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 2.6.18-53.1.14.el5 (RHEL 5),
7 2.6.22.14 vanilla (kernel.org).
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.16 - 2.6.21 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.40.7-sun1
12 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
13 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
14 removed cwd "./" (refer to Bugzilla 14399).
18 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
20 Severity : enhancement
22 Description: Update to support 2.6.22.14 vanilla kernel.
26 Frequency : rare, on recovery
27 Description: read procfs can produce deadlock in some situation
28 Details : Holding lprocfs lock which send rpc can produce block for destroy
29 obd objects and this also block reconnect with -EALREADY. This isn't
30 fix all lprocfs bugs - but make it rare.
32 Severity : enhancement
34 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
37 Frequency : frequent on X2 node
39 Description: mdc_set_open_replay_data LBUG
40 Details : Set replay data for requests that are eligible for replay.
44 Description: lustre_mgs: operation 101 on unconnected MGS
45 Details : When MGC is disconnected from MGS long enough, MGS will evict the
46 MGC, and late on MGC cannot successfully connect to MGS and a lot
47 of the error messages complaining that MGS is not connected.
50 Frequency : on start mds
52 Description: Implement get_info(last_id) in obdfilter.
55 Frequency : occasional
57 Description: Correctly check stale fid, not start epoch if ost not support SOM
58 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
59 overwrited with new generated fid, but mds can anwer with one of these
60 two fids and both is not stale. setattr incorectly start epoch and
61 assume will be called done_writeting, but without SOM done_writing
65 Frequency : rare, depends on device drivers and load
67 Description: MDS or OSS nodes crash due to stack overflow
68 Details : Code changes in 1.8.0 increased the stack usage of some functions.
69 In some cases, in conjunction with device drivers that use a lot
70 of stack the MDS (or possibly OSS) service threads could overflow
71 the stack. One change which was identified to consume additional
72 stack has been reworked to avoid the extra stack usage.
75 Frequency : occasional
77 Description: Do not fail import if osc_interpret_create gets -EAGAIN
78 Details : If osc_interpret_create got -EAGAIN it immediately exits and
79 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
80 oscc_has_objects and see OSC has no objests and call
81 oscc_internal_create to resend create request.
83 Severity : enhancement
85 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
87 Severity : enhancement
89 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
94 Description: Fix replay if there is an un-replied request and open
95 Details : In some cases, older replay request will revert the
96 mcd->mcd_last_xid on MDS which is used to record the client's
99 Severity : enhancement
101 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
103 Severity : enhancement
105 Description: Add rhel5 support to HEAD.
107 Serverity : enhancement
109 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
114 Description: Don't allow skipping OSTs if index has been specified.
115 Details : Don't allow skipping OSTs if index has been specified, make locking
116 in internal create lots better.
120 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
121 Details : access to bitfield in structure is always rounded to long
122 and this produce problem with not atomic change any bit.
126 Description: Lustre make rpms failed.
127 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
128 builds Lustre rpms from tarball.
131 Frequency : rare on shutdown ost
133 Description: If llog cancel was not send before clean_exports phase, this can
134 produce deadlock in llog code.
135 Details : If llog thread has last reference to obd and call class_import_put
136 this produce deadlock because llog_cleanup_commit_master wait when
137 last llog_commit_thread exited, but this never success because was
138 called from llog_commit_thread.
142 Description: allow userland application know is lost one of stripes.
143 Details : fill lvb_blocks with error code on ost and return it to
144 application if error flag found.
148 Description: NULL lov_tgts causing MDS oops
149 Details : more safe checks for NULL lov_tgts for avoid oops.
151 Severity : enhancement
153 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
157 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
158 Details : If an OST is down the MDS will hang indefinitely in
159 obd_statfs_rqset() waiting for the statfs data. While for
160 MDS QOS usage of statfs, it should not stuck in waiting.
162 Severity : enhancement
164 Description: remote_acl support
165 Details : Support ACL-based permission check for remote user.
166 Support setfacl/getfacl for remote user with the utils
167 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
168 the system "{s,g}etfacl" utils.
170 Severity : enhancement
172 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
174 Severity : enhancement
176 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
180 Description: make mgs_setparam() handle fsname containing dash
181 Details : fsname containing a dash does not work with lctl conf_param
183 Severity : enhancement
185 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
187 Severity : enhancement
189 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
193 Frequency : rare, at shutdown
194 Description: access already free / zero obd_namespace.
195 Details : if client_disconnect_export was called without force flag set,
196 and exist connect request in flight, this can produce access to
197 NULL pointer (or already free pointer) when connect_interpret
198 store ocd flags in obd_namespace.
202 Frequency : only at startup
203 Description: not alloc memory with spinlock held.
204 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
205 if any spinlock held.
207 Severity : enhancement
209 Description: make lustre randomly fail allocating memory
210 Details : Make lustre randomly failed allocating memory for testing purpose.
212 Severity : enhancement
214 Description: lost problems with lov objid file
215 Details : Fixes some scability and access to not inited memory problems
216 in work with lov objdid file.
221 Description: lfs find does not continue on file error
222 Details : Continue other files processing when a file/dir is absent.
226 Description: Inconsistent usage of lustre_pack_reply()
227 Details : Standardize the usage of lustre_pack_reply() such that it
228 always generate a CERROR on failure.
231 Frequency : occasional
233 Description: MDS hang or stay in waiting lock
234 Details : If client receive lock with CBPENDING flag ldlm need send lock
235 cancel as separate rpc, to avoid situation when cancel request
236 can't processed due all i/o threads stay in wait lock.
240 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
244 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
248 Description: Update to SLES9 kernel-2.6.5-7.287.3.
252 Description: Update to RHEL4 latest kernel.
254 Severity : enhancement
256 Description: Build SLES10 patchless client fails
257 Details : The configure was broken by run ./configure with
258 --with-linux-obj=.... argument for patchless client. When the
259 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
260 can't search header adequately. Use absolute path such as
261 -I($LINUX)/include instead.
265 Description: interrupt oig_wait produce painc on resend.
266 Details : brw_redo_request can be used for resend requests from ptlrpcd and
267 private set, and this produce situation when rq_ptlrpcd_data not
268 copyed to new allocated request and triggered LBUG on assert
269 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
270 ptlrpcd set if request is changed and can be safety changed to use
275 Description: LASSERT_{REQ,REP}SWAB macros are buggy
276 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
277 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
278 these macros and replace them with their difinitions instead.
282 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
283 Details : Update which_patch & target file for SLES10 latest kernel.
286 Description: Remove client patches from SLES 10 kernel.
287 Details : This causes SLES 10 clients to behave as patchless clients
288 even on a Lustre-patched (server) kernel.
290 Severity : enhancement
292 Description: self-adjustable client's lru lists
293 Details : use adaptive algorithm for managing client cached locks lru
294 lists according to current server load, other client's work
295 pattern, memory activities, etc. Both, server and client
296 side namespaces provide number of proc tunables for controlling
299 Severity : enhancement
301 Description: light-weight GSS support
302 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
303 and reduce performance overhead.
305 Severity : enhancement
307 Description: Linux keyring support
308 Details : Support using service of Linux keyring for Lustre GSS internal
309 context refresh/cache mechanism.
313 Description: Fix errors in lfs documentation
314 Details : Fixes man pages
318 Description: don't use GFP_* in generic Lustre code.
319 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
323 Description: obdclass is limited by single OBD_ALLOC(idarray)
324 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
328 Description: warnings with build patchless client with vanila 2.6.19 and up
329 Details : change old ctl_table style and replace ctl_table/ctl_table_header
330 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
334 Frequency : Only for SLES
335 Description: sanity_quota fail test_1
336 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
337 file, which makes TSTID[2] inunique.
341 Frequency : Only for SLES9
342 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
343 Details : Update target/ChangeLog/which_patch .
347 Frequency : Only for test_10 in sanity-quota.sh
348 Description: a bug in quota test code
349 Details : checking a test flag in a wrong place
353 Frequency : Only for test_13 in sanity-quota.sh
354 Description: a bug in quota test script
355 Details : shouldn't check overfull allocated quota
359 Description: jbd statistics
360 Details : Port older jbd statistics patch for sles10
362 Severity : enhancement
364 Description: Add printing inode info into message about error in writepage.
369 Description: reply_lock_interpret crash due to race with it and lock cancel.
370 Details : Do not replay locks that are being cancelled. Do not reference
371 locks by their address during replay, just by their handle.
375 Frequency : When flocks are used.
376 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
377 Details : Flock locks might destroy just granted lock if it could be merged
378 with another existing flock, this is done in completion handler,
379 so teach ldlm_cli_enquque_fini that this is a valid case for
385 Description: Oops in read and write path when failing to allocate lock.
386 Details : Check if lock allocation failed and return error back.
390 Description: lstripe command fails for valid OST index
391 Details : The stripe offset is compared to lov->desc.ld_tgt_count
392 instead of lov->desc.ld_active_tgt_count.
396 Description: sanity.sh failed test 103
397 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
402 Description: df doesn't work properly if diskfs blocksize != 4K
403 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
407 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
408 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
412 Description: obd_health_check_timeout too short
413 Details : set obd_health_check_timeout as 1.5x of obd_timeout
417 Description: llapi_file_create() does not allow some changes
418 Details : add llapi_file_open() that allows specifying the mode and
419 open flags, and also returns an open file handle.
423 Description: merge and cleanup kernel patches.
424 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
428 Description: Add journal checksum support.(Kernel part)
429 Details : The journal checksum feature adds two new flags i.e
430 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
431 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
432 indicates that the commit block contains the checksum for
433 the blocks described by the descriptor blocks. Now commit
434 record can be sent to disk without waiting for descriptor
435 blocks to be written to disk. This behavior is controlled
436 using JBD2_FEATURE_ASYNC_COMMIT flag.
440 Description: OSS needs mutliple precreate threads
441 Details : Add ability to start more than one create thread per OSS.
445 Description: Sanity.sh test_65k failed: setstripe should have succeeded
446 Details : execute lfs setstripe on client
450 Description: mds_obd_create error creating tmp object
451 Details : When the user sets quota on root, llog will be affected and can't
452 create files and write files.
455 Frequency : Always on ia64 patchless client, and possibly others.
457 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
458 Details : This allows the patchless client to be loaded on architectures
463 Description: RedHat Update kernel for RHEL5
464 Details : Add the kernel config file for RHEL5.
468 Description: RedHat Update kernel for RHEL5
469 Details : Modify the kernel config file more closer RHEL5.
473 Description: Build failure against Centos5 (RHEL5)
474 Details : Define PAGE_SIZE when it isn't present.
478 Description: test_7 on sanity-quota.sh can't be used on separate machines
479 Details : add facet to handle it
483 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
484 Details : with NFS, the anon dentry's parent was set to itself in
485 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
486 corresponding dentry other than use rec->ur_name.
488 Severity : enhancement
490 Description: lfs setstripe enhancement
491 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
495 Description: enable data checksumming by default
496 Details : enable checksum by default, allow --disable-checksum
497 configure option and "-o nochecksum" mount option. Checksums
498 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
502 Description: lustre support for RHEL5
503 Details : Add support for RHEL5.
507 Description: Client eviction due to failover config
508 Details : after a connection loss, the lustre client should attempt to
509 reconnect to the last active server first before trying the
510 other potential connections.
514 Description: when mds and osts use different quota unit(32bit and 64bit),
515 quota will be released repeatly.
516 Details : void sending multiple quota reqs to mds, which will keep the
517 status between the reqs.
521 Description: osts not allocated evenly to files
522 Details : change the condition to increase offset_idx
525 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
527 Description: Data corruption for OSTs that are formatted larger than 2TB
529 Details : When generating the bio request for lustre file writes the
530 sector number would overflow a temporary variable before being
531 used for the IO. The data reads correctly from Lustre (which
532 will overflow in a similar manner) but other file data or
533 filesystem metadata may be corrupted in some cases.
537 Description: Tune the kernel for good SCSI performance.
538 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
539 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
544 Description: rewrite ext2-derived code in obdclass/uuid.c
545 Details : rewrite inherited code (uuid parsing code from ext2 utils)
546 from scratch preserving functionality.
