1 tbd Sun Microsystems, Inc.
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
18 Details : If an OST is down the MDS will hang indefinitely in
19 obd_statfs_rqset() waiting for the statfs data. While for
20 MDS QOS usage of statfs, it should not stuck in waiting.
22 Severity : enhancement
24 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
26 Severity : enhancement
28 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
32 Description: make mgs_setparam() handle fsname containing dash
33 Details : fsname containing a dash does not work with lctl conf_param
35 Severity : enhancement
37 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
39 Severity : enhancement
41 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
45 Frequency : rare, at shutdown
46 Description: access already free / zero obd_namespace.
47 Details : if client_disconnect_export was called without force flag set,
48 and exist connect request in flight, this can produce access to
49 NULL pointer (or already free pointer) when connect_interpret
50 store ocd flags in obd_namespace.
54 Frequency : only at startup
55 Description: not alloc memory with spinlock held.
56 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
59 Severity : enhancement
61 Description: make lustre randomly fail allocating memory
62 Details : Make lustre randomly failed allocating memory for testing purpose.
64 Severity : enhancement
66 Description: lost problems with lov objid file
67 Details : Fixes some scability and access to not inited memory problems
68 in work with lov objdid file.
73 Description: lfs find does not continue on file error
74 Details : Continue other files processing when a file/dir is absent.
78 Description: Inconsistent usage of lustre_pack_reply()
79 Details : Standardize the usage of lustre_pack_reply() such that it
80 always generate a CERROR on failure.
83 Frequency : occasional
85 Description: MDS hang or stay in waiting lock
86 Details : If client receive lock with CBPENDING flag ldlm need send lock
87 cancel as separate rpc, to avoid situation when cancel request
88 can't processed due all i/o threads stay in wait lock.
92 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
96 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
100 Description: Update to SLES9 kernel-2.6.5-7.287.3.
104 Description: Update to RHEL4 latest kernel.
106 Severity : enhancement
108 Description: Build SLES10 patchless client fails
109 Details : The configure was broken by run ./configure with
110 --with-linux-obj=.... argument for patchless client. When the
111 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
112 can't search header adequately. Use absolute path such as
113 -I($LINUX)/include instead.
117 Description: interrupt oig_wait produce painc on resend.
118 Details : brw_redo_request can be used for resend requests from ptlrpcd and
119 private set, and this produce situation when rq_ptlrpcd_data not
120 copyed to new allocated request and triggered LBUG on assert
121 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
122 ptlrpcd set if request is changed and can be safety changed to use
127 Description: LASSERT_{REQ,REP}SWAB macros are buggy
128 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
129 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
130 these macros and replace them with their difinitions instead.
134 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
135 Details : Update which_patch & target file for SLES10 latest kernel.
138 Description: Remove client patches from SLES 10 kernel.
139 Details : This causes SLES 10 clients to behave as patchless clients
140 even on a Lustre-patched (server) kernel.
142 Severity : enhancement
144 Description: self-adjustable client's lru lists
145 Details : use adaptive algorithm for managing client cached locks lru
146 lists according to current server load, other client's work
147 pattern, memory activities, etc. Both, server and client
148 side namespaces provide number of proc tunables for controlling
151 Severity : enhancement
153 Description: light-weight GSS support
154 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
155 and reduce performance overhead.
157 Severity : enhancement
159 Description: Linux keyring support
160 Details : Support using service of Linux keyring for Lustre GSS internal
161 context refresh/cache mechanism.
165 Description: Fix errors in lfs documentation
166 Details : Fixes man pages
170 Description: don't use GFP_* in generic Lustre code.
171 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
175 Description: obdclass is limited by single OBD_ALLOC(idarray)
176 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
180 Description: warnings with build patchless client with vanila 2.6.19 and up
181 Details : change old ctl_table style and replace ctl_table/ctl_table_header
182 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
186 Frequency : Only for SLES
187 Description: sanity_quota fail test_1
188 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
189 file, which makes TSTID[2] inunique.
193 Frequency : Only for SLES9
194 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
195 Details : Update target/ChangeLog/which_patch .
199 Frequency : Only for test_10 in sanity-quota.sh
200 Description: a bug in quota test code
201 Details : checking a test flag in a wrong place
205 Frequency : Only for test_13 in sanity-quota.sh
206 Description: a bug in quota test script
207 Details : shouldn't check overfull allocated quota
211 Description: jbd statistics
212 Details : Port older jbd statistics patch for sles10
214 Severity : enhancement
216 Description: Add printing inode info into message about error in writepage.
221 Description: reply_lock_interpret crash due to race with it and lock cancel.
222 Details : Do not replay locks that are being cancelled. Do not reference
223 locks by their address during replay, just by their handle.
227 Frequency : When flocks are used.
228 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
229 Details : Flock locks might destroy just granted lock if it could be merged
230 with another existing flock, this is done in completion handler,
231 so teach ldlm_cli_enquque_fini that this is a valid case for
237 Description: Oops in read and write path when failing to allocate lock.
238 Details : Check if lock allocation failed and return error back.
242 Description: lstripe command fails for valid OST index
243 Details : The stripe offset is compared to lov->desc.ld_tgt_count
244 instead of lov->desc.ld_active_tgt_count.
