 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3654 - drop import inflight refcount on signal_completed_replay error (5255)
3655 - unlock page after async write error during send (3677)
3656 - return original error code on reconstructed replies (3761)
3657 - no transno return for symlink open, don't save no-trasno open (3440)
3659 - add pid to ldlm debugging output (4922)
3660 - bump the watchdog timeouts -- we can't handle 30sec yet
3661 - extra debugging for orphan dentry/inode bug (5259)
3663 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3666 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3667 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3668 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3669 - fix NULL dereference in /proc/sys/portals/routes (4827)
3670 - allow failed mdc_close() operations to be interrupted (4561)
3671 - stop precreate on OST before MDS would time out on it (4778)
3672 - don't send partial-page writes before EOF from client (4410)
3673 - discard client grant for sub-page writes on large-page clients (4520)
3674 - don't free dentries not owned by NFS code, check generation (4806)
3675 - fix lsm leak if mds_create_objects() fails (4801)
3676 - limit debug_daemon file size, always print CERROR messages (4789)
3677 - use transno after validating reply (3892)
3678 - process timed out requests if import state changes (3754)
3679 - update mtime on OST during writes, return in glimpse (4829)
3680 - add mkfsoptions to LDAP (4679)
3681 - use ->max_readahead method instead of zapping global ra (5039)
3682 - don't interrupt __l_wait_event() during strace
3684 - add software watchdogs to catch hung threads quickly (4941)
3685 - make lustrefs init script start after nfs is mounted
3686 - fix CWARN/ERROR duplication (4930)
3687 - return async write errors to application if possible (2248)
3688 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3689 - print NAL number in %x format (4645)
3690 - update barely-supported suse-2.4.21-171 series (4842)
3691 - support for sles 9 %post scripts
3692 - support for building 2.6 kernel-source packages
3693 - support for sles km_* packages
3695 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3698 - ignore -ENOENT errors in osc_destroy (3639)
3699 - notify osc create thread that OSC is being cleaned up (4600)
3700 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3701 - reconstruct ost_handle() like mds_handle() (4657)
3702 - create a new thread to do import eviction to avoid deadlock (3969)
3703 - let lconf resolve symlinked-to devices (4629)
3704 - don't unlink "objects" from directory with default EA (4554)
3705 - hold socknal file ref over connect in case target is down (4394)
3706 - allow more than 32000 subdirectories in a single directory (3244)
3707 - fix blocks count for O_DIRECT writes (3751)
3708 - OST returns ENOSPC from object create when no space left (4539)
3709 - don't send truncate RPC if file size isn't changing (4410)
3710 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3711 - bind to privileged port in socknal and tcpnal (3689)
3713 - rate limit CERROR/CWARN console message to avoid overload (4519)
3714 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3715 - basic mmap support (3918)
3716 - kernel patch series update from b1_4 (4711)
3718 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3721 - avoid crash during MDS cleanup with OST shut down (2775)
3722 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3723 - don't use bad inodes on the MDS (3744)
3724 - dynamic object preallocation to improve recovery speed (4236)
3725 - don't hold spinlock over lock dumping or change debug flags (4401)
3726 - don't zero obd_dev when it is force cleaned (3651)
3727 - print grants to console if they go negative (4431)
3728 - "lctl deactivate" will stop automatic recovery attempts (3406)
3729 - look for existing locks in ldlm_handle_enqueue() (3764)
3730 - don't resolve lock handle twice in recovery avoiding race (4401)
3731 - revalidate should check working dir is a directory (4134)
3733 - don't always mark "slow" obdfilter messages as errors (4418)
3735 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3738 - don't close LustreDB during write_conf until it is done (3860)
3739 - fix typo in lconf for_each_profile (3821)
3740 - allow dumping logs from multiple threads at one time (3820)
3741 - don't allow multiple threads in OSC recovery (3812)
3742 - fix debug_size parameters (3864)
3743 - fix mds_postrecov to initialize import for llog ctxt (3121)
3744 - replace config semaphore with spinlock (3306)
3745 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3746 - don't allow enqueue to complete on a destroyed export (3822)
3747 - down write_lock before checking llog header bitmap (3825)
3748 - recover from lock replay timeout (3764)
3749 - up llog sem before sending rpc (3652)
3750 - reduce ns lock hold times when setting kms (3267)
3751 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3752 - fix NULL deref and obd_dev leak on setup error (3312)
3753 - replace some LBUG about llog ops with error handling (3841)
3754 - don't match INVALID dentries from d_lookup and spin (3784)
3755 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3756 - fix invalid assertion in ptlrpc_set_wait (3880)
3758 - add libwrap support for the TCP acceptor (3996)
3759 - add /proc/sys/portals/routes for non-root route listing (3994)