550 Description: "lfs find -obd UUID" prints directories
551 Details : "lfs find -obd UUID" will return all directory names instead
552 of just file names. It is incorrect because the directories
553 do not reside on the OSTs.
557 Description: use i_size_read and i_size_write in 2.6 port
558 Details : replace inode->i_size access with i_size_read/write()
561 Frequency : when removing large files
563 Description: scheduling issue during removal of large Lustre files
564 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
565 It causes scheduling issues when removing large files (17TB in the
569 Frequency : only with liblustre clients on XT3
571 Description: evictions taking too long
572 Details : allow llrd to evict clients directly on OSTs
575 Frequency : only on ppc
577 Description: /proc/fs/lustre/devices broken on ppc
578 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
579 We should make sure the type of 'index' is loff_t and then cast
580 explicitly as needed below. Do not assign an explicitly cast
584 Frequency : only for rhel5
586 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
587 Details : Modify the target file & which_kernel.
589 Severity : enhancement
591 Description: omit set fsid for export NFS
592 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
593 as FS_REQUIRES_DEV for avoid problems with generate fsid.
597 Frequency : Always, for kernels after 2.6.16
598 Description: Fix warning idr_remove called for id=.. which is not allocated.
599 Details : Last kernels save old s_dev before kill super and not allow
600 to restore from callback - restore it before call kill_anon_super.
604 Description: buffer overruns could theoretically occur
605 Details : llapi_semantic_traverse() modifies the "path" argument by
606 appending values to the end of the origin string, and a
607 overrun may occur. Adding buffer overrun check in liblustreapi.
611 Description: Fix error on 'ls .' at the top of the Lustre mount.
612 Details : Don't revalidate dentry if it is a root dentry.
616 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
617 Details : Modify vm-tunables-rhel4.patch.
621 Description: Kernel config for 2.6.18-vanilla.
622 Details : Modify targets/2.6-vanilla.target.in.
623 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
624 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
625 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
626 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
630 Description: improve handling recoverable errors
631 Details : if request processig with error which can be recoverable on server
632 request should be resend, otherwise page released from cache and
637 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
638 Details : Modify target file & which_patch.
639 A flaw was found in the IA32 system call emulation provided
640 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
641 value could be stored in the %RAX register, which could trigger an
642 out-of-bounds system call table access. An untrusted local user
643 could exploit this flaw to run code in the kernel
644 (ie a root privilege escalation). (CVE-2007-4573).
648 Description: O_DIRECT bypasses client statistics.
649 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
650 read_ahead_stats were updated. copied the stats section from
651 osc_send_oap_rpc() into async_internal().
655 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
659 Description: change order of libsysio includes
660 Details : '#include sysio.h' should always come before '#include xtio.h'
664 Description: handle "serious error: objid * already exists" more gracefully
665 Details : If LAST_ID value on disk is smaller than the objects existing in
666 the O/0/d* directories, it indicates disk corruption and causes an
667 LBUG(). If the object is 0-length, then we should use the existing
668 object. This will help to avoid a full fsck in most cases.
670 Severity : enhancement
672 Description: adapt the lustre_config script to support the upgrade case
673 Details : Add "-u" option for lustre_config script to support upgrading 1.4
674 server targets to 1.6 in parallel.
678 Description: To avoid grant space > avaible space when the disk is almost
679 full. Without this patch you might see the error "grant XXXX >
680 available" or some LBUG about grant, when the disk is almost
682 Details : In filter_check_grant, for non_grant cache write, we should
683 check the left space by if (*left > ungranted + bytes), instead
684 of (*left > ungranted), because only we are sure the left space
685 is enough for another "bytes", then the ungrant space should be
686 increase. In client, we should update cl_avail_grant only there
687 is OBD_MD_FLGRANT in the reply.
691 Description: Update RHEL 4 kernel to fix local root privilege escalation.
692 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
693 described in CVE-2007-4573. This problem could allow untrusted
694 local users to gain root access.
697 Frequency : when using O_DIRECT and quotas
699 Description: Incorrect file ownership on O_DIRECT output files
700 Details : block usage reported by 'lfs quota' does not take into account
701 files that have been written with O_DIRECT.
706 Description: touch file failed when fs is not full
707 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
708 otherwise we can get ENOSP while fs is not full.
712 Description: parallel lock callbacks
713 Details : Instead of sending blocking and completion callbacks as separated
714 requests, adding them to a set and sending in parallel.
717 Frequency : only for Cray XT3
718 Bugzilla : 12829/13455
719 Description: Changing primary group doesn't change the group lustre assigns to
721 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
722 overridden with the primary group provided by the group upcall,
723 whereas the supplied fsgid can be trusted if it is in the list of
724 supplementary groups returned by the group upcall.
726 Severity : enhancement
728 Description: Allow masking D_WARNING, D_ERROR messages from console
729 Details : Console messages can now be disabled via lnet.printk.
733 Description: User code with malformed file open parameter crashes client node
734 Details : Before packing join_file req, all the related reference should be
735 checked carefully in case some malformed flags cause fake
736 join_file req on client.
740 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
741 Details : ldlm_completion_ast() assumes that a lock is granted when the req
742 mode is equal to the granted mode. However, it should also check
743 that LDLM_FL_CP_REQD is not set.
747 Description: Heavy nfs access might result in deadlocks
748 Details : After ELC code landed, it is now improper to enqueue any mds
749 locks under och_sem, because enqueue might want to decide to
750 cancel open locks for same inode we are holding och_sem for.
754 Description: Client eviction while running blogbench
755 Details : A lot of unlink operations with concurrent I/O can lead to a
756 deadlock causing evictions. To address the problem, the number of
757 oustanding OST_DESTROY requests is now throttled to
758 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
759 callbacks are processed in priority.
763 Description: enable ACLs on MDS by default
764 Details : ACLs must be enabled on MDS by default.
767 Frequency : PPC/PPC64 only
769 Description: conflicts between asm-ppc64/types.h and lustre_types.h
770 Details : fix duplicated definitions between asm-ppc64/types.h and
771 lustre_types.h on PPC.
774 Frequency : PPC/PPC64 only
776 Description: asm-ppc/segment.h does not exist
777 Details : fix compile issue on PPC.
781 Description: better handle error messages in extents code
784 Frequency : RHEL4 only
786 Description: mkfs is very slow on IA64/RHEL4
787 Details : A performance regression has been discovered in the MPT Fusion
788 driver between versions 3.02.73rh and 3.02.99.00rh. As a
789 consequence, we have downgraded the MPT Fusion driver in the RHEL4
790 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
792 Severity : enhancement
794 Description: SNMP support enhancement
795 Details : Adding total number of sampled request for an MDS node in snmp
798 Severity : enhancement
800 Description: Optimize ldlm waiting list processing for PR extent locks
801 Details : When processing waiting list for read extent lock and meeting read
802 lock that is same or wider to it that is not contended, skip
803 processing rest of the list and immediatelly return current
804 status of conflictness, since we are guaranteed there are no
805 conflicting locks in the rest of the list.
809 Description: Time out and refuse to reconnect
810 Details : When the failover node is the primary node, it is possible
811 to have two identical connections in imp_conn_list. We must
812 compare not conn's pointers but NIDs, otherwise we can defeat
813 connection throttling.
817 Description: port llog fixes from b1_6 into HEAD
818 Details : Port llog reference couting and some llog cleanups from b1_6
819 (bug 10800) into HEAD, for protect from panic and access to already
820 free llog structures.
824 Description: Detect stride IO mode in read-ahead
825 Details : When a client does stride read, read-ahead should detect that and
826 read-ahead pages according to the detected stride pattern.
830 Description: data checksumming impacts single node performance
831 Details : add support for several checksum algorithm. Currently, only CRC32
832 and Adler-32 are supported. The checksum type can be changed on
833 the fly via /proc/fs/lustre/osc/*/checksum_type.
837 Description: use adler32 for page checksums
838 Details : when available, use the Adler-32 algorithm instead of CRC32 for
843 Description: build for x2 fails
844 Details : fix compile issue on Cray systems.
848 Description: Properly match for duplicate locks
849 Details : Due to different lock order from skiplists code, we need to
850 traverse entire list for now
853 Frequency : only on PPC/SLES10
855 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
856 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
857 be defined. Add a hack in mkfs_lustre.c to work around this compile
862 Description: LASSERT on MDS when client holding flock lock dies
863 Details : ldlm pool logic depends on number of granted locks equal to
864 number of released locks which is not true for flock locks, so
865 just exclude such locks from consideration.
869 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
870 Details : Use fsfilt_sync() for both the whole filesystem sync and
871 individual file sync to eliminate dangerous inode locking
872 with I_LOCK that can lead to a deadlock.
874 --------------------------------------------------------------------------------
876 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
878 * Support for kernels:
879 2.6.5-7.283 (SLES 9),
880 2.6.9-55.EL (RHEL 4),
881 2.6.16.46-0.14 (SLES 10),
882 2.6.18.8 vanilla (kernel.org)
883 * Client support for unpatched kernels:
884 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
885 2.6.16 - 2.6.22 vanilla (kernel.org)
886 * Due to recently discovered recovery problems, we do not recommend
887 using patchless RHEL 4 clients with this or any earlier release.
888 * Recommended e2fsprogs version: 1.39.cfs8
889 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
890 * Starting with this release, the ldiskfs backing filesystem required
891 by Lustre is now in its own package, lustre-ldiskfs. This package
892 should be installed. It is versioned separately from Lustre and
893 may be released separately in future.
897 Description: block reactivating mgc import until all deactivates complete
898 Details : Fix race when failing back MDT/MGS to itself (testing)
901 Frequency : at statup only
903 Description: mds_lov_synchronize race leads to various problems
904 Details : simultaneous MDT->OST connections at startup can cause the
905 sync to abort, leaving the OSC in a bad state.
907 Severity : enhancement
909 Description: add optional extra BUILD_VERSION info
910 Details : add a new environment variable (namely LUSTRE_VERS) which allows
911 to override the lustre version.
914 Frequency : 2.6.18 servers only
916 Description: ll_kern_mount() doesn't release the module reference
917 Details : The ldiskfs module reference count never drops down to 0
918 because ll_kern_mount() doesn't release the module reference.
923 Description: server LBUG when using old ost_num_threads parameter
924 Details : Accept the old ost_num_threads parameter but warn that it
925 is deprecated, and fix an off-by-one error that caused an LBUG.
930 Description: Transient SCSI error results in persistent IO issue
931 Details : iobuf->dr_error is not reinitialized to 0 between two
935 Frequency : sometimes when underlying device returns I/O errors
937 Description: OSTs not going read-only during write failures
938 Details : OSTs are not remounted read-only when the journal commit threads
939 get I/O errors because fsfilt_ext3 calls journal_start/stop()
940 instead of the ext3 wrappers.
944 Description: poor connect scaling with increasing client count
945 Details : Don't run filter_grant_sanity_check for more than 100 exports
946 to improve scaling for large numbers of clients.
949 Frequency : SLES10 only
951 Description: sanity-quota.sh quotacheck failed: rc = -22
952 Details : Quotas cannot be enabled on SLES10.
955 Frequency : liblustre clients only
957 Description: getdirentries does not give error when run on compute nodes
958 Details : getdirentries does not fail when the size specified as an argument
959 is too small to contain at least one entry
961 Severity : enhancement
963 Description: Add LNET router traceability for debug purposes
964 Details : If a checksum failure occurs with a router as part of the
965 IO path, the NID of the last router that forwarded the bulk data
966 is printed so it can be identified.
971 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
972 Details : Due to a race condition, liblustre clients were occasionally
975 Severity : enhancement
977 Description: lfs setstripe use optional parameters instead of postional
980 Severity : enhancement
982 Description: Nanosecond timestamp support for ldiskfs
983 Details : The on-disk ldiskfs filesystem has added support for nanosecond
984 resolution timestamps. There is not yet support for this at
985 the Lustre filesystem level.
988 Frequency : during server recovery
990 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
991 Details : request with rq_no_resend flag not awake l_wait_event if they get a
995 Frequency : nfs export on patchless client
997 Description: connectathon hang when test nfs export over patchless client
998 Details : Disconnected dentry cannot be found with lookup, so we do not need
999 to unhash it or make it invalid
1002 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1003 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1004 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1009 Description: open req refcounting wrong on reconnect
1010 Details : If reconnect happened between getting open reply from server and
1011 call to mdc_set_replay_data in ll_file_open, we will schedule
1012 replay for unreferenced request that we are about to free.