248 Description: sanity.sh failed test 103
249 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
254 Description: df doesn't work properly if diskfs blocksize != 4K
255 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
259 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
260 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
264 Description: obd_health_check_timeout too short
265 Details : set obd_health_check_timeout as 1.5x of obd_timeout
269 Description: llapi_file_create() does not allow some changes
270 Details : add llapi_file_open() that allows specifying the mode and
271 open flags, and also returns an open file handle.
275 Description: merge and cleanup kernel patches.
276 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
280 Description: Add journal checksum support.(Kernel part)
281 Details : The journal checksum feature adds two new flags i.e
282 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
283 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
284 indicates that the commit block contains the checksum for
285 the blocks described by the descriptor blocks. Now commit
286 record can be sent to disk without waiting for descriptor
287 blocks to be written to disk. This behavior is controlled
288 using JBD2_FEATURE_ASYNC_COMMIT flag.
292 Description: OSS needs mutliple precreate threads
293 Details : Add ability to start more than one create thread per OSS.
297 Description: Sanity.sh test_65k failed: setstripe should have succeeded
298 Details : execute lfs setstripe on client
302 Description: mds_obd_create error creating tmp object
303 Details : When the user sets quota on root, llog will be affected and can't
304 create files and write files.
307 Frequency : Always on ia64 patchless client, and possibly others.
309 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
310 Details : This allows the patchless client to be loaded on architectures
315 Description: RedHat Update kernel for RHEL5
316 Details : Add the kernel config file for RHEL5.
320 Description: RedHat Update kernel for RHEL5
321 Details : Modify the kernel config file more closer RHEL5.
325 Description: Build failure against Centos5 (RHEL5)
326 Details : Define PAGE_SIZE when it isn't present.
330 Description: test_7 on sanity-quota.sh can't be used on separate machines
331 Details : add facet to handle it
335 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
336 Details : with NFS, the anon dentry's parent was set to itself in
337 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
338 corresponding dentry other than use rec->ur_name.
340 Severity : enhancement
342 Description: lfs setstripe enhancement
343 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
347 Description: enable data checksumming by default
348 Details : enable checksum by default, allow --disable-checksum
349 configure option and "-o nochecksum" mount option. Checksums
350 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
354 Description: lustre support for RHEL5
355 Details : Add support for RHEL5.
359 Description: Client eviction due to failover config
360 Details : after a connection loss, the lustre client should attempt to
361 reconnect to the last active server first before trying the
362 other potential connections.
366 Description: when mds and osts use different quota unit(32bit and 64bit),
367 quota will be released repeatly.
368 Details : void sending multiple quota reqs to mds, which will keep the
369 status between the reqs.
373 Description: osts not allocated evenly to files
374 Details : change the condition to increase offset_idx
377 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
379 Description: Data corruption for OSTs that are formatted larger than 2TB
381 Details : When generating the bio request for lustre file writes the
382 sector number would overflow a temporary variable before being
383 used for the IO. The data reads correctly from Lustre (which
384 will overflow in a similar manner) but other file data or
385 filesystem metadata may be corrupted in some cases.
389 Description: Tune the kernel for good SCSI performance.
390 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
391 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
396 Description: rewrite ext2-derived code in obdclass/uuid.c
397 Details : rewrite inherited code (uuid parsing code from ext2 utils)
398 from scratch preserving functionality.
402 Description: "lfs find -obd UUID" prints directories
403 Details : "lfs find -obd UUID" will return all directory names instead
404 of just file names. It is incorrect because the directories
405 do not reside on the OSTs.
409 Description: use i_size_read and i_size_write in 2.6 port
410 Details : replace inode->i_size access with i_size_read/write()
413 Frequency : when removing large files
415 Description: scheduling issue during removal of large Lustre files
416 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
417 It causes scheduling issues when removing large files (17TB in the
421 Frequency : only with liblustre clients on XT3
423 Description: evictions taking too long
424 Details : allow llrd to evict clients directly on OSTs
427 Frequency : only on ppc
429 Description: /proc/fs/lustre/devices broken on ppc
430 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
431 We should make sure the type of 'index' is loff_t and then cast
432 explicitly as needed below. Do not assign an explicitly cast
436 Frequency : only for rhel5
438 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
439 Details : Modify the target file & which_kernel.
441 Severity : enhancement
443 Description: omit set fsid for export NFS
444 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
445 as FS_REQUIRES_DEV for avoid problems with generate fsid.
449 Frequency : Always, for kernels after 2.6.16
450 Description: Fix warning idr_remove called for id=.. which is not allocated.
451 Details : Last kernels save old s_dev before kill super and not allow
452 to restore from callback - restore it before call kill_anon_super.
456 Description: buffer overruns could theoretically occur
457 Details : llapi_semantic_traverse() modifies the "path" argument by
458 appending values to the end of the origin string, and a
459 overrun may occur. Adding buffer overrun check in liblustreapi.
463 Description: Fix error on 'ls .' at the top of the Lustre mount.
464 Details : Don't revalidate dentry if it is a root dentry.
468 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
469 Details : Modify vm-tunables-rhel4.patch.
473 Description: Kernel config for 2.6.18-vanilla.
474 Details : Modify targets/2.6-vanilla.target.in.
475 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
476 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
477 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
478 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
482 Description: improve handling recoverable errors
483 Details : if request processig with error which can be recoverable on server
484 request should be resend, otherwise page released from cache and
489 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
490 Details : Modify target file & which_patch.
491 A flaw was found in the IA32 system call emulation provided
492 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
493 value could be stored in the %RAX register, which could trigger an
494 out-of-bounds system call table access. An untrusted local user
495 could exploit this flaw to run code in the kernel
496 (ie a root privilege escalation). (CVE-2007-4573).