3760 - allow setting MDS UUID in .xml (2580)
3761 - print the stack of a process that LBUGs (4228)
3763 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3766 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3767 - make sure to unset replay flag from failed open requests (3440)
3768 - if default stripe count is 0, use OST count for inode size (3636)
3769 - update parent mtime/ctime on client for create/unlink (2611)
3770 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3771 - free recovery state on server during a forced cleanup (3571)
3772 - unregister_reply for resent reqs (3063)
3773 - loop back devices mounting and status check on 2.6 (3563)
3774 - fix resource-creation race that can provoke i_size == 0 (3513)
3775 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3776 - more debugging for page-accounting assertion (3746)
3777 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3778 - avoid deadlock after precreation failure (3758)
3779 - fix race and lock order deadlock in orphan handling (3450, 3750)
3780 - add validity checks when grabbing inodes from l_ast_data (3599)
3782 - add /proc/.../recovery_status to obdfilter (3428)
3783 - lightweight CDEBUG infrastructure, debug daemon (3668)
3784 - change default OSC RPC parameters to be better on small clusters
3785 - turn off OST read cache for files smaller than 32MB
3786 - install man pages and include them in rpms (3100)
3787 - add new init script for (un)mounting lustre filesystems (2593)
3788 - run chkconfig in %post for init scripts (3701)
3789 - drop scimac NAL (unmaintained)
3791 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3794 - clean kiobufs before and after use (3485)
3795 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3796 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3797 - change init script to fail more clearly if not run as root (1528)
3798 - allow clients to reconnect during replay (1742)
3799 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3800 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3801 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3802 - fix cond_resched (3554)
3803 - only evict unfinished clients after recovery (3515)
3804 - allow bulk resend, prevent data loss (3570)
3805 - dynamic ptlrpc request buffer allocation (2102)
3806 - don't allow unlinking open directory if it isn't empty (2904)
3807 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3808 - remove extraneous obd dereference causing LASSERT failure (3334)
3809 - don't use get_cycles() when creating temp. files on the mds (3156)
3810 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3811 - handle EEXIST for set-stripe, set proper directory name (3336)
3813 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3814 - fix ksocknal_fmb_callback() error messages (2918)
3816 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3819 - don't copy lvb into (possibly NULL) reply on error (2983)
3820 - don't deref dentry after dput, don't free lvb on error (2922)
3821 - use the kms to determine writeback rpc length (2947)
3822 - increment oti_logcookies when osc is inactive (2948)
3823 - update client's i_blocks count via lvb messages (2543)
3824 - handle intent open/close of special files properly (1557)
3825 - mount MDS with errors=remount-ro, like obdfilter (2009)
3826 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3827 - don't use cancelling-locks' kms values (2947)
3828 - use highest lock extent for kms, not last one (2925)
3829 - don't dereference ERR_PTR() dentry in error handling path (3107)
3830 - fix thread race in portals_debug_dumplog() (3122)
3831 - create lprocfs device entries at setup instead of at attach (1519)
3832 - common AST error handler, don't evict client on completion race (3145)
3833 - zero nameidata in detach_mnt in 2.6 (3118)
3834 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3835 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3836 - initialize RPC timeout timer earlier for 2.6 (3219)
3837 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3838 - print nal/nid for unknown nid (3258)
3839 - additional checks for oscc recovery before doing precreate (3284)
3840 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3841 - don't crash in mdc_close for bad permissions on open (3285)
3842 - zero i_rdev for non-device files (3147)
3843 - clear page->private before handing to FS, better assertion (3119)
3844 - tune the read pipeline (3236)
3845 - fix incorrect decref of invalidated dentry (2350)
3846 - provide read-ahead stats and refine rpc in flight stats (3328)
3847 - don't hold journal transaction open across create RPC (3313)
3848 - update atime on MDS at close time (3265)
3849 - close LDAP connection when recovering to avoid server load (3315)
3850 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3851 - don't leak open file on MDS after open resend (3325)
3852 - serialize filter_precreate and filter_destroy_precreated (3329)
3853 - loop device shouldn't call sync_dev() for nul device (3092)
3854 - clear page cache after eviction (2766)
3855 - resynchronize MDS->OST in background (2824)
3856 - refuse to mount the same filesystem twice on same mountpoint (3394)
3857 - allow llmount to create routes for mounting behind routers (3320)
3858 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3859 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3860 - fix rare problem with rename on htree directories (3417)