1013 Subsequent close will crash in variety of ways.
1014 Check that request is still eligible for replay in
1015 mdc_set_replay_data().
1020 Description: disable writes to filesystem when reading health_check file
1021 Details : the default for reading the health_check proc file has changed
1022 to NOT do a journal transaction and write to disk, because this
1023 can cause reads of the /proc file to hang and block HA state
1024 checking on a healthy but otherwise heavily loaded system. It
1025 is possible to return to the previous behaviour during configure
1026 with --enable-health-write.
1028 Severity : enhancement
1030 Description: 64-bit inode version
1031 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1032 to the inode. This will be required for version-based recovery.
1037 Description: MDS fails to start if a duplicate client export is detected
1038 Details : in some rare cases it was possible for a client to connect to
1039 an MDS multiple times. Upon recovery the MDS would detect this
1040 and fail during startup. Handle this more gracefully.
1042 Severity : enhancement
1044 Description: Add -o localflock option to simulate old noflock
1046 Details : This will achieve local-only flock/fcntl locks
1052 Description: log_commit_thread vs filter_destroy race leads to crash
1053 Details : Take import reference before releasing llog record semaphore
1058 Description: Wrong request locking in request set processing
1059 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1060 imp_delayed_list, in this place should be used imp_lock.
1063 Frequency : when reconnection
1065 Description: Grant Leak when osc reconnect to OST
1066 Details : When osc reconnect ost, OST(filter) should check whether it
1067 should grant more space to client by comparing fed_grant and
1068 cl_avail_grant, and return the granted space to client instead
1069 of "new granted" space, because client will call osc_init_grant
1070 to update the client grant space info.
1073 Frequency : when client reconnect to OST
1075 Description: Grant Leak when osc do resend and replay bulk write
1076 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1077 bulk write request, because the grant info while be sync between
1078 OSC and OST when reconnect, and we should ignore the grant info
1079 these of resend/replay write req.
1084 Description: Grant space more than avaiable left space sometimes.
1085 Details : When then OST is about to be full, if two bulk writing from
1086 different clients came to OST. Accord the avaliable space of the
1087 OST, the first req should be permitted, and the second one
1088 should be denied by ENOSPC. But if the seconde arrived before
1089 the first one is commited. The OST might wrongly permit second
1090 writing, which will cause grant space > avaiable space.
1093 Frequency : when client is evicted
1095 Description: Grant might be wrongly erased when osc is evicted by OST
1096 Details : when the import is evicted by server, it will fork another
1097 thread ptlrpc_invalidate_import_thread to invalidate the
1098 import, where the grant will be set to 0. While the original
1099 thread will update the grant it got when connecting. So if
1100 the former happened latter, the grant will be wrongly errased
1101 because of this race.
1106 Description: Checking Stale with correct fid
1107 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1108 to check whether it is stale, because sometimes, we want the
1109 enqueue happened anyway, and op_data.fid2 will not be initialized.
1111 Severity : enhancement
1113 Description: update patchless client
1114 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1117 Frequency : only with 2.4 kernel
1119 Description: random memory corruption
1120 Details : size of struct ll_inode_info is to big for union inode.u and this
1121 can be cause of random memory corruption.
1126 Description: Memory leak in recovery
1127 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1128 It should also check obd_fail before fsfilt_start, otherwise if
1129 fsfilt_start return -EROFS,(failover mds during mds recovery).
1130 then the req will return with repmsg->transno = 0 and rc = EROFS.
1131 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1132 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1133 be freed no matter whether fsfilt_commit success or not.
1136 Frequency : only with huge count clients
1138 Description: Prevents from taking the superblock lock in llap_from_page for
1140 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1141 for llap_from_page call in ll_removepage prevents from taking the
1142 superblock lock for a soon died page.
1147 Description: Not check open intent error before release open handle
1148 Details : in some rare cases, the open intent error is not checked before
1149 release open handle, which may cause
1150 ASSERTION(open_req->rq_transno != 0), because it tries to release
1151 the failed open handle.
1156 Description: Set cat log bitmap only after create log success.
1157 Details : in some rare cases, the cat log bitmap is set too early. and it
1158 should be set only after create log success.
1162 Description: Accessing a block bevice can re-enable I/O when Lustre is
1163 tearing down a device.
1164 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1169 Description: service threads may hog cpus when there are a lot of requests
1170 Details : Insert cond_resched to give other threads a chance to use some CPU
1175 Description: the cat log was not initialized in recovery
1176 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1177 unlink log on mds will not be initialized until mds post
1178 recovery. And also in mds post recovery, the unlink log will
1179 initialization will be done asynchronausly, so there will be race
1180 between add unlink log and unlink log initialization.
1184 Description: brw_stats were being printed incorrectly
1185 Details : brw_stats were being printed as log2 but all of them were not
1186 recorded as log2. Also remove some code duplication arising from
1187 filter_tally_{read,write}.
1191 Frequency : rare, only in recovery.
1192 Description: ASSERTION(req->rq_type != LI_POISON) failed
1193 Details : imp_lock should be held while iterating over imp_sending_list for
1194 prevent destroy request after get timeout in ptlrpc_queue_wait.
1198 Description: replay-single.sh test 52 fails
1199 Details : A lock's skiplist need to be cleanup when it being unlinked
1200 from its resource list.
1204 Description: Short directio read returns full requested size rather than
1206 Details : Direct I/O operations should return actual amount of bytes
1207 transferred rather than requested size.
1209 Severity : enhancement
1211 Description: metadata RPC reduction (e.g. for rm performance)
1212 Details : decrease the amount of synchronous RPC between clients and servers
1213 by canceling conflicing lock before the operation on the client
1214 and packing thier handles into the main operation RPC to server.
1216 Severity : enhancement
1218 Description: Async OSC create to avoid the blocking unnecessarily.
1219 Details : If a OST has no remain object, system will block on the creating
1220 when need to create a new object on this OST. Now, ways use
1221 pre-created objects when available, instead of blocking on an
1222 empty osc while others are not empty. If we must block, we block
1223 for the shortest possible period of time.
1225 Severity : enhancement
1227 Description: refine locking for avoid write wrong info into lov_objid file
1228 Details : fix possible races with add new target and write/update data in
1231 --------------------------------------------------------------------------------
1233 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1238 Frequency : on some architectures
1240 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1241 Details : Uninitialized flags sometimes cause configuration commands to
1245 Frequency : patchless clients only
1247 Description: missing __iget() symbol export
1248 Details : The __iget() symbol export is missing. To avoid the need for
1249 this on patchless clients the deathrow inode reaper is turned
1250 off, and we depend on the VM to clean up old inodes. This
1251 dependency was during via the fix for bug 12181.
1253 --------------------------------------------------------------------------------
1255 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1257 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1258 INTEROPERATE with older versions automatically. In many cases a
1259 special upgrade step is needed. Please read the
1260 user documentation before upgrading any part of a live system.
1261 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1262 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1263 * WARNING: Lustre configuration and startup changes are required with
1264 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1270 Severity : enhancement
1272 Description: MountConf
1273 Details : Lustre configuration is now managed via mkfs and mount
1274 commands instead of lmc and lconf. New obd types (MGS, MGC)
1275 are added for dynamic configuration management. See
1276 https://mail.clusterfs.com/wikis/lustre/MountConf for
1279 Severity : enhancement
1281 Description: dynamic OST addition
1282 Details : OSTs can now be added to a live filesystem
1284 Severity : enhancement
1286 Description: startup order invariance
1287 Details : MDTs and OSTs can be started in any order. Clients only
1288 require the MDT to complete startup.
1290 Severity : enhancement
1292 Description: parallel, asynchronous orphan cleanup
1293 Details : orphan cleanup is now performed in separate threads for each
1294 OST, allowing parallel non-blocking operation.
1296 Severity : enhancement
1298 Description: optimized stripe assignment
1299 Details : stripe assignments are now made based on ost space available,
1300 ost previous usage, and OSS previous usage, in order to try
1301 to optimize storage space and networking resources.
1303 Severity : enhancement
1305 Description: Permanently set tunables
1306 Details : All writable /proc/fs/lustre tunables can now be permanently
1307 set on a per-server basis, at mkfs time or on a live system.
1309 Severity : enhancement
1311 Description: Lustre message v2
1312 Details : Add lustre message format v2.
1314 Severity : enhancement
1316 Description: client OST exclusion list
1317 Details : Clients can be started with a list of OSTs that should be
1318 declared "inactive" for known non-responsive OSTs.
1322 Description: SPEC SFS validation failure on NFS v2 over lustre.
1323 Details : Changes the blocksize for regular files to be 2x RPC size,
1324 and not depend on stripe size.
1326 Severity : enhancement
1328 Description: Multiple MD RPCs in flight.
1329 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1330 To avoid overly-overloading MDS, introduce a limit on number of
1331 MDS RPCs in flight for a single client and add /proc controls
1332 to adjust this limit.
1334 Severity : enhancement
1336 Description: client read/write statistics
1337 Details : Add client read/write call usage stats for performance
1338 analysis of user processes.
1339 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1340 file access. extents_stats shows chunk size distribution.
1341 extents_stats_per_process show chunk size distribution per
1344 Severity : enhancement
1346 Description: mds statistics
1347 Details : Add detailed mds operations statistics in
1348 /proc/fs/lustre/mds/*/stats.
1352 Description: Failure of copying files with lustre special EAs.
1353 Details : Client side always return success for setxattr call for lustre
1354 special xattr (currently only "trusted.lov").
1359 Description: Refcount LNET uuids
1360 Details : The global LNET uuid list grew linearly with every startup;
1361 refcount repeated list entries instead of always adding to
1364 Severity : enhancement
1366 Description: Dynamic service threads
1367 Details : Within a small range, start extra service threads
1368 automatically when the request queue builds up.
1371 Frequency : mixed-endian client/server environments
1373 Description: mixed-endian crashes
1374 Details : The new msg_v2 system had some failures in mixed-endian
1377 Severity : enhancement
1379 Description: Easy OST removal
1380 Details : OSTs can be permanently deactivated with e.g. 'lctl
1381 conf_param lustre-OST0001.osc.active=0'
1383 Severity : enhancement
1385 Description: MGS proc entries
1386 Details : Added basic proc entries for the MGS showing what filesystems
1389 Severity : enhancement
1391 Description: provide MGS failover
1392 Details : Added config lock reacquisition after MGS server failover.
1394 Severity : enhancement
1396 Description: add Linux 2.4 support
1397 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1401 Description: a large application tries to do I/O to the same resource and dies
1402 in the middle of it.
1403 Details : Check the req->rq_arrival time after the call to
1404 ost_brw_lock_get(), but before we do anything about
1405 processing it & sending the BULK transfer request. This
1406 should help move old stale pending locks off the queue as
1407 quickly as obd_timeout.
1410 Frequency : when an incorrect nid is specified during startup
1412 Description: ptlrpc connect to non-existant node causes kernel crash
1413 Details : LNET can't be re-entered from an event callback, which
1414 happened when we expire a message after the export has been
1415 cleaned up. Instead, hand the zombie cleanup off to another
1418 Severity : enhancement
1420 Description: plain/inodebits lock performance improvement
1421 Details : Grouping plain/inodebits in granted list by their request modes
1422 and bits policy, thus improving the performance of search through
1426 Frequency : only if OST filesystem is corrupted
1428 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1429 Details : for a short time RPCs with bulk IO are in the replay list,
1430 but replay of bulk IOs is unimplemented. If the OST filesystem
1431 is corrupted due to disk cache incoherency and then replay is
1432 started it is possible to trip an assertion. Avoid putting
1433 committed RPCs into the replay list at all to avoid this issue.
1436 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1438 Description: System hang on startup
1439 Details : This bug allowed the liblustre (e.g. catamount) client to
1440 return to the app before handling all startup RPCs. This
1441 could leave the node unresponsive to lustre network traffic
1442 and manifested as a server ptllnd timeout.