500 Description: O_DIRECT bypasses client statistics.
501 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
502 read_ahead_stats were updated. copied the stats section from
503 osc_send_oap_rpc() into async_internal().
507 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
511 Description: change order of libsysio includes
512 Details : '#include sysio.h' should always come before '#include xtio.h'
516 Description: handle "serious error: objid * already exists" more gracefully
517 Details : If LAST_ID value on disk is smaller than the objects existing in
518 the O/0/d* directories, it indicates disk corruption and causes an
519 LBUG(). If the object is 0-length, then we should use the existing
520 object. This will help to avoid a full fsck in most cases.
522 Severity : enhancement
524 Description: adapt the lustre_config script to support the upgrade case
525 Details : Add "-u" option for lustre_config script to support upgrading 1.4
526 server targets to 1.6 in parallel.
530 Description: To avoid grant space > avaible space when the disk is almost
531 full. Without this patch you might see the error "grant XXXX >
532 available" or some LBUG about grant, when the disk is almost
534 Details : In filter_check_grant, for non_grant cache write, we should
535 check the left space by if (*left > ungranted + bytes), instead
536 of (*left > ungranted), because only we are sure the left space
537 is enough for another "bytes", then the ungrant space should be
538 increase. In client, we should update cl_avail_grant only there
539 is OBD_MD_FLGRANT in the reply.
543 Description: Update RHEL 4 kernel to fix local root privilege escalation.
544 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
545 described in CVE-2007-4573. This problem could allow untrusted
546 local users to gain root access.
549 Frequency : when using O_DIRECT and quotas
551 Description: Incorrect file ownership on O_DIRECT output files
552 Details : block usage reported by 'lfs quota' does not take into account
553 files that have been written with O_DIRECT.
558 Description: touch file failed when fs is not full
559 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
560 otherwise we can get ENOSP while fs is not full.
564 Description: parallel lock callbacks
565 Details : Instead of sending blocking and completion callbacks as separated
566 requests, adding them to a set and sending in parallel.
569 Frequency : only for Cray XT3
570 Bugzilla : 12829/13455
571 Description: Changing primary group doesn't change the group lustre assigns to
573 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
574 overridden with the primary group provided by the group upcall,
575 whereas the supplied fsgid can be trusted if it is in the list of
576 supplementary groups returned by the group upcall.
578 Severity : enhancement
580 Description: Allow masking D_WARNING, D_ERROR messages from console
581 Details : Console messages can now be disabled via lnet.printk.
583 --------------------------------------------------------------------------------
585 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
587 * Support for kernels:
588 2.6.5-7.283 (SLES 9),
589 2.6.9-55.EL (RHEL 4),
590 2.6.16.46-0.14 (SLES 10),
591 2.6.18.8 vanilla (kernel.org)
592 * Client support for unpatched kernels:
593 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
594 2.6.16 - 2.6.22 vanilla (kernel.org)
595 * Due to recently discovered recovery problems, we do not recommend
596 using patchless RHEL 4 clients with this or any earlier release.
597 * Recommended e2fsprogs version: 1.39.cfs8
598 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
599 * Starting with this release, the ldiskfs backing filesystem required
600 by Lustre is now in its own package, lustre-ldiskfs. This package
601 should be installed. It is versioned separately from Lustre and
602 may be released separately in future.
606 Description: block reactivating mgc import until all deactivates complete
607 Details : Fix race when failing back MDT/MGS to itself (testing)
610 Frequency : at statup only
612 Description: mds_lov_synchronize race leads to various problems
613 Details : simultaneous MDT->OST connections at startup can cause the
614 sync to abort, leaving the OSC in a bad state.
616 Severity : enhancement
618 Description: add optional extra BUILD_VERSION info
619 Details : add a new environment variable (namely LUSTRE_VERS) which allows
620 to override the lustre version.
623 Frequency : 2.6.18 servers only
625 Description: ll_kern_mount() doesn't release the module reference
626 Details : The ldiskfs module reference count never drops down to 0
627 because ll_kern_mount() doesn't release the module reference.
632 Description: server LBUG when using old ost_num_threads parameter
633 Details : Accept the old ost_num_threads parameter but warn that it
634 is deprecated, and fix an off-by-one error that caused an LBUG.
639 Description: Transient SCSI error results in persistent IO issue
640 Details : iobuf->dr_error is not reinitialized to 0 between two
644 Frequency : sometimes when underlying device returns I/O errors
646 Description: OSTs not going read-only during write failures
647 Details : OSTs are not remounted read-only when the journal commit threads
648 get I/O errors because fsfilt_ext3 calls journal_start/stop()
649 instead of the ext3 wrappers.
653 Description: poor connect scaling with increasing client count
654 Details : Don't run filter_grant_sanity_check for more than 100 exports
655 to improve scaling for large numbers of clients.
658 Frequency : SLES10 only
660 Description: sanity-quota.sh quotacheck failed: rc = -22
661 Details : Quotas cannot be enabled on SLES10.
664 Frequency : liblustre clients only
666 Description: getdirentries does not give error when run on compute nodes
667 Details : getdirentries does not fail when the size specified as an argument
668 is too small to contain at least one entry
670 Severity : enhancement
672 Description: Add LNET router traceability for debug purposes
673 Details : If a checksum failure occurs with a router as part of the
674 IO path, the NID of the last router that forwarded the bulk data
675 is printed so it can be identified.