3862 - allow default OST striping configuration per directory (1414)
3863 - fix compilation for qswnal for 2.6 kernels (3125)
3864 - increase maximum number of MDS request buffers for large systems
3865 - change liblustreapi to be useful for external progs like lfsck (3098)
3866 - increase local configuration timeout for slow disks (3353)
3867 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3869 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3872 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3873 - don't overwrite extent policy data in reply if lock was blocked (2901)
3874 - drop filter export grants atomically with removal from device (2663)
3875 - del obd_self_export from work_list in class_disconnect_exports (2908)
3876 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3877 - swab reply message in mdc_close, other PPC fixes (2464)
3878 - fix destroying of named logs (2325)
3879 - overwrite old logs when running lconf --write_conf (2264)
3880 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3881 - fix race in target_handle_connect (2898)
3882 - mds_reint_create() should take same inode create lock (2926)
3883 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3884 - don't close files for self_export to avoid uninitialized obd (2936)
3885 - allow MDS with the same name as client node (2939)
3886 - hold dentry reference for closed log files for unlink (2325)
3887 - reserve space for all logs during transactions (2059)
3888 - don't evict page beyond end of stripe extent (2925)
3889 - don't oops on a deleted current working directory (2399)
3890 - handle hard links to targets without a parent properly (2517)
3891 - don't dereference NULL lock when racing during eviction (2867)
3892 - don't grow lock extents when lots of conflicting locks (2919)
3894 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3897 - account for cache space usage on clients to avoid data loss (974)
3898 - lfsck support in lustre kernel code (2349)
3899 - reduce journal credits needed for BRW writes (2370)
3900 - orphan handling to avoid losing space on client/server crashes
3901 - ptlrpcd can be blocked, stopping ALL progress (2477)
3902 - use lock value blocks to assist in proper KMS, faster stat (1021)
3903 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3904 - recovery for initial connections (2355)
3905 - fixes for mds_cleanup_orphans (1934)
3906 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3907 - block all file creations until orphan recovery completes (1901)
3908 - client remove rq_connection from request struct (2423)
3909 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3910 - recovery timer race (2670)
3911 - mdc_close recovey bug (2532)
3912 - ptlrpc cleanup bug (2710)
3913 - mds timeout on local locks (2588)
3914 - namespace lock held during RPCs (2431)
3915 - handle interrupted sync write properly (2503)
3916 - don't try to handle a message that hasn't been replied to (2699)
3917 - client assert failure during cleanup after abort recovery (2701)
3918 - leak mdc device after failed mount (2712)
3919 - ptlrpc_check_set allows timedout requests to complete (2714)
3920 - wait for inflight reqs when ptlrpcd finishes (2710)
3921 - make sure unregistered services are removed from the srv_list
3922 - reset bulk XID's when resending them (caught by 1138 test)
3923 - unregister_bulk after timeout
3924 - fix lconf error (2694)
3925 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3926 - readahead locks pages, leaves pending causing memory pressure (2673)
3927 - increase OST request buffers to 4096 on large machines (2729)
3928 - fix up permission of existing directories in simple_mkdir (2661)
3929 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3930 - don't assign transno to errored transactions (2742)
3931 - don't delete objects on OST if given a bogus objid from MDS (2751)
3932 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3933 - if rq_no_resend, then timeout request after recovery (2432)
3934 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3935 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3936 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3937 - MDS recovery shouldn't delete objects that it has given out (2730)
3938 - if enqueue arrives after completion, don't clobber LVB (2819)
3939 - don't unlock pages twice when trigger_group_io returns error (2814)
3940 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3941 - don't write pages to disk if there was an error (1450)
3942 - don't ping imports that have recovery disabled (2676)
3943 - take buffered bytes into account when balancing socknal conn (2817)
3944 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3945 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3946 - remove little-endian swabbing of llog records (1987)
3947 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3948 - retry reposting request buffers if they fail (1191)
3949 - grow extent at grant time to avoid granting a revoked lock (2809)
3950 - lock revoke doesn't evict page if covered by a second lock (2765)
3951 - disable VM readahead to avoid reading outside lock extents (2805)
3953 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3954 - updated kernel patches for hp-2.4.20 kernel (2681)
3956 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3959 - fix truncated write corruption (2366)