1444 Severity : enhancement
1446 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1447 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1448 boolean to control whether to print peer debug info when a
1449 client's RPC times out.
1452 Frequency : only for kernels with patches from Lustre below 1.4.3
1454 Description: Remove old rdonly API
1455 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1458 Frequency : only for devices with external journals
1460 Description: Set external device read-only also
1461 Details : During a commanded failover stop, we set the disk device
1462 read-only while the server shuts down. We now also set any
1463 external journal device read-only at the same time.
1466 Frequency : when upgrading from 1.4 while trying to change parameters
1468 Description: The wrong (new) MDC name was used when setting parameters for
1469 upgraded MDT's. Also allows changing of OSC (and MDC)
1470 parameters if --writeconf is specified at tunefs upgrade time.
1473 Frequency : when setting specific ost indicies
1475 Description: QOS code breaks on skipped indicies
1476 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1477 created with --index need not be sequential.
1479 Severity : enhancement
1481 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1482 Details : The uninit_groups feature works in conjunction with the kernel
1483 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1484 up the pass1 processing of e2fsck. This is a read-only feature
1485 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1486 mount filesystems that have had this feature enabled.
1488 Severity : enhancement
1490 Description: Improve multi-block allocation algorithm to avoid fragmentation
1491 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1492 allocation locality and avoid filesystem fragmentation.
1494 ------------------------------------------------------------------------------
1496 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1498 * Support for kernels:
1499 2.4.21-47.0.1.EL (RHEL 3)
1500 2.6.5-7.283 (SLES 9)
1501 2.6.9-42.0.10.EL (RHEL 4)
1502 2.6.12.6 vanilla (kernel.org)
1503 2.6.16.27-0.9 (SLES 10)
1504 * Recommended e2fsprogs version: 1.39.cfs5
1506 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1510 Frequency : occasional, depends on client load and configuration
1511 Bugzilla : 12181, 12203
1512 Description: data loss for recently-modified files
1514 Details : In some cases it is possible that recently written or created
1515 files may not be written to disk in a timely manner (this should
1516 normally be within 30s unless client IO load is very high).
1517 The problem appears as zero-length files or files that are a
1518 multiple of 1MB in size after a client crash or client eviction
1519 that are missing data at the end of the file.
1521 This problem is more likely to be hit on clients where files are
1522 repeatedly created and unlinked in the same directory, clients
1523 have a large amount of RAM, have many CPUs, the filesystem has
1524 many OSTs, the clients are rebooted frequently, and/or the files
1525 are not accessed by other nodes after being written.
1527 The presence of the problem can be detected by looking at
1528 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1529 smaller than the second (nr_unused) then dirty files will not
1530 be flushed automatically to disk. "sync; sleep 10" should be
1531 run several times on the node before unmounting it to update
1532 Lustre (this is also safe to run on nodes without this problem).
1534 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1535 that can cause this same problem, so customers using that kernel
1536 also need to update the kernel in addition to Lustre. In order
1537 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1539 It is normal that files written just before a client crash (less
1540 than 30s) may not yet have been flushed to disk, even for local
1544 Frequency : frequent on thin XT3 nodes
1546 Description: UUID collision on thin XT3 Linux nodes
1547 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1548 randomly, since we relied on an insufficiently-seeded PRNG.
1553 Description: OSS hangs after "All ost request buffers busy"
1554 Details : A deadlock between quota and journal operations caused OSS
1555 hangs after printing "All ost request buffers busy."
1558 Frequency : always on liblustre builds
1560 Description: Cleanup compiler warnings on liblustre
1563 Frequency : always on liblustre builds on XT3
1565 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1569 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1570 > 32000 subdirectories
1571 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1572 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1573 subdirectory count crosses 32000. This will aid e2fsck to
1574 correctly handle more than 32000 subdirectories.
1577 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1579 Description: System hang on startup
1580 Details : This bug allowed the liblustre (e.g. catamount) client to
1581 return to the app before handling all startup RPCs. This
1582 could leave the node unresponsive to lustre network traffic
1583 and manifested as a server ptllnd timeout.
1585 Severity : enhancement
1587 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1588 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1589 boolean to control whether to print peer debug info when a
1590 client's RPC times out.
1595 Description: make O_SYNC working on 2.6 kernels
1596 Details : 2.6 kernels use different method for mark pages for write,
1597 so need add a code to lustre for O_SYNC work.
1602 Description: Failure to close file and release space on NFS
1603 Details : Put inode details into lock acquired in ll_intent_file_open.
1604 Use mdc_intent_lock in ll_intent_open to properly
1605 detect all kind of errors unhandled by mdc_enqueue.
1610 Description: proc file read during shutdown sometimes raced obd removal,
1612 Details : Add lock to prevent obd access after proc file removal.
1615 Frequency : Only for files larger than 4GB on 32-bit clients.
1617 Description: improperly doing page alignment of locks
1618 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1619 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1624 Description: RPCs being resent when they shouldn't be
1625 Details : Some RPCs that should not be resent are being resent. This
1626 can cause inconsistencies in the RPC state machine. Do not
1627 resend such requests.
1630 Frequency : rare, only with NFS export
1632 Description: Crash on NFS re-export node
1633 Details : under very unusual load conditions an assertion is hit in
1634 ll_intent_file_open()
1637 Frequency : only if OST filesystem is corrupted
1639 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1640 Details : for a short time RPCs with bulk IO are in the replay list,
1641 but replay of bulk IOs is unimplemented. If the OST filesystem
1642 is corrupted due to disk cache incoherency and then replay is
1643 started it is possible to trip an assertion. Avoid putting
1644 committed RPCs into the replay list at all to avoid this issue.
1649 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1650 Details : Large single O_DIRECT read and write calls can fail to allocate
1651 a sufficiently large buffer to process the request. In case of
1652 allocation failure the allocation is retried with a smaller
1653 buffer and broken into smaller requests.
1655 Severity : enhancement
1657 Description: Add -o localflock option to simulate old noflock behaviour.
1658 Details : This will achieve local-only flock/fcntl locks coherentness.
1663 Description: versioning check is incomplete
1664 Details : Checking the version difference of client vs. server, report
1665 error if the gap is too big.
1670 Description: add support PG_writeback bit
1671 Details : add support for PG_writeback bit for Lustre, for more carefull
1672 work with page cache in 2.6 kernel. This also fix some deadlocks
1673 and remove hack for work O_SYNC with 2.6 kernel.
1675 Severity : enhancement
1677 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1678 Details : The uninit_groups feature works in conjunction with the kernel
1679 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1680 up the pass1 processing of e2fsck. This is a read-only feature
1681 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1682 mount filesystems that have had this feature enabled.
1684 Severity : enhancement
1686 Description: Improve multi-block allocation algorithm to avoid fragmentation
1687 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1688 allocation locality and avoid filesystem fragmentation.
1690 ------------------------------------------------------------------------------
1692 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1694 * Support for kernels:
1695 2.6.9-42.0.3.EL (RHEL 4)
1696 2.6.5-7.276 (SLES 9)
1697 2.4.21-47.0.1.EL (RHEL 3)
1698 2.6.12.6 vanilla (kernel.org)
1699 2.6.16.21-0.8 (SLES10)
1700 * Recommended e2fsprogs version: 1.39.cfs2-0
1702 * The backwards-compatible /proc/sys/portals symlink has been removed
1703 in this release. Before upgrading, please ensure that you change
1704 any configuration scripts or /etc/sysctl.conf files that access
1705 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1706 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1707 in advance of the upgrade on any system running Lustre 1.4.6 or
1708 newer, since /proc/sys/lnet was added in that version.
1709 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1713 Frequency : only when quota is used
1715 Description: avoid scanning export list for quota master
1716 Details : Change the algorithms to avoid scanning export list in order
1717 to improve the efficiency.
1720 Frequency : MDS failover only, very rarely
1722 Description: "went back in time" messages on mds failover
1723 Details : The greatest transno may be lost when the current operation
1724 finishes with an error (transno==0) and the client's last_rcvd
1725 record is over-written. Save the greatest transno in the
1726 mds_last_transno for this case.
1729 Frequency : always for specific kernels and striping counts
1731 Description: client may get "Matching packet too big" without ACL support
1732 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1733 when trying to access files in certain configurations. The
1734 clients should in fact be denied when mounting because they do
1735 not understand ACLs.
1738 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1740 Description: many clients connecting with IO in progress causes connect timeouts
1741 Details : Avoid synchronous journal commits to avoid delays caused by many
1742 clients connecting/disconnecting when bulk IO is in progress.
1743 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1744 OST_IO_PORTAL to avoid delays behind potentially many pending
1748 Frequency : occasionally with multiple writers to a single file
1750 Description: shared writes to file may result in wrong size reported by stat()
1751 Details : Allow growing of kms when extent lock is cancelled
1754 Frequency : always with random mmap IO to multi-striped file
1756 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1757 Details : Only if the hole is at the end of OST object so that kms is too
1758 small. Fix is to increase kms accordingly in ll_nopage.
1761 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1763 Description: writes to a missing object would leak memory on the OST
1764 Details : If there is an inconsistency between the MDS and OST filesystems,
1765 such that the MDS references an object that doesn't exist, writes
1766 to that object will leak memory due to incorrect cleanup in the
1767 error handling path, eventually running out of memory on the OST.
1772 Description: Creating too long symlink causes lustre errors
1773 Details : Check symlink and name lengths before sending requests to MDS.
1776 Frequency : only if flock is enabled (not on by default)
1778 Description: posix locks not released on fd closure on 2.6.9+
1779 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1780 kernels, this caused such locks not to be released on fd close and
1781 then assertions on fs unmount about still used locks.
1784 Frequency : MDS failover only, very rarely
1786 Description: clients may get ASSERTION(granted_lock != NULL)
1787 Details : When request was taking a long time, and a client was resending
1788 a getattr by name lock request. The were multiple lock requests
1789 with the same client lock handle and
1790 mds_getattr_name->fixup_handle_for_resent_request found one of the
1791 lock handles but later failed with ASSERTION(granted_lock != NULL).
1796 Description: handle->h_buffer_credits > 0, assertion failure
1797 Details : h_buffer_credits is zero after truncate, causing assertion
1798 failure. This patch extends the transaction or creates a new
1802 Frequency : NFS re-export or patchless client
1803 Bugzilla : 11179, 10796
1804 Description: Crash on NFS re-export node (__d_move)
1805 Details : We do not want to hash the dentry if we don't have a lock.
1806 But if this dentry is later used in d_move, we'd hit uninitialised
1807 list head d_hash, so we just do this to init d_hash field but
1808 leave dentry unhashed.
1811 Frequency : NFS re-export or patchless client
1813 Description: NFS exports has problem with symbolic link
1814 Details : lustre client didn't properly install dentry when re-exported
1815 to NFS or running patchless client.
1818 Frequency : NFS re-export or patchless client
1820 Description: Various nfs/patchless fixes.
1821 Details : fixes reuse disconected alias for lookup process - this fixes
1822 warning "find_exported_dentry: npd != pd",
1823 fix permission error with open files at nfs.
1824 fix apply umask when do revalidate.
1827 Frequency : occasional
1829 Description: Crash on NFS re-export node
1830 Details : calling clear_page() on the wrong pointer triggered oops in
1831 generic_mapping_read().
1834 Frequency : rarely, using O_DIRECT IO
1836 Description: unaligned directio crashes client with LASSERT
1837 Details : check for unaligned buffers before trying any requests.
1840 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1842 Description: stale data returned from RAID cache
1843 Details : If only a small amount of IO is done to the RAID device before
1844 reading it again it is possible to get stale data from the RAID
1845 cache instead of reading it from disk.
1848 Frequency : always for sles10 kernel
1850 Description: sles10 support
1851 Details : ll_follow_link: compile fixes and using of nd_set_link
1852 under newer kernels.
1855 Frequency : depends on arch, kernel and compiler version, always on sles10
1858 Description: recursive or deep enough symlinks cause stack overflow
1859 Details : getting rid of large stack-allocated variable in
1863 Frequency : depends on hardware
1865 Description: lustre write performance loss in the SLES10 kernel
1866 Details : the performance loss is caused by using of write barriers in the
1867 ext3 code. The SLES10 kernel turns barrier support on by
1868 default. The fix is to undo that change for ldiskfs.