680 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
681 Details : Due to a race condition, liblustre clients were occasionally
684 Severity : enhancement
686 Description: lfs setstripe use optional parameters instead of postional
689 Severity : enhancement
691 Description: Nanosecond timestamp support for ldiskfs
692 Details : The on-disk ldiskfs filesystem has added support for nanosecond
693 resolution timestamps. There is not yet support for this at
694 the Lustre filesystem level.
697 Frequency : during server recovery
699 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
700 Details : request with rq_no_resend flag not awake l_wait_event if they get a
704 Frequency : nfs export on patchless client
706 Description: connectathon hang when test nfs export over patchless client
707 Details : Disconnected dentry cannot be found with lookup, so we do not need
708 to unhash it or make it invalid
711 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
712 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
713 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
718 Description: open req refcounting wrong on reconnect
719 Details : If reconnect happened between getting open reply from server and
720 call to mdc_set_replay_data in ll_file_open, we will schedule
721 replay for unreferenced request that we are about to free.
722 Subsequent close will crash in variety of ways.
723 Check that request is still eligible for replay in
724 mdc_set_replay_data().
729 Description: disable writes to filesystem when reading health_check file
730 Details : the default for reading the health_check proc file has changed
731 to NOT do a journal transaction and write to disk, because this
732 can cause reads of the /proc file to hang and block HA state
733 checking on a healthy but otherwise heavily loaded system. It
734 is possible to return to the previous behaviour during configure
735 with --enable-health-write.
737 Severity : enhancement
739 Description: 64-bit inode version
740 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
741 to the inode. This will be required for version-based recovery.
746 Description: MDS fails to start if a duplicate client export is detected
747 Details : in some rare cases it was possible for a client to connect to
748 an MDS multiple times. Upon recovery the MDS would detect this
749 and fail during startup. Handle this more gracefully.
751 Severity : enhancement
753 Description: Add -o localflock option to simulate old noflock
755 Details : This will achieve local-only flock/fcntl locks
761 Description: log_commit_thread vs filter_destroy race leads to crash
762 Details : Take import reference before releasing llog record semaphore
767 Description: Wrong request locking in request set processing
768 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
769 imp_delayed_list, in this place should be used imp_lock.
772 Frequency : when reconnection
774 Description: Grant Leak when osc reconnect to OST
775 Details : When osc reconnect ost, OST(filter) should check whether it
776 should grant more space to client by comparing fed_grant and
777 cl_avail_grant, and return the granted space to client instead
778 of "new granted" space, because client will call osc_init_grant
779 to update the client grant space info.
782 Frequency : when client reconnect to OST
784 Description: Grant Leak when osc do resend and replay bulk write
785 Details : When osc reconnect to OST, OST(filter)should clear grant info of
786 bulk write request, because the grant info while be sync between
787 OSC and OST when reconnect, and we should ignore the grant info
788 these of resend/replay write req.
793 Description: Grant space more than avaiable left space sometimes.
794 Details : When then OST is about to be full, if two bulk writing from
795 different clients came to OST. Accord the avaliable space of the
796 OST, the first req should be permitted, and the second one
797 should be denied by ENOSPC. But if the seconde arrived before
798 the first one is commited. The OST might wrongly permit second
799 writing, which will cause grant space > avaiable space.
802 Frequency : when client is evicted
804 Description: Grant might be wrongly erased when osc is evicted by OST
805 Details : when the import is evicted by server, it will fork another
806 thread ptlrpc_invalidate_import_thread to invalidate the
807 import, where the grant will be set to 0. While the original
808 thread will update the grant it got when connecting. So if
809 the former happened latter, the grant will be wrongly errased
810 because of this race.
815 Description: Checking Stale with correct fid
816 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
817 to check whether it is stale, because sometimes, we want the
818 enqueue happened anyway, and op_data.fid2 will not be initialized.
820 Severity : enhancement
822 Description: update patchless client
823 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
826 Frequency : only with 2.4 kernel
828 Description: random memory corruption
829 Details : size of struct ll_inode_info is to big for union inode.u and this
830 can be cause of random memory corruption.
835 Description: Memory leak in recovery
836 Details : Lov_mds_md was not free in an error handler in mds_create_object.
837 It should also check obd_fail before fsfilt_start, otherwise if
838 fsfilt_start return -EROFS,(failover mds during mds recovery).
839 then the req will return with repmsg->transno = 0 and rc = EROFS.
840 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
841 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
842 be freed no matter whether fsfilt_commit success or not.
845 Frequency : only with huge count clients
847 Description: Prevents from taking the superblock lock in llap_from_page for
849 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
850 for llap_from_page call in ll_removepage prevents from taking the
851 superblock lock for a soon died page.
856 Description: Not check open intent error before release open handle
857 Details : in some rare cases, the open intent error is not checked before
858 release open handle, which may cause
859 ASSERTION(open_req->rq_transno != 0), because it tries to release
860 the failed open handle.
865 Description: Set cat log bitmap only after create log success.
866 Details : in some rare cases, the cat log bitmap is set too early. and it
867 should be set only after create log success.
871 Description: Accessing a block bevice can re-enable I/O when Lustre is
872 tearing down a device.