3960 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3962 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3963 - ensure len(uuid) < 37 in lmc (1171)
3964 - fix ia64 OOPS in llog_test (2255)
3965 - zero end of page at obdfilter for partial page writes (2648)
3966 - don't leave stale dentries around after renames (bug 2428)
3967 - fix timeouts when evicting a client with a single lock held (2642)
3968 - set deadline for the initial HELLO message to drain (2634)
3969 - print out dotted-quad IP addresses in the socknal (2302)
3971 - additional debugging for MDS client eviction problem (2443)
3972 - fix mkfsoptions support for osts (2603, 2604)
3974 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3977 - add series for the vanilla 2.6.0 kernel
3978 - add series for the vanilla 2.4.24 kernel
3979 - add series for a cray x86/64 UL kernel drop
3980 - fix xattr patches for the vanilla 2.4.19 series
3982 - generate true UUIDs in lmc (1171)
3983 - have portals stack dumping break in UML (2466)
3984 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3985 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3986 - wake callback waiting threads on client eviction (2460)
3987 - Add --ptldebug and --subsystem to lmc (1719)
3988 - update assertion to allow safe interrupt allocation
3989 - set rq_no_resend for cancel requests (2432)
3990 - recalculate ptlrpcd timeout after resend (2494)
3991 - call vfs_rmdir when removing pending directories (2368)
3992 - fix renaming a file to itself (2429)
3993 - lmc creates a default one-stripe lov (2454)
3994 - expand procfs space to handle large clusters (2326)
3995 - increase UML stack to avoid overflow
3996 - update lconf's list of debug and subsystem masks
3997 - fix lfs find --obd (2510)
3998 - /proc tunable for disabling filter read caching (2591)
3999 - stop rpm packages from altering slapd.conf (2301)
4000 - disable nagle in the socknal under 0conf (2578)
4001 - choose mds inode size based on stripe count (2572)
4002 - fix kernel-source rpm problems (2516)
4004 - add --disable-doc to avoid pdf generation (2421)
4005 - update documentation, tests, type-os, comments
4006 - avoid format warnings on ia64
4007 - remove the TOE NAL
4008 - tiny code cleanups by removing unused fields
4010 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4013 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4014 - remove the most blatant lies from BUILDING file (2371)
4015 - change default debug level to reasonable production setting
4016 - reduce client side cache size to reduce cache flush time
4017 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4018 - make TCP zerocopy and pinger support enabled by default (2476)
4019 - sync writes completed after process exits caused crashes (2319)
4020 - maintain correct mount count on the MDS (2356)
4021 - backout 1557, because 2316 wasn't really fixed
4022 - better file I/O statistics gathering in /proc
4023 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4024 - another kernel patch to fix zero-copy TCP function export
4025 - don't take duplicate lock when processing re-sent getattr (2420)
4026 - lctl uses obd_self_export instead of creating new conn (2353)
4027 - MDS/OST recovery case which requires object creation asserted (2425)
4028 - move lfs from /usr/sbin to /usr/bin in packages
4029 - fix race between mds_client_add and mds_client_free (2417)
4030 - use kmalloc instead of slabs in portals (2430)
4031 - don't create duplicate records when a failover MDS is present (2442)
4032 - remove unnecessary mount age check (2332)
4033 - don't remove directory inodes from locks prematurely (2451)
4034 - don't break if MDS service name is the same as hostname (2103)
4035 - fix races in client write RPC generation when cache full (2482)
4037 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4040 - remove now-unused request->rq_obd (278)
4041 - if an allocation fails, print out how much memory we've used (1933)
4042 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4043 - add the "configurable stack size" patch to most series files (1256)
4044 - ability to write large log records, for 100+ OST configs (2306)
4045 - fix NULL deref when filter_prep fails (2314)
4046 - fix operator precedence error in filter_sync
4047 - dynamic allocation of socknal TX descriptors (2315)
4048 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4049 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4050 - ensure that CWARN messages in Portals always get to the syslog
4051 - __init/__exit are not for prototype decls (ldlm_init/exit)
4052 - x86-64 compile warning fixes
4053 - fix gateway LMC keyword conflict (2318)
4054 - fix MDS lock inversions in getattr/reint paths (1844)
4055 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4056 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4057 - DLM race condition prevented some lock evictions (2328)
4058 - ENOMEM detection and retry on socknal sends (2230)
4059 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4060 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4061 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4062 - better error messages when a client is rejected during recovery (1505)
4063 - avoid cancelling locks which were never granted, after failure (2330)
4064 - fix i_sem/journal inversion in mds_client_add (2333)
4065 - fix truncate/getattr lock cycle deadlock (2334)
4066 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4067 - fix two rare exit paths which could leak an l_lock() ref (2321)