1870 ------------------------------------------------------------------------------
1872 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1874 * Support for kernels:
1875 2.6.9-42.0.3EL (RHEL 4)
1876 2.6.5-7.276 (SLES 9)
1877 2.4.21-47.0.1.EL (RHEL 3)
1878 2.6.12.6 vanilla (kernel.org)
1882 Frequency : quota enabled and large files being deleted
1884 Description: releasing more than 4GB of quota at once hangs OST
1885 Details : If a user deletes more than 4GB of files on a single OST it
1886 will cause the OST to spin in an infinite loop. Release
1887 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1892 Description: statfs data retrieved from /proc may be stale or zero
1893 Details : When reading per-device statfs data from /proc, in the
1894 {kbytes,files}_{total,free,avail} files, it may appear
1895 as zero or be out of date.
1898 Frequency : systems with MD RAID1 external journal devices
1900 Description: lconf's call to blkid is confused by RAID1 journal devices
1901 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1902 of returning all block devices that match the journal UUID.
1905 Frequency : always, for aggregate stripe size over 4GB
1907 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1908 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1909 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1912 Frequency : always if "lfs find" used on a local file/directory
1914 Description: "lfs find" segfaults if used on a local file/directory
1915 Details : The case where a directory component was not specified wasn't
1916 handled correctly. Handle this properly.
1919 Frequency : always on ppc64
1921 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1922 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1923 perform endianness conversion.
1926 Frequency : rarely (truncate to non-zero file size after write under load)
1927 Bugzilla : 10730, 10687
1928 Description: Files padded with zeros to next 4K multiple
1929 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1930 it is possible that files that are truncated to a non-zero size
1931 immediately after being written are filled with zero bytes beyond
1932 the truncated size. No file data is lost.
1934 Severity : enhancement
1936 Description: Allow recovery/failover for liblustre clients.
1937 Details : liblustre clients were unaware of failover configurations until
1940 Severity : enhancement
1942 Description: user file locks should fail when not mounting with flock option
1943 Details : Set up an error-returning stub in ll_file_operations.lock field
1944 to prevent incorrect behaviour when client is mounted without
1945 flock option. Also, set up properly f_op->flock field for
1949 Frequency : always on ia64
1951 Description: "lfs df" loops on printing out MDS statfs information
1952 Details : The obd_ioctl_data was not initialized and in some systems
1953 this caused a failure during the ioctl that did not return
1954 an error. Initialize the struct and return an error on failure.
1957 Frequency : SLES 9 only
1959 Description: Error of copying files with lustre special EAs as root
1960 Details : Client side always return success for setxattr call for lustre
1961 special xattr (currently only "trusted.lov").
1964 Frequency : rarely on clusters with both ia64+i386 clients
1966 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1967 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1968 clients are doing concurrent writes to the same file it is
1969 possible that the ia64 clients may LASSERT because the OST
1970 extent locks are not PAGE_SIZE aligned. Ensure that grown
1971 locks are always aligned on the request boundary.
1974 Frequency : specific use, occasional
1976 Description: Overwriting in use executable truncates on-disk binary image
1977 Details : If one node attempts to overwrite an executable in use by
1978 another node, we now correctly return ETXTBSY instead of
1979 truncating the file.
1981 Severity : enhancement
1983 Description: Async OSC create to avoid the blocking unnecessarily.
1984 Details : If a OST has no remain object, system will block on the creating
1985 when need to create a new object on this OST. Now, ways use
1986 pre-created objects when available, instead of blocking on an
1987 empty osc while others are not empty. If we must block, we block
1988 for the shortest possible period of time.
1993 Description: chmod on Lustre root is propagated to other clients
1994 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1995 invalid by the follow_mount time.
1998 Frequency : liblustre clients only
2000 Description: Race in 'instant cancel' lock handling could lead to such locks
2001 never to be granted in case of SMP MDS
2002 Details : Do not destroy not yet granted but cbpending locks in
2006 Frequency : replay/resend of open
2008 Description: non null lock assetion failure in mds_intent_policy
2009 Details : Trying to replay/resend lockless open requests resulted in
2010 mds_open() returning 0 with no lock. Now it sets a flag if
2011 a lock is going to be returned.
2013 Severity : enhancement
2015 Description: Checksum enhancements
2016 Details : New checksum enhancements allow for resending RPCs that failed
2019 Severity : enhancement
2021 Description: Tunables on number of dirty pages in cacche
2022 Details : Allow to set limit on number of dirty pages cached.
2027 Description: client crash on unmount - lock still has references
2028 Details : In some error handling cases it was possible to leak a lock
2029 reference on a client while accessing a file. This was not
2030 harmful to the client during operation, but would cause the
2031 client to crash when the filesystem is unmounted.
2034 Frequency : specific case, rare
2036 Description: ETXTBSY on mds though file not in use by client
2037 Details : ETXTBSY is no longer incorrectly returned when attempting to
2038 chmod or chown a directory that the user previously tried to
2039 execute or a currently-executing binary.
2042 Frequency : extremely rare except on liblustre-based clients
2044 Description: Lustre space not freed when files are deleted
2045 Details : Clean up open-unlinked files after client eviction. Previously
2046 the unlink was skipped and the files remained as orphans.
2051 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2052 Details : In some cases it was possible to send a blocking callback to
2053 a client doing a glimpse, even though that client didn't get
2054 a lock granted. When the glimpse lock is cancelled on the OST
2055 the freed lock is left on the waiting list and corrupted the list.
2058 Frequency : all core dumps
2060 Description: Broke core dumps to lustre
2061 Details : Negative dentry may be unhashed if parent does not have UPDATE
2062 lock, but some callers, e.g. do_coredump, expect dentry to be
2063 hashed after successful create, hash it in ll_create_it.
2065 ------------------------------------------------------------------------------
2067 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2069 * Support for kernels:
2070 2.6.9-42.0.2.EL (RHEL 4)
2071 2.6.5-7.276 (SLES 9)
2072 2.4.21-40.EL (RHEL 3)
2073 2.6.12.6 vanilla (kernel.org)
2077 Frequency : always on RHEL 3
2079 Description: Number of open files grows over time
2080 Details : The number of open files grows over time, whether or not
2081 Lustre is started. This was due to a filp leak introduced
2082 by one of our kernel patches.
2084 ------------------------------------------------------------------------------
2086 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2088 * Support for kernels:
2089 2.6.9-42.EL (RHEL 4)
2090 2.6.5-7.276 (SLES 9)
2091 2.4.21-40.EL (RHEL 3)
2092 2.6.12.6 vanilla (kernel.org)
2097 Bugzilla : 5719, 9635, 9792, 9684
2098 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2099 Details : If a server is under heavy load and cannot reply to new
2100 connection requests before the client resends the (re)connect,
2101 the connection handling code can behave badly if two service
2102 threads are concurrently handing separate (re)connections from
2103 the same client. Add better locking to the connection handling
2104 code, and ensure that only a single connection will be processed
2105 for a given client UUID, even if the lock is dropped.
2107 Severity : enhancement
2109 Description: add TCP zero-copy support to kernel
2110 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2111 sends if the hardware supports scatter-gather and checksumming.
2112 This allows socklnd to do client-write and server-read more
2113 efficiently and reduce CPU utilization from skbuf copying.
2116 Frequency : only if NFS exporting from client
2118 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2119 Details : If setting attributes on a file created under NFS that had
2120 never been opened it would be possible to oops the client
2121 if the file had no objects.
2124 Frequency : always for liblustre
2126 Description: liblustre client does MDS+OSTs setattr RPC for each write
2127 Details : When doing a write from a liblustre client, the client
2128 incorrectly issued an RPC to the MDS and each OST the file was
2129 striped over in order to update the timestamps. When writing
2130 with small chunks and many clients this could overwhelm the MDS
2131 with RPCs. In all cases it would slow down the write because
2132 these RPCs are unnecessary.
2134 Severity : enhancement
2136 Description: allow number of MDS service threads to be changed at module load
2137 Details : It is now possible to change the number of MDS service threads
2138 running. Adding "options mds mds_num_threads={N}" to the MDS's
2139 /etc/modprobe.conf will set the number of threads for the next
2140 time Lustre is restarted (assuming the "mds" module is also
2141 reloaded at that time). The default number of threads will
2142 stay the same, 32 for most systems.
2147 Description: OST crash if filesystem is unformatted or corrupt
2148 Details : If an OST is started on a device that has never been formatted
2149 or if the filesystem is corrupt and cannot even mount then the
2150 error handling cleanup routines would dereference a NULL pointer.
2155 Description: NULL pointer deref in llap_from_page.
2156 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2157 incorrect) mapping if the page was truncated/reclaimed while it was
2158 searched for. Check for this condition and skip such pages when
2159 doing readahead. Introduce extra check to llap_from_page() to
2160 verify page->mapping->host is non-NULL (so page is not anonymous).
2163 Frequency : Sometimes when using sys_sendfile
2165 Description: "page not covered by a lock" warnings from ll_readpage
2166 Details : sendfile called ll_readpage without right page locks present.
2167 Now we introduced ll_file_sendfile that does necessary locking
2168 around call to generic_file_sendfile() much like we do in
2172 Frequency : with certain MDS communication failures at client mount time
2174 Description: NULL pointer deref after failed client mount
2175 Details : a client connection request may delayed by the network layer
2176 and not be sent until after the PTLRPC layer has timed out the
2177 request. If the client fails the mount immediately it will try
2178 to clean up before the network times out the request. Add a
2179 reference from the request import to the obd device and delay
2180 the cleanup until the network drops the request.
2183 Frequency : occasionally during client (re)connect
2185 Description: assertion failure during client (re)connect
2186 Details : processing a client connection request may be delayed by the
2187 client or server longer than the client connect timeout. This
2188 causes the client to resend the connection request. If the
2189 original connection request is replied in this interval, the
2190 client may trip an assertion failure in ptlrpc_connect_interpret()
2191 which thought it would be the only running connect process.
2194 Frequency : only with obd_echo servers and clients that are rebooted
2196 Description: kernel BUG accessing uninitialized data structure
2197 Details : When running an obd_echo server it did not start the ping_evictor
2198 thread, and when a client was evicted an uninitialized data
2199 structure was accessed. Start the ping_evictor in the RPC
2200 service startup instead of the OBD startup.
2202 Severity : enhancement
2203 Bugzilla : 10193 (patchless)
2204 Description: Remove dependency on various unexported kernel interfaces.
2205 Details : No longer need reparent_to_init, exit_mm, exit_files,
2206 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2209 Frequency : rare (only users of deprecated and unsupported LDAP config)
2211 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2212 Details : LDAP apparently contains 'lustreName' attributes instead of
2213 'name'. A simple remapping of the name is sufficient.
2216 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2218 Description: waiting_locks_callback trips kernel BUG if client is evicted
2219 Details : Running with the dump_on_timeout debug flag turned on makes
2220 it possible that the waiting_locks_callback() can try to dump
2221 the Lustre kernel debug logs from an interrupt handler. Defer
2222 this log dumping to the expired_lock_main() thread.
2224 Severity : enhancement
2226 Description: Support NFS exporting on 2.6 kernels.
2227 Details : Implement non-rawops metadata methods for NFS server to use without
2228 changing NFS server code.
2231 Frequency : very rare (synthetic metadata workload only)
2233 Description: two racing renames might cause an MDS thread to deadlock
2234 Details : Running the "racer" program may cause one MDS thread to rename
2235 a file from being the source of a rename to being the target of
2236 a rename at exactly the same time that another thread is doing
2237 so, and the second thread has already enqueued these locks after
2238 doing a lookup of the target and is trying to relock them in
2239 order. Ensure that we don't try to re-lock the same resource.
2242 Frequency : only very large systems with liblustre clients
2244 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2245 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2246 all OSTs in parallel. This allows the eviction of stale liblustre
2247 clients to proceed much faster than if they were done in series,
2248 and also offers similar improvements for other set_info RPCs.
2253 Description: excessive CPU usage during initial read phase on client
2254 Details : During the initial read phase on a client, it would agressively
2255 retry readahead on the file, consuming too much CPU and impacting
2256 performance (since 1.4.5.8). Improve the readahead algorithm
2257 to avoid this, and also improve some other common cases (read
2258 of small files in particular, where "small" is files smaller than
2259 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2264 Description: MDS crash when receiving packet with unknown intent.