873 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
878 Description: service threads may hog cpus when there are a lot of requests
879 Details : Insert cond_resched to give other threads a chance to use some CPU
884 Description: the cat log was not initialized in recovery
885 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
886 unlink log on mds will not be initialized until mds post
887 recovery. And also in mds post recovery, the unlink log will
888 initialization will be done asynchronausly, so there will be race
889 between add unlink log and unlink log initialization.
893 Description: brw_stats were being printed incorrectly
894 Details : brw_stats were being printed as log2 but all of them were not
895 recorded as log2. Also remove some code duplication arising from
896 filter_tally_{read,write}.
900 Frequency : rare, only in recovery.
901 Description: ASSERTION(req->rq_type != LI_POISON) failed
902 Details : imp_lock should be held while iterating over imp_sending_list for
903 prevent destroy request after get timeout in ptlrpc_queue_wait.
907 Description: replay-single.sh test 52 fails
908 Details : A lock's skiplist need to be cleanup when it being unlinked
909 from its resource list.
913 Description: Short directio read returns full requested size rather than
915 Details : Direct I/O operations should return actual amount of bytes
916 transferred rather than requested size.
918 Severity : enhancement
920 Description: metadata RPC reduction (e.g. for rm performance)
921 Details : decrease the amount of synchronous RPC between clients and servers
922 by canceling conflicing lock before the operation on the client
923 and packing thier handles into the main operation RPC to server.
925 Severity : enhancement
927 Description: Async OSC create to avoid the blocking unnecessarily.
928 Details : If a OST has no remain object, system will block on the creating
929 when need to create a new object on this OST. Now, ways use
930 pre-created objects when available, instead of blocking on an
931 empty osc while others are not empty. If we must block, we block
932 for the shortest possible period of time.
934 Severity : enhancement
936 Description: refine locking for avoid write wrong info into lov_objid file
937 Details : fix possible races with add new target and write/update data in
940 --------------------------------------------------------------------------------
942 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
947 Frequency : on some architectures
949 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
950 Details : Uninitialized flags sometimes cause configuration commands to
954 Frequency : patchless clients only
956 Description: missing __iget() symbol export
957 Details : The __iget() symbol export is missing. To avoid the need for
958 this on patchless clients the deathrow inode reaper is turned
959 off, and we depend on the VM to clean up old inodes. This
960 dependency was during via the fix for bug 12181.
962 --------------------------------------------------------------------------------
964 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
966 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
967 INTEROPERATE with older versions automatically. In many cases a
968 special upgrade step is needed. Please read the
969 user documentation before upgrading any part of a live system.
970 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
971 version will not interoperate with 1.6 betas before beta5 (1.5.95).
972 * WARNING: Lustre configuration and startup changes are required with
973 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
979 Severity : enhancement
981 Description: MountConf
982 Details : Lustre configuration is now managed via mkfs and mount
983 commands instead of lmc and lconf. New obd types (MGS, MGC)
984 are added for dynamic configuration management. See
985 https://mail.clusterfs.com/wikis/lustre/MountConf for
988 Severity : enhancement
990 Description: dynamic OST addition
991 Details : OSTs can now be added to a live filesystem
993 Severity : enhancement
995 Description: startup order invariance
996 Details : MDTs and OSTs can be started in any order. Clients only
997 require the MDT to complete startup.
999 Severity : enhancement
1001 Description: parallel, asynchronous orphan cleanup
1002 Details : orphan cleanup is now performed in separate threads for each
1003 OST, allowing parallel non-blocking operation.
1005 Severity : enhancement
1007 Description: optimized stripe assignment
1008 Details : stripe assignments are now made based on ost space available,
1009 ost previous usage, and OSS previous usage, in order to try
1010 to optimize storage space and networking resources.
1012 Severity : enhancement
1014 Description: Permanently set tunables
1015 Details : All writable /proc/fs/lustre tunables can now be permanently
1016 set on a per-server basis, at mkfs time or on a live system.
1018 Severity : enhancement
1020 Description: Lustre message v2
1021 Details : Add lustre message format v2.
1023 Severity : enhancement
1025 Description: client OST exclusion list
1026 Details : Clients can be started with a list of OSTs that should be
1027 declared "inactive" for known non-responsive OSTs.
1031 Description: SPEC SFS validation failure on NFS v2 over lustre.
1032 Details : Changes the blocksize for regular files to be 2x RPC size,
1033 and not depend on stripe size.
1035 Severity : enhancement
1037 Description: Multiple MD RPCs in flight.
1038 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1039 To avoid overly-overloading MDS, introduce a limit on number of
1040 MDS RPCs in flight for a single client and add /proc controls
1041 to adjust this limit.
1043 Severity : enhancement
1045 Description: client read/write statistics
1046 Details : Add client read/write call usage stats for performance
1047 analysis of user processes.
1048 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1049 file access. extents_stats shows chunk size distribution.
1050 extents_stats_per_process show chunk size distribution per
1053 Severity : enhancement
1055 Description: mds statistics
1056 Details : Add detailed mds operations statistics in
1057 /proc/fs/lustre/mds/*/stats.
1061 Description: Failure of copying files with lustre special EAs.
1062 Details : Client side always return success for setxattr call for lustre
1063 special xattr (currently only "trusted.lov").