4068 - fencepost error in MDS/OST orphan recovery (2226)
4069 - make log record alignment 8 bytes (1988)
4070 - lstripe now fails when requested offset > ost_count (2237)
4071 - ensure that all kernel series have a complete list.h (1607)
4072 - fix crashes in special-file operations (2316)
4073 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4075 - allow configurable automake binary, for testing new versions
4076 - small update to the lfs documentation
4078 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4080 * fix negative export reference count in fsfilt_sync (2312)
4082 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4083 * release candidate 0.9.1
4085 - orphans are moved into the PENDING directory for possible recovery
4086 - replayed opens now open by fid for orphan/rename safety (1042)
4087 - last close of an orphan inode generates a transno (683)
4088 - chdir() and mount() now pin the directory entry (1020)
4089 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4090 - discard very old requests without processing them (1502)
4091 - remove some common, well-understood CERRORs (1505)
4092 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4093 - clear "grant" flags in OST replies until OST grant code lands (1644)
4094 - fix read performance by not clobbering i_blksize on client (1598)
4095 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4096 - make LRU size a /proc tunable, clears locks when reduced (707)
4097 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4098 - server locks take references on exports now (1558)
4099 - build fixes for 2.4.20-rh trees (1663)
4100 - return an error from lov_create if all OSCs are inactive (1751)
4101 - fix import levels when a reconnect happens without a timeout (1597)
4102 - exit early from mds_open if we get a lookup error (1749)
4103 - partial page read at EOF wouldn't wait for disk before sending (1642)
4104 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4105 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4106 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4107 - write/truncate lock inversion (1639)
4108 - Don't auto-load obdclass, portals modules during cleanup (1495)
4109 - fix timestamps from jumping to "now" (1763)
4110 - extra journal assertions (1648)
4111 - add an extra multiunlink test (1771)
4112 - fix read_record/write_record API (1776)
4113 - fix leak of offset_extent, possible incorrect i_size later (1772)
4114 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4115 - Debugging for the kqswnal_get_idle_tx problems (1820)
4116 - Allow recovery to be attempted multiple times (1536)
4117 - Write out MDS last_rcvd file after it is first created (1600)
4118 - Fix tx_descriptor leak in failed transmit situations (1827)
4119 - ext3 journaling fixes for assertion failure after IO error (1871)
4120 - class_export_put() on freed export after completion AST error (1896)
4121 - Fix revalidate looping in VFS (1322)
4122 - Don't access a freed export during MDS_REINT timeout (1521)
4123 - Add open-unlink recovery support on the MDS (1673,1764)
4124 - Return an error if no MDS data was read from last_rcvd (1946)
4125 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4126 - Don't setup a disk device that doesn't match exported UUID (317)
4127 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4128 - avoid committing NULL handle in force close
4129 - local.sh is now a one-stripe LOV configuration
4130 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4131 - fix ext3 htree duplicate directory entry corruption (1516)
4132 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4133 - update to new LOV EA format (2097)
4134 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4136 2003-06-15 Phil Schwan <phil@clusterfs.com>
4139 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4140 - per-import/export recovery handling (958, 931, 959)
4141 - multiple last-rcvd slots, for serving multiple FSes (949)
4142 - connections are again shared between multiple imp/exports (963, 964)
4143 - "umount -f" would hang if any requests needed to be sent (393, 978)
4144 - avoid pinning large req buffer by copying for queued messages (989)
4145 - add "uuid" to "lctl device" command to help upcalls (991)
4146 - "open" RPCs with transnos would confuse recovery counters (1037)
4147 - do proper endian conversion of all wire messages (288, 340, 891)
4148 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4149 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4150 - fix ptlrpc_connection leak in target_handle_connect (1174)
4151 - fix import refcounting bug in OST and MDS cleanup (1134)
4152 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4153 - fix very unlikely obd_types race condition (501)
4154 - remove osc_open hack for echo_client (1187)
4155 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4156 - a failure in read_inode2 leads to deadlock (1139)
4157 - cancel ack-locks as soon as transaction is committed (1072)
4158 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4159 - make sure to commitrw after any preprw to avoid deadlock (1162)
4160 - failing to execute a file in a lustre FS would lock inode (1203)
4161 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4162 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4163 - obd_uuid2tgt would walk off the end of the list (1255)
4164 - on IA64 the osc would give portals incorrect bulk size (1258)