2265 Details : Do not LBUG in unknown intent case, just return -EFAULT
2267 Severity : enhancement
2268 Bugzilla : 9293, 9385
2269 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2270 Details : Do not serialize getattr (non-intent version) and statfs.
2273 Frequency : occasional, when OST network is overloaded/intermittent
2275 Description: client evicted by OST after bulk IO timeout
2276 Details : If a client sends a bulk IO request (read or write) the OST
2277 may evict the client if it is unresposive to its data GET/PUT
2278 request. This is incorrect if the network is overloaded (takes
2279 too long to transfer the RPC data) or dropped the OST GET/PUT
2280 request. There is no need to evict the client at all, since
2281 the pinger and/or lock callbacks will handle this, and the
2282 client can restart the bulk request.
2285 Frequency : Always when mmapping file with no objects
2287 Description: client crashes when mmapping file with no objects
2288 Details : Check that we actually have objects in a file before doing any
2289 operations on objects in ll_vm_open, ll_vm_close and
2295 Description: Request leak when working with deleted CWD
2296 Details : Introduce advanced request refcount tracking for requests
2297 referenced from lustre intent.
2299 Severity : Enhancement
2301 Description: Cache open file handles on client.
2302 Details : MDS now will return special lock along with openhandle, if
2303 requested and client is allowed to hold openhandle, even if unused,
2304 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2305 closing files for every read/write openration.
2307 Severity : Enhancement
2309 Description: Cache open negative dentries on client when possible.
2310 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2311 negative dentries on lock revocation.
2316 Description: Remounting a client read-only wasn't possible with a zconf mount
2317 Details : It wasn't possible to remount a client read-only with llmount.
2319 Severity : enhancement
2320 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2321 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2322 MPICH is included to add Lustre-specific ADIO interfaces.
2323 This is based closely on the UFS ADIO layer and only differs
2324 in file creation, in order to allow the OST striping to be set.
2325 This is user-contributed code and not supported by CFS.
2330 Description: extended inode attributes (immutable, append-only) work improperly
2331 when 2.4 and 2.6 kernels are used on client/server or vice versa
2332 Details : Introduce kernel-independent values for these flags.
2334 Severity : enhancement
2337 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2338 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2339 granularity. Now, Fractional MB granularity is be supported,
2340 this is very useful for low memory system.
2342 Severity : enhancement
2344 Description: Getattr by fid
2345 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2346 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2347 have locks and avoids getattr rpc requests for it on every stat.
2350 Frequency : Always, for filesystems larger than 2TB
2352 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2353 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2354 filesystems larger than 2TB to be mounted with mballoc enabled.
2357 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2359 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2360 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2361 filesystems larger than 2TB is mounted then the kernel will
2362 silently corrupt the start of the filesystem. CONFIG_LBD is
2363 enabled for all CFS-supported kernels, but the possibility of
2364 this happening with a modified kernel config exists.
2366 Severity : enhancement
2368 Description: add client O_DIRECT support for 2.6 kernels
2369 Details : It is now possible to do O_DIRECT reads and writes to files
2370 in the Lustre client mountpoint on 2.6 kernel clients.
2372 Severity : enhancement
2374 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2375 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2376 parallel, not waiting for response from every OST before sending
2377 a rpc to the next OST.
2382 Description: setattr vs write race when updating file timestamps
2383 Details : Client processes that update a file timestamp into the past
2384 right after writing to the file (e.g. tar) it is possible that
2385 the updated file modification time can be reset to the current
2386 time due to a race between processing the setattr and write RPC.
2388 Severity : enhancement
2390 Description: Bring 'lfs find' closer in line with regular Linux find.
2391 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2392 -print0 options and obtains all the needed info through the lustre
2395 Severity : enhancement
2397 Description: support up to 1024 configured devices on one node
2398 Details : change obd_dev array from statically allocated to dynamically
2399 allocated structs as they are first used to reduce memory usage
2404 Description: Flush dirty partially truncated pages during truncate
2405 Details : Immediatelly flush partially truncated pages in filter_setattr,
2406 this way we completely avoid having any pages in page cache on OST
2407 and can retire ugly workarounds during writes to flush such pages.
2412 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2413 Details : protect inode from truncation within vfs_unlink() context
2414 just take a reference before calling vfs_unlink() and release it
2415 when parent's i_sem is free.
2420 Description: last_id value checked outside lock on OST caused LASSERT failure
2421 Details : If there were multiple MDS->OST object precreate requests in
2422 flight, it was possible that the OST's last object id was checked
2423 outside a lock and incorrectly tripped an assertion. Move checks
2424 inside locks, and discard old precreate requests.
2427 Frequency : always, if extents are used on OSTs
2429 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2430 Details : OSTs using the extents format would not zero the high 16 bits of
2431 the index physical block number. This is not a problem for any
2432 OST filesystems smaller than 16TB, and no kernels support ext3
2433 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2434 new/modified files) and can be fixed for existing filesystems
2435 with e2fsprogs-1.39-cfs1.
2440 Description: import connection selection may be incorrect if timer wraps
2441 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2442 import connections to be ignored if the 32-bit jiffies counter
2443 wraps. Use a 64-bit jiffies counter.
2446 Frequency : very large clusters immediately after boot
2448 Description: LNET request buffers exhausted under heavy short-term load
2449 Details : If a large number of client requests are generated on a service
2450 that has previously never seen so many requests it is possible
2451 that the request buffer growth cannot keep up with the spike in
2452 demand. Instead of dropping incoming requests, they are held in
2453 the LND until the RPC service can accept more requests.
2456 Frequency : Sometimes during replay
2458 Description: Assertion failure in ll_local_open after replay.
2459 Details : If replay happened on an open request reply before we were able
2460 to set replay handler, reply will become not swabbed tripping the
2461 assertion in ll_local_open. Now we set the handler right after
2462 recognising of open request
2465 Frequency : very rare
2467 Description: kernel reports "badness in vsnprintf"
2468 Details : Reading from the "recovery_status" /proc file in small chunks
2469 may cause a negative length in lprocfs_obd_rd_recovery_status()
2470 call to vsnprintf() (which is otherwise harmless). Exit early
2471 if there is no more space in the output buffer.
2473 Severity : enhancement
2475 Description: clear OBD RPC statistics by writing to them
2476 Details : It is now possible to clear the OBD RPC statistics by writing
2477 to the "stats" file.
2482 Description: Client mtime is not the same on different clients after utimes
2483 Details : In some cases, the client was using the utimes() syscall on
2484 a file cached on another node. The clients now validate the
2485 ctime from the MDS + OSTs to determine which one is right.
2490 Description: Inability to activate failout mode
2491 Details : lconf script incorrectly assumed that in pythong string's numeric
2492 value is used in comparisons.
2495 Frequency : always with multiple stripes per file
2497 Description: Inefficient object allocation for mutli-stripe files
2498 Details : When selecting which OSTs to stripe files over, for files with
2499 a stripe count that divides evenly into the number of OSTs,
2500 the MDS is always picking the same starting OST for each file.
2501 Return the OST selection heuristic to the original design.
2506 Description: mount failures may take full timeout to return an error
2507 Details : Under some heavy load conditions it is possible that a
2508 failed mount can wait for the full obd_timeout interval,
2509 possibly several minutes, before reporting an error.
2510 Instead return an error as soon as the status is known.
2512 Frequency : quota enabled and large files being deleted
2514 Description: releasing more than 4GB of quota at once hangs OST
2515 Details : If a user deletes more than 4GB of files on a single OST it
2516 will cause the OST to spin in an infinite loop. Release
2517 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2522 Description: statfs data retrieved from /proc may be stale or zero
2523 Details : When reading per-device statfs data from /proc, in the
2524 {kbytes,files}_{total,free,avail} files, it may appear
2525 as zero or be out of date.
2528 Frequency : systems with MD RAID1 external journal devices
2530 Description: lconf's call to blkid is confused by RAID1 journal devices
2531 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2532 of returning all block devices that match the journal UUID.
2535 Frequency : always, for aggregate stripe size over 4GB
2537 Description: assertion fails when trying to use 4GB stripe size
2538 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2539 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2542 Frequency : always on ppc64
2544 Description: the first write on an ext3 filesystem with mballoc got stuck
2545 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2546 perform endianness conversion.
2548 ------------------------------------------------------------------------------
2550 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2552 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2553 INTEROPERATE with older versions automatically. Please read the
2554 user documentation before upgrading any part of a live system.
2555 * WARNING: Lustre networking configuration changes are required with
2556 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2559 * Support for kernels:
2560 2.6.9-22.0.2.EL (RHEL 4)
2561 2.6.5-7.244 (SLES 9)
2562 2.6.12.6 vanilla (kernel.org)
2565 Severity : enhancement
2566 Bugzilla : 7981/8208
2567 Description: Introduced Lustre Networking (LNET)
2568 Details : LNET is new networking infrastructure for Lustre, it includes
2569 a reorganized network configuration mode (see the user
2570 documentation for full details) as well as support for routing
2571 between different network fabrics. Lustre Networking Devices
2572 (LNDs) for the supported network fabrics have also been
2573 created for this new infrastructure.
2575 Severity : enhancement
2576 Description: Introduced Access control lists
2577 Details : clients can set ACLs on files and directories in order to have
2578 more fine-grained permissions than the standard Unix UGO+RWX.
2579 The MDS must be started with the "-o acl" mount option.
2581 Severity : enhancement
2582 Description: Introduced filesystem quotas
2583 Details : Administrators may now establish per-user quotas on the
2586 Severity : enhancement
2588 Description: Configuration change for the XT3
2589 The PTLLND is now used to run Lustre over Portals on the XT3
2590 The configure option(s) --with-cray-portals are no longer used.
2591 Rather --with-portals=<path-to-portals-includes> is used to
2592 enable building on the XT3. In addition to enable XT3 specific
2593 features the option --enable-cray-xt3 must be used.
2598 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2599 Details : A race between allocating a new llcd and re-getting the llcd_lock
2600 allowed another thread to grab newly-allocated llcd.
2602 Severity : enhancement
2604 Description: 2.6 OST async journal commit and locking fix to improve performance
2605 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2606 2.6 kernels are now async as they already were in 2.4 kernels so
2607 that they can commit concurrently with the network bulk transfer.
2608 For block-allocated files the filter allocation semaphore is held
2609 to avoid filesystem fragmentation during allocation. BKL lock
2610 removed for 2.6 xattr operations where it is no longer needed.
2615 Description: lconf incorrectly determined whether two IP networks could talk
2616 Details : In some more complicated routing and multiple-network
2617 configurations, lconf will avoid trying to make a network
2618 connection to a disjoint part of the IP space. It was doing the
2619 math incorrectly for one set of cases.
2624 Description: Fix for potential infinite loop processing records in an llog.
2625 Details : If an llog record is corrupted/zeroed, it is possible to loop
2626 forever in llog_process(). Validate the llog record length
2627 and skip the remainder of the block on error.
2630 Frequency : occasional (liblustre only)
2632 Description: liblustre could not open files whose last component is a symlink
2633 Details : sysio_path_walk() would incorrectly pass the open intent to
2634 intermediate path components.
2637 Frequency : rare (liblustre only with non-standard tuning)
2638 Bugzilla : 7201 (7350)
2639 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2640 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2641 after it has been released, only do this for referenced locks
2643 Severity : enhancement
2645 Description: specifying an (invalid) directory default stripe_size of -1
2646 would reset the directory default striping
2647 Details : stripe_size -1 was used internally to signal directory stripe
2648 removal, now use "all default" to signal dir stripe removal
2649 as a directory striping of "all default" is not useful
2652 Frequency : common for large clusters running liblustre clients
2654 Description: doing an ls when liblustre clients are running is slow
2655 Details : sending a glimpse AST to a liblustre client waits for every AST
2656 to time out, as liblustre clients will not respond. Since they
2657 cannot cache data we refresh the OST lock LVB from disk instead.