1068 Description: Refcount LNET uuids
1069 Details : The global LNET uuid list grew linearly with every startup;
1070 refcount repeated list entries instead of always adding to
1073 Severity : enhancement
1075 Description: Dynamic service threads
1076 Details : Within a small range, start extra service threads
1077 automatically when the request queue builds up.
1080 Frequency : mixed-endian client/server environments
1082 Description: mixed-endian crashes
1083 Details : The new msg_v2 system had some failures in mixed-endian
1086 Severity : enhancement
1088 Description: Easy OST removal
1089 Details : OSTs can be permanently deactivated with e.g. 'lctl
1090 conf_param lustre-OST0001.osc.active=0'
1092 Severity : enhancement
1094 Description: MGS proc entries
1095 Details : Added basic proc entries for the MGS showing what filesystems
1098 Severity : enhancement
1100 Description: provide MGS failover
1101 Details : Added config lock reacquisition after MGS server failover.
1103 Severity : enhancement
1105 Description: add Linux 2.4 support
1106 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1110 Description: a large application tries to do I/O to the same resource and dies
1111 in the middle of it.
1112 Details : Check the req->rq_arrival time after the call to
1113 ost_brw_lock_get(), but before we do anything about
1114 processing it & sending the BULK transfer request. This
1115 should help move old stale pending locks off the queue as
1116 quickly as obd_timeout.
1119 Frequency : when an incorrect nid is specified during startup
1121 Description: ptlrpc connect to non-existant node causes kernel crash
1122 Details : LNET can't be re-entered from an event callback, which
1123 happened when we expire a message after the export has been
1124 cleaned up. Instead, hand the zombie cleanup off to another
1127 Severity : enhancement
1129 Description: plain/inodebits lock performance improvement
1130 Details : Grouping plain/inodebits in granted list by their request modes
1131 and bits policy, thus improving the performance of search through
1135 Frequency : only if OST filesystem is corrupted
1137 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1138 Details : for a short time RPCs with bulk IO are in the replay list,
1139 but replay of bulk IOs is unimplemented. If the OST filesystem
1140 is corrupted due to disk cache incoherency and then replay is
1141 started it is possible to trip an assertion. Avoid putting
1142 committed RPCs into the replay list at all to avoid this issue.
1145 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1147 Description: System hang on startup
1148 Details : This bug allowed the liblustre (e.g. catamount) client to
1149 return to the app before handling all startup RPCs. This
1150 could leave the node unresponsive to lustre network traffic
1151 and manifested as a server ptllnd timeout.
1153 Severity : enhancement
1155 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1156 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1157 boolean to control whether to print peer debug info when a
1158 client's RPC times out.
1161 Frequency : only for kernels with patches from Lustre below 1.4.3
1163 Description: Remove old rdonly API
1164 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1167 Frequency : only for devices with external journals
1169 Description: Set external device read-only also
1170 Details : During a commanded failover stop, we set the disk device
1171 read-only while the server shuts down. We now also set any
1172 external journal device read-only at the same time.
1175 Frequency : when upgrading from 1.4 while trying to change parameters
1177 Description: The wrong (new) MDC name was used when setting parameters for
1178 upgraded MDT's. Also allows changing of OSC (and MDC)
1179 parameters if --writeconf is specified at tunefs upgrade time.
1182 Frequency : when setting specific ost indicies
1184 Description: QOS code breaks on skipped indicies
1185 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1186 created with --index need not be sequential.
1188 Severity : enhancement
1190 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1191 Details : The uninit_groups feature works in conjunction with the kernel
1192 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1193 up the pass1 processing of e2fsck. This is a read-only feature
1194 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1195 mount filesystems that have had this feature enabled.
1197 Severity : enhancement
1199 Description: Improve multi-block allocation algorithm to avoid fragmentation
1200 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1201 allocation locality and avoid filesystem fragmentation.
1203 ------------------------------------------------------------------------------
1205 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1207 * Support for kernels:
1208 2.4.21-47.0.1.EL (RHEL 3)
1209 2.6.5-7.283 (SLES 9)
1210 2.6.9-42.0.10.EL (RHEL 4)
1211 2.6.12.6 vanilla (kernel.org)
1212 2.6.16.27-0.9 (SLES 10)
1213 * Recommended e2fsprogs version: 1.39.cfs5
1215 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1219 Frequency : occasional, depends on client load and configuration
1220 Bugzilla : 12181, 12203
1221 Description: data loss for recently-modified files
1223 Details : In some cases it is possible that recently written or created
1224 files may not be written to disk in a timely manner (this should
1225 normally be within 30s unless client IO load is very high).
1226 The problem appears as zero-length files or files that are a
1227 multiple of 1MB in size after a client crash or client eviction
1228 that are missing data at the end of the file.
1230 This problem is more likely to be hit on clients where files are
1231 repeatedly created and unlinked in the same directory, clients
1232 have a large amount of RAM, have many CPUs, the filesystem has
1233 many OSTs, the clients are rebooted frequently, and/or the files
1234 are not accessed by other nodes after being written.
1236 The presence of the problem can be detected by looking at
1237 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1238 smaller than the second (nr_unused) then dirty files will not
1239 be flushed automatically to disk. "sync; sleep 10" should be
1240 run several times on the node before unmounting it to update
1241 Lustre (this is also safe to run on nodes without this problem).
1243 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1244 that can cause this same problem, so customers using that kernel
1245 also need to update the kernel in addition to Lustre. In order
1246 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1248 It is normal that files written just before a client crash (less
1249 than 30s) may not yet have been flushed to disk, even for local
1253 Frequency : frequent on thin XT3 nodes
1255 Description: UUID collision on thin XT3 Linux nodes
1256 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1257 randomly, since we relied on an insufficiently-seeded PRNG.