4165 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4166 - fix lock inversion caused by new llite matching code (1282)
4167 - limit the number of dirty pages on a client to 10MB (1286)
4168 - timed out locks were not being corrected cancelled (1289)
4169 - fix O_DIRECT above 4GB on IA-32 (1292)
4170 * major user-visible changes
4171 - fail out/fail over policy now controlled by the upcall (993)
4173 - add OBD_PING to check server availability and failure (954)
4174 - lustre messages are now sent in sending host order (288, 340, 891)
4175 - add eadatalen to MDS getattr reply (340)
4176 - OST read replies may contain second buffer, with per-page status (593)
4178 2003-03-11 Phil Schwan <phil@clusterfs.com>
4181 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4182 - failed open()s could cause deadlock; fixed (867, 869)
4183 - stop cancelling OST locks when files are closed (481)
4184 - overlapping XID spaces caused network corruption (851, 853)
4185 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4186 - setattr_raw added, to do single-RPC, server-side setattrs
4187 - lmc/lconf syntax change for OST UUIDs
4188 - fix crashy race condition between ptlrpc_free_req and osc_close
4189 - don't use request in mdc_enqueue if we hit a timeout (889)
4190 - don't set the inode i_size for regular files from the MDS (896)
4191 - handle out of order completion AST (842)
4192 - don't LBUG if a lock request times out after receiving AST (913)
4193 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4194 - if a bad lock AST arrives, send an error instead of dropping entirely
4195 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4196 - fix leak in bulk IO when only partially completed (899, 900, 926)
4197 - fix O_DIRECT for ia64 (55)
4198 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4199 - C-z after timeout could hang a process forever; fixed (977)
4201 - client-side I/O cache (678, 924, 929, 941, 970)
4203 - READPAGE and SETATTRs which don't take server-side locks get
4206 2003-02-11 Phil Schwan <phil@clusterfs.com>
4209 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4210 - Fix fsfilt_extN_readpage() to read a full page of directory
4211 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4212 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4213 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4214 - Most intent operations were removed, in favour of a new RPC mode
4215 that does a single RPC to the server and bypasses most of the VFS
4216 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4217 - Aggressively cancel local locks on DLM servers
4218 - mds_reint_unlink sends EA to the client if it's the last nlink.
4219 client uses that EA to unlink OST objects.
4220 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4221 - recursive symlinks were fixed (439)
4222 - fixed NULL deref in DEBUG_REQ
4223 - filter_update_lastobjid no longer calls sync, which annoyed extN
4224 - fixed multi-client small-writes to a single file problem (445)
4225 - fixed mtime updates during file writes (607)
4226 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4227 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4228 - fixed Linux OST size reporting problem (444, 656)
4229 - OST now updates object mtime with writes or setattr (607, 619)
4230 - client verifies file size before zeroing page past EOF (445)
4231 - OST now writes last allocated objid to disk with allocation (108)
4232 - LOV on echo now works (409)
4234 - mds_reint_unlink sends a new buffer, with the EA included. this
4235 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4236 set if a regular file was being unlinked, and it was the last link
4237 - use PtlGet from the target for bulk writes (315)
4238 - OST now updates object mtime with writes or setattr (607, 619)
4239 - LDLM now has a grant-time callback to revalidate locked items, if
4241 - Many MDS operations were reorganized to combat race conditions
4243 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4244 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4246 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4249 - Fully reactivate OST imports after reconnection (512, others)
4250 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4251 - More graceful error handling for truncating on dead OST (515)
4252 - Don't error out unless we're actually accessing dead stripes (474)
4253 - Fix garbage sizes when stripes are missing (410)
4254 - LRU counters were broken, causing constant lock purge (433, 432)
4255 - garbage on read from stripes with failed OSTs (441)
4256 - mark OSCs as active before reconnecting during recovery (438)
4257 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4258 - lfind did not preserve OST order in output (443)
4259 - symlinks cause hung clients, incorrect data (439)
4260 - stop dereferencing request after dropping refcount (457)
4261 - don't LASSERT(spin_is_locked) on non-SMP (455)
4262 - fixes for many rename() bugs
4263 - fstat didn't correctly synchronize attributes (399)
4264 - server must handle lock cancellation during blocking AST prep (487)
4265 - bulk descriptors were free()d too soon (511)
4266 - fix paths in lconf, which would load incorrect modules (451, 507)
4267 - fix confusing lconf 'host not found' error message (386)
4268 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4269 - fix race condition in mdc_blocking_ast() for inode access (526)