2659 Severity : enhancement
2661 Description: doing an ls at the same time as file IO can be slow
2662 Details : enqueue and other "small" requests can be blocked behind many
2663 large IO requests. Create a new OST IO portal for non-IO
2664 requests so they can be processed faster.
2667 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2669 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2670 created file with mode 000
2671 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2672 allow an ftruncate() as a non-root user to a file with mode 000.
2673 Lustre now respects this flag to disable mode checks when
2674 truncating a file owned by the user
2677 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2680 Description: Revoking locks from clients that went dead or catatonic might take
2682 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2683 cancellation of such locks instant on servers without waiting for
2684 any reply from clients. Clients drops these locks when cancel
2685 notification from server is received without replying.
2688 Frequency : liblustre-only, when liblustre client dies or becomes busy
2690 Description: Doing ls on Linux clients can take a long time with active
2692 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2693 granting such locks to it in the first place if possible. Locks
2694 are taken by proxy on the OST during the read or write and
2695 dropped immediately afterward. Add connect flags handling, do
2696 not grant locks to liblustre clients for glimpse ASTs.
2698 Severity : enhancement
2700 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2701 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2702 sequential reads and behaves badly with applications doing random
2703 reads. Improve it by reading ahead at least read region, and
2704 avoiding excessive large RPC for small reads.
2706 Severity : enhancement
2708 Description: Creating more than 1000 files for a single job may cause a load
2709 imbalance on the OSTs if there are also a large number of OSTs.
2710 Details : qos_prep_create() uses an OST index reseed value that is an
2711 even multiple of the number of available OSTs so that if the
2712 reseed happens in the middle of the object allocation it will
2713 still utilize the OSTs as uniformly as possible.
2718 Description: OST or MDS may oops in ping_evictor_main()
2719 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2720 but doesn't restart at beginning of obd_exports_timed list
2723 Severity : enhancement
2725 Description: improve by-nid export eviction on the MDS and OST
2726 Details : allow multiple exports with the same NID to be evicted at one
2727 time without re-searching the exports list.
2730 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2732 Description: MDS may oops in groups_free()
2733 Details : in rare race conditions a newly allocated group_info struct is
2734 freed again, and this can be NULL. The 2.4 compatibility code
2735 for groups_free() checked for a NULL pointer, but 2.6 did not.
2738 Frequency : common for liblustre clients doing little filesystem IO
2739 Bugzilla : 9352, 7313
2740 Description: server may evict liblustre clients accessing contended locks
2741 Details : if a client is granted a lock or receives a completion AST
2742 with a blocking AST already set it would not reply to the AST
2743 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2746 Frequency : lfs setstripe, only systems with more than 160 OSTs
2748 Description: unable to set striping with a starting offset beyond OST 160
2749 Details : llapi_create_file() incorrectly limited the starting stripe
2750 index to the maximum single-file stripe count.
2753 Frequency : LDAP users only
2755 Description: lconf did not handle in-kernel recovery with LDAP properly
2756 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2758 Severity : enhancement
2760 Description: bind OST threads to NUMA nodes to improve performance
2761 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2762 node and do their allocations there to localize memory access
2764 Severity : enhancement
2766 Description: llmount can determine client NID directly from Myrinet (GM)
2767 Details : the client NID code from gmnalnid was moved directly into
2768 llmount, removing the need to use this or specifying the
2769 client NID explicitly when mounting GM clients with zeroconf
2772 Frequency : if client is started with down MDS
2774 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2775 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2776 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2777 ptlrpc_import_delayed_req() because we didn't check if the
2778 request was interrupted, and we also didn't break out of the
2779 event loop if there was no timeout
2784 Description: data loss during non-page-aligned writes to a single file from
2785 both multiple nodes and multiple threads on one node at same time
2786 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2787 inconsistency led to false short-reads, that were cached and later
2788 used by ->prepare_write() to fill in partially written page,
2789 leading to data loss.
2792 Frequency : always, if lconf --abort_recovery used
2794 Description: lconf --abort_recovery fails with 'Operation not supported'
2795 Details : lconf was attempting to abort recovery on the MDT device and not
2798 Severity : enhancement
2800 Description: remove cleanup logs
2801 Details : replace lconf-generated cleanup logs with lustre internal
2802 cleanup routines. Eliminates the need for client-cleanup and
2805 Severity : enhancement
2807 Description: add support for EAs (user and system) on lustre filesystems
2808 Details : it is now possible to store extended attributes in the Lustre
2809 client filesystem, and with the user_xattr mount option it
2810 is possible to allow users to store EAs on their files also
2812 Severity : enhancement
2814 Description: Add possibility (config option) to show minimal available OST free
2816 Details : When compiled with --enable-mindf configure option, statfs(2)
2817 (and so, df) will return least minimal free space available from
2818 all OSTs as amount of free space on FS, instead of summary of
2819 free spaces of all OSTs.
2821 Severity : enhancement
2823 Description: do not expand extent locks acquired on OST-side
2824 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2825 by server: they are not cached anyway.
2828 Frequency : when mmap is used/binaries executed from Lustre
2830 Description: Unmmap pages before throwing them away from read cache.
2831 Details : llap_shrink cache now attempts to unmap pages before discarding
2832 them (if unmapping failed - do not discard). SLES9 kernel has
2833 extra checks that trigger if this unmapping is not done first.
2838 Description: lconf didn't resolve symlinks before checking to see whether a
2839 given mountpoint was already in use
2842 Frequency : when migrating failover services
2843 Bugzilla : 6395, 9514
2844 Description: When migrating a subset of services from a node (e.g. failback
2845 from a failover service node) the remaining services would
2846 time out and evict clients.
2847 Details : lconf --force (implied by --failover) sets the global obd_timeout
2848 to 5 seconds in order to quickly disconnect, but this caused
2849 other RPCs to time out too quickly. Do not change the global
2850 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2852 Severity : enhancement
2853 Frequency : if MDS is started with down OST
2854 Bugzilla : 9439,5706
2855 Description: Allow startup/shutdown of an MDS without depending on the
2856 availability of the OSTs.
2857 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2858 Add appropriate locking and lov-osc refcounts for safe
2859 cleaning. Add osc abort_inflight calls in case the
2860 synchronize never started.
2863 Frequency : occasional (Cray XT3 only)
2865 Description: root not authorized to access files in CRAY_PORTALS environment
2866 Details : The client process capabilities were not honoured on the MDS in
2867 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2868 been accessed by an authorized user then root was able to access
2869 the file on the local client also. The root user capabilities
2870 are now allowed on the MDS, as this environment has secure UID.
2873 Frequency : occasional
2875 Description: ldiskfs "too long searching" message happens too often
2876 Details : A debugging message (otherwise harmless) prints too often on
2877 the OST console. This has been reduced to only happen when
2878 there are fragmentation problems on the filesystem.
2883 Description: Division by zero in statfs when all OSCs are inactive
2884 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2885 lov_statfs divides by value returned by lov_get_stripecnt().
2889 Bugzilla : 9489, 3273
2890 Description: First write from each client to each OST was only 4kB in size,
2891 to initialize client writeback cache, which caused sub-optimal
2892 RPCs and poor layout on disk for the first writen file.
2893 Details : Clients now request an initial cache grant at (re)connect time
2894 and so that they can start streaming writes to the cache right
2895 away and always do full-sized RPCs if there is enough data.
2896 If the OST is rebooted the client also re-establishes its grant
2897 so that client cached writes will be honoured under the grant.
2902 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2903 Details : Now I/O RPCs go to different portal number and (presumably) fast
2904 lock requests (and glimses) and other RPCs get their own service
2905 threads pool that should be able to service those RPCs
2908 Severity : enhancement
2910 Description: Ability to exchange lustre version between client and servers and
2911 issue warnings at client side if client is too old. Also for
2912 liblustre clients there is ability to refuse connection of too old
2914 Details : New 'version' field is added to connect data structure that is
2915 filled with version info. That info is later checked by server and
2919 Frequency : rare, liblustre only.
2920 Bugzilla : 9296, 9581
2921 Description: Two simultaneous writes from liblustre at offset within same page
2922 might proceed at the same time overwriting eachother with stale
2924 Details : I/O lock withing llu_file_prwv was released too early, before data
2925 actually was hitting the wire. Extended lock-holding time until
2926 server acknowledges receiving data.
2929 Frequency : extremely rare. Never observed in practice.
2931 Description: avoid generating lustre_handle cookie of 0.
2932 Details : class_handle_hash() generates handle cookies by incrementing
2933 global counter, and can hit 0 occasionaly (this is unlikely, but
2934 not impossible, because initial value of cookie counter is
2935 selected randonly). Value of 0 is used as a sentinel meaning
2936 "unassigned handle" --- avoid it. Also coalesce two critical
2937 sections in this function into one.
2939 Severity : enhancement
2941 Description: allow liblustre clients to delegate truncate locking to OST
2942 Details : To avoid overhead of locking, liblustre client instructs OST to
2943 take extent lock in ost_punch() on client's behalf. New connection
2944 flag is added to handle backward compatibility.
2946 Severity : enhancement
2947 Bugzilla : 4928, 7341, 9758
2948 Description: allow number of OST service threads to be specified
2949 Details : a module parameter allows the number of OST service threads
2950 to be specified via "options ost ost_num_threads={N}" in the
2951 OSS's /etc/modules.conf or /etc/modprobe.conf.
2955 Bugzilla : 6146, 9635, 9895
2956 Description: servers crash with bad pointer in target_handle_connect()
2957 Details : In rare cases when a client is reconnecting it was possible that
2958 the connection request was the last reference for that export.
2959 We would temporarily drop the export reference and get a new
2960 one, but this may have been the last reference and the export
2961 was just destroyed. Get new reference before dropping old one.
2963 Severity : enhancement
2964 Frequency : if client is started with failover MDS
2966 Description: Allow multiple MDS hostnames in the mount command
2967 Details : Try to read the configuration from all specified MDS
2968 hostnames during a client mount in case the "primary"
2971 Severity : enhancement
2973 Description: Stop sending data to evicted clients as soon as possible.
2974 Details : Check if the client we are about to send or are sending data to
2975 was evicted already. (Check is done every second of waiting,
2976 for which l_wait_event interface was extended to allow checking
2977 of exit condition at specified intervals).
2980 Frequency : rare, normally only when NFS exporting is done from client
2982 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2984 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2987 Frequency : always, if srand() is called before liblustre initialization
2989 Description: Liblustre uses system PRNG disturbing its usage by user application
2990 Details : Introduce internal to lustre fast and high-quality PRNG for
2991 lustre usage and make liblustre and some other places in generic
2992 lustre code to use it.
2994 Severity : enhancement
2995 Bugzilla : 9477, 9557, 9870
2996 Description: Verify that the MDS configuration logs are updated when xml is
2997 Details : Check if the .xml configuration logs are newer than the config
2998 logs stored on the MDS and report an error if this is the case.
2999 Request --write-conf, or allow starting with --old_conf.
3001 Severity : enhancement
3003 Description: Handle symlinks in the path when checking if Lustre is mounted.
3004 Details : Resolve intermediate symlinks when checking if a client has
3005 mounted a filesystem to avoid duplicate client mounts.
3010 Description: lconf can hit an error exception but still return success.
3011 Details : The lconf command catches the Command error exception at the top
3012 level script context and will exit with the associated exit
3013 status, but doesn't ensure that this exit status is non-zero.
3018 Description: failure of ptlrpc thread startup can cause oops
3019 Details : Starting a ptlrpc service thread can fail if there are a large
3020 number of threads or the server memory is very fragmented.
3021 Handle this without oopsing.
3024 Frequency : always, only if liblustre and non-default acceptor port was used
3026 Description: liblustre cannot connect to servers with non-default acceptor port
3027 Details : tcpnal_set_default_params() was not called and was therefore
3028 ignoring the environment varaible TCPNAL_PORT, as well as other
3029 TCPNAL_ environment variables
3034 Description: two objects could be created on the same OST for a single file
3035 Details : If an OST is down, in some cases it was possible to create two
3036 objects on a single OST for a single file. No problems other
3037 than potential performance impact and spurious error messages.