1262 Description: OSS hangs after "All ost request buffers busy"
1263 Details : A deadlock between quota and journal operations caused OSS
1264 hangs after printing "All ost request buffers busy."
1267 Frequency : always on liblustre builds
1269 Description: Cleanup compiler warnings on liblustre
1272 Frequency : always on liblustre builds on XT3
1274 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1278 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1279 > 32000 subdirectories
1280 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1281 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1282 subdirectory count crosses 32000. This will aid e2fsck to
1283 correctly handle more than 32000 subdirectories.
1286 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1288 Description: System hang on startup
1289 Details : This bug allowed the liblustre (e.g. catamount) client to
1290 return to the app before handling all startup RPCs. This
1291 could leave the node unresponsive to lustre network traffic
1292 and manifested as a server ptllnd timeout.
1294 Severity : enhancement
1296 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1297 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1298 boolean to control whether to print peer debug info when a
1299 client's RPC times out.
1304 Description: make O_SYNC working on 2.6 kernels
1305 Details : 2.6 kernels use different method for mark pages for write,
1306 so need add a code to lustre for O_SYNC work.
1311 Description: Failure to close file and release space on NFS
1312 Details : Put inode details into lock acquired in ll_intent_file_open.
1313 Use mdc_intent_lock in ll_intent_open to properly
1314 detect all kind of errors unhandled by mdc_enqueue.
1319 Description: proc file read during shutdown sometimes raced obd removal,
1321 Details : Add lock to prevent obd access after proc file removal.
1324 Frequency : Only for files larger than 4GB on 32-bit clients.
1326 Description: improperly doing page alignment of locks
1327 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1328 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1333 Description: RPCs being resent when they shouldn't be
1334 Details : Some RPCs that should not be resent are being resent. This
1335 can cause inconsistencies in the RPC state machine. Do not
1336 resend such requests.
1339 Frequency : rare, only with NFS export
1341 Description: Crash on NFS re-export node
1342 Details : under very unusual load conditions an assertion is hit in
1343 ll_intent_file_open()
1346 Frequency : only if OST filesystem is corrupted
1348 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1349 Details : for a short time RPCs with bulk IO are in the replay list,
1350 but replay of bulk IOs is unimplemented. If the OST filesystem
1351 is corrupted due to disk cache incoherency and then replay is
1352 started it is possible to trip an assertion. Avoid putting
1353 committed RPCs into the replay list at all to avoid this issue.
1358 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1359 Details : Large single O_DIRECT read and write calls can fail to allocate
1360 a sufficiently large buffer to process the request. In case of
1361 allocation failure the allocation is retried with a smaller
1362 buffer and broken into smaller requests.
1364 Severity : enhancement
1366 Description: Add -o localflock option to simulate old noflock behaviour.
1367 Details : This will achieve local-only flock/fcntl locks coherentness.
1372 Description: versioning check is incomplete
1373 Details : Checking the version difference of client vs. server, report
1374 error if the gap is too big.
1379 Description: add support PG_writeback bit
1380 Details : add support for PG_writeback bit for Lustre, for more carefull
1381 work with page cache in 2.6 kernel. This also fix some deadlocks
1382 and remove hack for work O_SYNC with 2.6 kernel.
1384 Severity : enhancement
1386 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1387 Details : The uninit_groups feature works in conjunction with the kernel
1388 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1389 up the pass1 processing of e2fsck. This is a read-only feature
1390 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1391 mount filesystems that have had this feature enabled.
1393 Severity : enhancement
1395 Description: Improve multi-block allocation algorithm to avoid fragmentation
1396 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1397 allocation locality and avoid filesystem fragmentation.
1399 ------------------------------------------------------------------------------
1401 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1403 * Support for kernels:
1404 2.6.9-42.0.3.EL (RHEL 4)
1405 2.6.5-7.276 (SLES 9)
1406 2.4.21-47.0.1.EL (RHEL 3)
1407 2.6.12.6 vanilla (kernel.org)
1408 2.6.16.21-0.8 (SLES10)
1409 * Recommended e2fsprogs version: 1.39.cfs2-0
1411 * The backwards-compatible /proc/sys/portals symlink has been removed
1412 in this release. Before upgrading, please ensure that you change
1413 any configuration scripts or /etc/sysctl.conf files that access
1414 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1415 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1416 in advance of the upgrade on any system running Lustre 1.4.6 or
1417 newer, since /proc/sys/lnet was added in that version.
1418 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1422 Frequency : only when quota is used
1424 Description: avoid scanning export list for quota master
1425 Details : Change the algorithms to avoid scanning export list in order
1426 to improve the efficiency.
1429 Frequency : MDS failover only, very rarely
1431 Description: "went back in time" messages on mds failover
1432 Details : The greatest transno may be lost when the current operation
1433 finishes with an error (transno==0) and the client's last_rcvd
1434 record is over-written. Save the greatest transno in the
1435 mds_last_transno for this case.
1438 Frequency : always for specific kernels and striping counts
1440 Description: client may get "Matching packet too big" without ACL support
1441 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1442 when trying to access files in certain configurations. The
1443 clients should in fact be denied when mounting because they do
1444 not understand ACLs.