4270 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4271 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4272 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4273 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4274 - fix lstripe in several ways (kernel side) (527)
4275 - fix request leak in ldlm_cli_enqueue (262)
4276 - incorrect OSC was marked inactive after OST failure
4277 - call mds_fs_cleanup before unmounting filesystem (524)
4278 - fix races between taking ns_lock and ldlm_lock_change_resource
4279 - fix races updating LOV export open file list
4280 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4281 - fix recovery NULL deref in ldlm_cli_cancel_unused
4282 - fix some DLM races by using new hash table for lock handles (419)
4283 - permit the client to specify desired inodes, at replay
4284 - duplicate requests when we queue them for replay reintegration
4285 - fix last_rcvd offset calculation
4286 - sync after each recovered transaction, so we always make progress
4287 - never, not always, ERESTART requests without transnos
4288 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4290 - skip replay if the MDS says that the client is already connected
4291 - don't check for a recovery-enabled export to match lctl's UUID
4292 - don't INC_USE_COUNT for phantom exports
4293 - don't crash when cleaning up phantom exports (567)
4294 - don't double-finish or set replay data for errored mdc_open requests
4295 - abort requests when they time out, so we don't get old replies
4296 - send/receive replies for AST messages again
4297 - if the client says that it doesn't have the lock, cancel it on the
4299 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4300 mark it as destroyed, it will all work out when decref is called
4301 - fix module use counts (22, 581)
4303 - ASTs now expect a reply (server cancels lock on error reply)
4305 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4308 - fix many simultaneous client startup (392)
4309 - fix dentry->d_it clobbering
4310 - credentials weren't being shipped for readdir/getattr operations
4311 - remove invalid assertions triggered during some concurrent MD
4313 - proper Lustre versions added (336, 389)
4314 - fix memory leak for create error case (398)
4315 - fix LOV locking bug that would get cli/srv out of sync
4316 - fix echo client over LOV (409)
4317 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4318 - fix double-O_EXCL intent crash (424)
4319 - avoid sending multiple lock CANCELs (352)
4321 - MDS can do multi-client recovery (modulo bugs in new code)
4323 - many updates, edits, cleanups
4325 2002-11-18 Phil Schwan <phil@clusterfs.com>
4328 - fix null d_it dereference (346)
4329 - fix full OST/dbench hang (333)
4330 - fix permission problem with file removal (286)
4331 - fix removal of OSCs from LOV when they fail
4332 - fix NULL deref during bulk timeout (214)
4333 - fix problems related to multiple filesystems on one MDS (241)
4334 - fixed serious subtle metadata locking bugs
4335 - free locks on clients when inodes are removed due to memory
4337 - fix inode pointer in lock data (285)
4338 - partial support for multiple MDS on a single host (241)
4339 - data locks weren't cancelled at clear_inode time (290, 311)
4340 - intent locks could lead to unbounded lock growth (205)
4341 - added a maximum lock count, an LRU list, and a flusher
4342 - fix multiple rename (365)
4343 - properly abstracted the echo client
4344 - OSC locked 1 byte too many; fixed
4345 - rewrote brw callback code:
4346 - fixed recovery bugs related to LOVs (306)
4347 - fixed too-many-pages-in-one-write crash (191)
4348 - fixed (again) crash in sync_io_timeout (214)
4349 - probably fixed callback-related race (385)
4351 - Add capability to MDS protocol
4352 - LDLM cancellations and callbacks on different portals
4354 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4357 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4358 - defer open object destruction to close time (601981, 138)
4359 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4360 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4361 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4362 - don't call obd_setattr in ll_file_release for destroyed objects
4363 * protocol change to lustre_msg: move |version| and add |flags|
4364 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4365 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4366 * added replay of create, unlink, link and rename operations during
4367 MDS failover; recovery should be much more robust now
4368 * remove failed OSCs from LOVs (only lov_create uses this so far)
4369 * the lustre-HOWTO was brought (more) up to date (582544)
4371 2002-10-23 Phil Schwan <phil@clusterfs.com>
4374 - in-use dentries weren't being reused properly (617851)
4375 - prevent multiple LDLM setup (599178)
4376 - fix LOV size calculations for truncate (617853)
4377 - fix client handling of MDS intent errors (POSIX)
4378 - fix permission bug in lovstripe.c test (624321)
4379 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4380 - truncate past end of file could corrupt data
4381 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4382 - a race in recovery could return ETIMEDOUT to apps (623947)
4383 - building outside the source directory was fixed
4384 * the lustre-HOWTO was brought (more) up to date (582544)