3041 Bugzilla : 5681, 9562
3042 Description: Client may oops in ll_unhash_aliases
3043 Details : Client dcache may become inconsistent in race condition.
3044 In some cases "getcwd" can fail if the current directory is
3050 Description: Inode refcounting problems in NFS export code
3051 Details : link_raw functions used to call d_instantiate without obtaining
3052 extra inode reference first.
3056 Bugzilla : 9942, 9903
3057 Description: Referencing freed requests leading to crash, memleaks with NFS.
3058 Details : We used to require that call to ll_revalidate_it was always
3059 followed by ll_lookup_it. Also with revalidate_special() it is
3060 possible to call ll_revalidate_it() twice for the same dentry
3061 even if first occurence returned success. This fix changes semantic
3062 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3063 reference on a request referred from the intent.
3064 ll_intent_release() then releases such a request.
3067 Frequency : rare, normally benchmark loads only
3069 Description: unlinked inodes were kept in memory on the client
3070 Details : If a client is repeatedly creating and unlinking files it
3071 can accumulate a lot of stale inodes in the inode slab cache.
3072 If there is no other client load running this can cause the
3073 client node to run out of memory. Instead flush old inodes
3074 from client cache that have the same inode number as a new inode.
3077 Frequency : SLES9 2.6.5 kernel and long filenames only
3078 Bugzilla : 9969, 10379
3079 Description: utime reports stale NFS file handle
3080 Details : SLES9 uses out-of-dentry names in some cases, which confused
3081 the lustre dentry revalidation. Change it to always use the
3085 Frequency : rare, unless heavy write-truncate concurrency is continuous
3086 Bugzilla : 4180, 6984, 7171, 9963, 9331
3087 Description: OST becomes very slow and/or deadlocked during object unlink
3088 Details : filter_destroy() was holding onto the parent directory lock
3089 while truncating+unlinking objects. For very large objects this
3090 may block other threads for a long time and slow overall OST
3091 responsiveness. It may also be possible to get a lock ordering
3092 deadlock in this case, or run out of journal credits because of
3093 the combined truncate+unlink. Solution is to do object truncate
3094 first in one transaction without parent lock, and then do the
3095 final unlink in a new transaction with the parent lock. This
3096 reduces the lock hold time dramatically.
3099 Frequency : rare, 2.4 kernels only
3101 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3102 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3103 before unmounting their filesystem in order to avoid deadlock.
3104 The kernel_locked() function in 2.4 kernels only checks whether
3105 the kernel lock is held, not whether it is this process that is
3106 holding it as 2.6 kernels do.
3111 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3112 Details : The client ptlrpc code may be trying to reconnect to a down
3113 server before a previous connection attempt has timed out.
3114 Increase the reconnect interval to be longer than the connection
3115 timeout interval to avoid sending duplicate connections to
3119 Frequency : echo_client brw_test command
3121 Description: fix echo_client to work with OST preallocated code
3122 Details : OST preallocation code (5137) didn't take echo_client IO path
3123 into account: echo_client calls filter methods outside of any
3124 OST thread and, hence, there is no per-thread preallocated
3125 pages and buffers to use. Solution: hijack pga pages for IO. As
3126 a byproduct, this avoids unnecessary data copying.
3130 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3131 Description: Client can oops in mdc_commit_close() after open replay
3132 Details : It was possible for the MDS to return an open request with no
3133 transaction number in mds_finish_transno() if the client was
3134 evicted, but without actually returning an error. Clients
3135 would later try to replay that open and may trip an assertion
3136 Simplify the client close codepath, and always return an error
3137 from the MDS in case the open is not successful.
3140 Frequency : rare, 2.6 OSTs only
3142 Description: OST may deadlock under high load on fragmented files
3143 Details : If there was a heavy load and highly-fragmented OST filesystems
3144 it was possible to have all the OST threads deadlock waiting on
3145 allocation of biovecs, because the biovecs were not released
3146 until the entire RPC IO was completed. Instead, release biovecs
3147 as soon as they are complete to ensure forward IO progress.
3149 Severity : enhancement
3151 Description: Support for specifying external journal device at mount
3152 Details : If an OST or MDS device is formatted with an external journal
3153 device, this device major/minor is stored in the ext3 superblock
3154 and may not be valid for failover. Allow detecting and
3155 specifying the external journal at mount time.
3160 Description: Mounting an MDS with pending unlinked files may cause oops
3161 Details : target_finish_recovery() calls mds_postrecov() which returned
3162 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3163 considers this an error and immediately begins cleaning up the
3164 lov, just after starting the mds_lov process
3166 Severity : enhancement
3168 Description: Implement 'lfs df' to report actual free space on per-OST basis
3169 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3170 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3171 usage of objects; '-h' to report in human readable format.
3173 ------------------------------------------------------------------------------
3175 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3182 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3183 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3184 at startup during orphan unlinking. Instead, skip buddy update
3185 before setup, as it will be regenerated after recovery is complete.
3190 Description: If an OST is inactive, its locks might reference stale inodes.
3191 Details : lov_change_cbdata() must iterate over all namespaces, even if
3192 they are inactive to clear inode references from the lock.
3194 Severity : enhancement
3195 Frequency : occasional, if non-standard max_dirty_mb used
3197 Description: Client will block write RPCs if not enough grant
3198 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3199 then the client will block writes while waiting for another RPC
3200 to complete instead of consuming its dirty limit. With change
3201 we get improved performance when max_dirty_mb is small.
3203 Severity : enhancement
3204 Bugzilla : 3389, 6253
3205 Description: Add support for supplementary groups on the MDS.
3206 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3207 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3208 lookups for user supplementary groups into a cache.
3212 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3213 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3214 for liblustre O_EXCL is enforced by MDS.
3218 Description: readdir never returns NULL in liblustre.
3219 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3220 dentry in struct dirent.
3224 Frequency : liblustre only, depends on application IO pattern
3225 Description: liblustre clients evicted if not contacting servers
3226 Details : Don't put liblustre clients into the ping_evictor list, so
3227 they will not be evicted by the pinger ever.
3229 Severity : enhancement
3231 Description: Add ability to evict clients by NID from MDS.
3232 Details : By echoing "nid:$NID" string into
3233 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3234 $NID would be instantly evicted from this MDS and from all active
3235 OSTs connected to it.
3239 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3240 Details : lookup_it_finish() used to query file size from OSTs that was not
3245 Description: service threads change working directory to that of init
3246 Details : Starting lustre service threads may pin the working directory
3247 of the parent thread, making that filesystem busy. Threads
3248 now change to the working directory of init to avoid this.
3252 Frequency : during shutdown only
3253 Description: shutdown with a failed MDS or OST can cause unmount to hang
3254 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3259 Frequency : 2.6 only
3260 Description: chmod/chown may include an extra supplementary group
3261 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3262 supplementary group and if none is specified this is used.
3265 Bugzilla : 5479 (6816)
3267 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3268 Details : If lookup is in progress on a file that is unlinked we might try
3269 to revalidate the inode and fail in revalidate after lookup is
3270 complete and ll_file_open() enqueues the open again but
3271 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3274 Frequency : always, if lconf --abort_recovery used
3276 Description: lconf --abort_recovery fails with 'Operation not supported'
3277 Details : lconf was attempting to abort recovery on the MDT device and not
3280 ------------------------------------------------------------------------------
3282 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3287 Frequency : rare (only unsupported configurations with a node running as an
3289 Bugzilla : 6514, 5137
3290 Description: Mounting a Lustre file system on a node running as an OST could
3292 Details : OSTs now preallocates memory needed to write out data at
3293 startup, instead of when needed, to avoid having to
3294 allocate memory in possibly low memory situations.
3295 Specifically, if the file system is mounted on on OST,
3296 memory pressure could force it to try to write out data,
3297 which it needed to allocate memory to do. Due to the low
3298 memory, it would be unable to do so and the node would
3299 become unresponsive.
3301 Severity : enhancement
3303 Description: Addition of lconf --service command line option
3304 Details : lconf now accepts a '--service <arg>' option, which is
3305 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3307 Severity : enhancement
3309 Description: Failover mode is now the default for OSTs.
3310 Details : By default, OSTs will now run in failover mode. To return to
3311 the old behaviour, add '--failout' to the lmc line for OSTs.
3313 Severity : enhancement
3315 Description: Health checks are now provided for MDS and OSTs
3316 Details : Additional detailed health check information on MSD and OSTs
3317 is now provided through the procfs health_check value.
3320 Frequency : occasional, depends on IO load
3322 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3323 after numerous create/delete cycles
3324 Details : The ext3 inode allocation policy would not allocate new inodes
3325 very well on the OSTs because there are no new directories
3326 being created. Instead we look for groups with free space if
3327 the parent directories are nearly full.
3332 Description: Network or server problems during mount may cause partially
3333 mounted clients instead of returning an error.
3334 Details : The config llog parsing code may overwrite the error return
3335 code during mount error handling, returning success instead
3341 Description: MDS can fail to allocate large reply buffers
3342 Details : After long uptimes the MDS can fail to allocate large reply
3343 buffers (e.g. zconf client mount config records) due to memory
3344 fragmentation or consumption by the buffer cache. Preallocate
3345 some large reply buffers so that these replies can be sent even
3346 under memory pressure.
3350 Frequency : rare (liblustre)
3351 Description: fsx running with liblustre complained that using truncate() to
3352 extend the file doesn't work. This patch corrects that issue.
3353 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3354 ATTR_SIZE and lsm use in llu_setattr_raw.
3358 Frequency : rare, only 2.6 kernels
3359 Description: Unusual file access patterns on the MDS may result in inode
3360 data being lost in very rare circumstances.
3361 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3362 code in the 2.6 kernel caused the inode and/or EA data not to
3363 be read from disk, causing single-file corruption.
3367 Frequency : rare, only 2.6 filesystems using extents
3368 Description: Heavy concurrent write and delete load may cause data corruption.
3369 Details : It was possible under high-load situations to have an extent
3370 metadata block in the block device cache from a just-unlinked
3371 file overwrite a newly-allocated data block. We now unmap any
3372 metadata buffers that alias just-allocated data blocks.
3376 Frequency : filesystems with default stripe_count larger than 77
3377 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3378 Details : lconf specifies an inode size of 4096 bytes when the default
3379 stripe_count is larger than 77. This conflicts with the default
3380 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3381 this case to avoid pinning too much memory for large EAs.
3383 ------------------------------------------------------------------------------
3385 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3390 Frequency : rare (extremely heavy IO load with hundreds of clients)
3392 Description: Client is evicted, gets IO error writing to file
3393 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3394 caused clients to be evicted for AST timeouts. The fixes in
3395 bug 5192 mean we no longer need to have such short AST timeouts
3396 so ldlm_timeout has been increased.
3399 Frequency : occasional during --force or --failover shutdown under load
3400 Bugzilla : 5949, 4834
3401 Description: Server oops/LBUG if stopped with --force or --failover under load
3402 Details : a collection of import/export refcount and cleanup ordering
3403 issues fixed for safer force cleanup
3406 Frequency : only filesystems larger than 120 OSTs
3407 Bugzilla : 5990, 6223
3408 Description: lfs getstripe would oops on a very large filesystem
3409 Details : lov_getconfig used kfree on vmalloc'd memory
3412 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3413 Bugzilla : 6242, 6243
3414 Description: reading from files that had been truncated to a non-zero size
3415 but never opened returned no data
3416 Details : ll_file_read() reads zeros from no-object files to EOF
3421 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3422 Details : There's little harm in aborting MDS/OSS recovery and letting it
3423 try again, so I removed the LASSERT and return an error instead.
3425 Severity : enhancement
3427 Description: New debugging infrastructure for tracking down data corruption
3428 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3429 (b) cover more of the client-side code path, and (c) try to narrow
3430 down where problems occurred
3434 Bugzilla : 3819, 4364, 4397, 6313
3435 Description: Racing close and eviction MDS could cause assertion in mds_close
3436 Details : It was possible to get multiple mfd references during close and
3437 client eviction, leading to one thread referencing a freed mfd.
3439 Severity: : enhancement
3440 Bugzilla : 3262, 6359
3441 Description: Attempts to reconnect to servers are now more aggressive.
3442 Details : This builds on the enhanced upcall-less recovery that was added
3443 in 1.4.2. When trying to reconnect to servers, clients will
3444 now try each server in the failover group every 10 second