4385 * major progress was made on recovery functionality
4387 2002-10-10 Phil Schwan <phil@clusterfs.com>
4390 - recovery deadlock fix
4391 - rm -rf causes LBUG fix (617817)
4392 - file open by multiple tasks fix (618962)
4393 - directory permissions bugs (602707 and 620007)
4394 - journal_stop fixed with locking (611313)
4395 - O_APPEND failures resolved (618273, perhaps 614459)
4396 - lconf PATH fix (619770)
4397 - IA64 build fix (621450)
4398 - RPC buffer sizes scale with amount of memory
4400 2002-10-01 Phil Schwan <phil@clusterfs.com>
4403 - locks would be cancelled without throwing away data pages,
4404 resulting in inconsistent data (605627)
4405 - inode attributes were not always being refreshed (605627, 612449)
4406 - lconf now continues to cleanup after lctl reports an error
4407 - MDS now enforces user permissions (602707)
4408 - lprocfs cleanup fixed, but not yet enabled (614157)
4409 - fixed infinite server hang, should a client not respond to an AST
4410 - avoid going into recovery if user calls readlink() with a buffer
4411 that's too small (613941)
4412 - AST RPCs no longer require replies (614867) -- this may be changed
4413 - don't crash server if client sends an IOV that's too big (611336)
4414 - fixed lock conversion deadlock (611892)
4415 - fixed the following of symlinks (614622)
4416 * recovery: the server can remove locks from a client that dies, other
4417 clients can make progress
4418 * more extN patch fixes
4419 * compile-time configurable ptlrpc buffer allocations
4421 - collaborative read cache document
4422 - Lustre Lite Performance CDR document-in-progress
4424 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4427 - fix typo in patch-2.4.18
4429 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4432 - clear ptlrpc request each time in handle_incoming_request()
4433 - unlink of files now destroys the object on the OST
4435 2002-09-19 Peter Braam <braam@clusterfs.com>
4437 * add hard link support
4438 * change obdfile creation method
4439 * kernel patch changed
4441 2002-09-19 Peter Braam <braam@clusterfs.com>
4444 - stack overflow bug in extN fixed
4446 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4448 * documentation updates
4449 - add man pages for config tools
4450 - update tests/README to describe testing with new config tools
4451 - finish metadata API descriptions
4452 * bug fixes and cleanups
4453 - statfs workaround for 16TB limit
4454 - LOV stripe allocation improved, can stripe on subset of OSTs
4455 - LOV file size/IO offset was wrong for files > 4GB in size
4456 - object EA data was being dropped, caused files to be unreadable
4457 - memory overflow with non-LOV OST caused memory corruption
4458 - fixed regression tests to work with new config tools, obdfilter
4459 - fixed bug when directory size became larger than 1 block
4460 - fixed bug (for single client case) when PWD was deleted
4461 - invalidate local directory pages when doing intent-based ops
4462 - avoid LDLM oops when lock callback contained bad data
4464 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4466 * documentation updates
4467 * bug fixes and cleanups
4468 - configuration tools
4471 - 64-bit compile warnings
4472 - 64-bit internal statfs data
4474 * test_brw on persistent OST devices
4476 * lprocfs (disabled)
4478 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4480 * documentation updates
4481 * bug fixes and cleanups
4482 * configuration tools
4484 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4487 * many small fixes to 0_5_4
4488 * io/network handling
4489 * thinkos in MDS operations
4491 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4494 * crucial basic fixes to 0.5.3
4495 * IOR, Iozone work over Elan
4498 2002-08-07 Phil Schwan <phil@clusterfs.com>
4499 * version 0_5_3, our first alpha
4500 * we use the new Portals iovs
4501 * documentation updates
4502 * bug fixes and cleanups
4503 * small changes in the DLM wire protocol
4505 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4506 * version 0_5_1 with some initial stability,
4507 * locking on MD and file I/O.
4508 * documentation updates
4509 * several bug fixes since 0.5.0
4510 * small changes in wire protocol
4512 2002-07-18 Phil Schwan <phil@clusterfs.com>
4514 * delivered as Lustre Light Alpha
4515 * fixed a crash after handling invalid MDS requests
4516 * fixed directory pages for architectures with non-4k pages sizes
4518 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4519 * release version v0_4_4
4520 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4522 * Fixes a number of interruption problems with OST operations.
4523 * Update documentation for portals header changes
4524 * Move all wire protocol structs/defines to lustre_idl.h
4525 * Fixes symlink length bug.
4526 * Add tcpdump to repository.
4528 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4529 * release version v0_4_3
4530 * Fixes statfs for inodes on extN.
4531 * Fixes bug in runtests which would delete /etc/hosts.
4532 * Use 64-bit object IDs wherever possible (not into VFS though)
4533 Remove ost_get_info, which is unused by lustre, and out of date.
4535 2002-07-03 Peter Braam <braam@clusterfs.com>
4536 * release version v0_4_2 Fixes a lookup error (type not passed)
4537 * move forward to head of Portals
4538 * move forward to latest Lustre kernel
4540 2002-06-25 Peter Braam <braam@clusterfs.com>
4541 * release version v0_4_1. Hopefully stable on single node use.