1447 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1449 Description: many clients connecting with IO in progress causes connect timeouts
1450 Details : Avoid synchronous journal commits to avoid delays caused by many
1451 clients connecting/disconnecting when bulk IO is in progress.
1452 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1453 OST_IO_PORTAL to avoid delays behind potentially many pending
1457 Frequency : occasionally with multiple writers to a single file
1459 Description: shared writes to file may result in wrong size reported by stat()
1460 Details : Allow growing of kms when extent lock is cancelled
1463 Frequency : always with random mmap IO to multi-striped file
1465 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1466 Details : Only if the hole is at the end of OST object so that kms is too
1467 small. Fix is to increase kms accordingly in ll_nopage.
1470 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1472 Description: writes to a missing object would leak memory on the OST
1473 Details : If there is an inconsistency between the MDS and OST filesystems,
1474 such that the MDS references an object that doesn't exist, writes
1475 to that object will leak memory due to incorrect cleanup in the
1476 error handling path, eventually running out of memory on the OST.
1481 Description: Creating too long symlink causes lustre errors
1482 Details : Check symlink and name lengths before sending requests to MDS.
1485 Frequency : only if flock is enabled (not on by default)
1487 Description: posix locks not released on fd closure on 2.6.9+
1488 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1489 kernels, this caused such locks not to be released on fd close and
1490 then assertions on fs unmount about still used locks.
1493 Frequency : MDS failover only, very rarely
1495 Description: clients may get ASSERTION(granted_lock != NULL)
1496 Details : When request was taking a long time, and a client was resending
1497 a getattr by name lock request. The were multiple lock requests
1498 with the same client lock handle and
1499 mds_getattr_name->fixup_handle_for_resent_request found one of the
1500 lock handles but later failed with ASSERTION(granted_lock != NULL).
1505 Description: handle->h_buffer_credits > 0, assertion failure
1506 Details : h_buffer_credits is zero after truncate, causing assertion
1507 failure. This patch extends the transaction or creates a new
1511 Frequency : NFS re-export or patchless client
1512 Bugzilla : 11179, 10796
1513 Description: Crash on NFS re-export node (__d_move)
1514 Details : We do not want to hash the dentry if we don't have a lock.
1515 But if this dentry is later used in d_move, we'd hit uninitialised
1516 list head d_hash, so we just do this to init d_hash field but
1517 leave dentry unhashed.
1520 Frequency : NFS re-export or patchless client
1522 Description: NFS exports has problem with symbolic link
1523 Details : lustre client didn't properly install dentry when re-exported
1524 to NFS or running patchless client.
1527 Frequency : NFS re-export or patchless client
1529 Description: Various nfs/patchless fixes.
1530 Details : fixes reuse disconected alias for lookup process - this fixes
1531 warning "find_exported_dentry: npd != pd",
1532 fix permission error with open files at nfs.
1533 fix apply umask when do revalidate.
1536 Frequency : occasional
1538 Description: Crash on NFS re-export node
1539 Details : calling clear_page() on the wrong pointer triggered oops in
1540 generic_mapping_read().
1543 Frequency : rarely, using O_DIRECT IO
1545 Description: unaligned directio crashes client with LASSERT
1546 Details : check for unaligned buffers before trying any requests.
1549 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1551 Description: stale data returned from RAID cache
1552 Details : If only a small amount of IO is done to the RAID device before
1553 reading it again it is possible to get stale data from the RAID
1554 cache instead of reading it from disk.
1557 Frequency : always for sles10 kernel
1559 Description: sles10 support
1560 Details : ll_follow_link: compile fixes and using of nd_set_link
1561 under newer kernels.
1564 Frequency : depends on arch, kernel and compiler version, always on sles10
1567 Description: recursive or deep enough symlinks cause stack overflow
1568 Details : getting rid of large stack-allocated variable in
1572 Frequency : depends on hardware
1574 Description: lustre write performance loss in the SLES10 kernel
1575 Details : the performance loss is caused by using of write barriers in the
1576 ext3 code. The SLES10 kernel turns barrier support on by
1577 default. The fix is to undo that change for ldiskfs.
1579 ------------------------------------------------------------------------------
1581 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1583 * Support for kernels:
1584 2.6.9-42.0.3EL (RHEL 4)
1585 2.6.5-7.276 (SLES 9)
1586 2.4.21-47.0.1.EL (RHEL 3)
1587 2.6.12.6 vanilla (kernel.org)
1591 Frequency : quota enabled and large files being deleted
1593 Description: releasing more than 4GB of quota at once hangs OST
1594 Details : If a user deletes more than 4GB of files on a single OST it
1595 will cause the OST to spin in an infinite loop. Release
1596 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1601 Description: statfs data retrieved from /proc may be stale or zero
1602 Details : When reading per-device statfs data from /proc, in the
1603 {kbytes,files}_{total,free,avail} files, it may appear
1604 as zero or be out of date.
1607 Frequency : systems with MD RAID1 external journal devices
1609 Description: lconf's call to blkid is confused by RAID1 journal devices
1610 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1611 of returning all block devices that match the journal UUID.
1614 Frequency : always, for aggregate stripe size over 4GB
1616 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1617 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1618 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1621 Frequency : always if "lfs find" used on a local file/directory
1623 Description: "lfs find" segfaults if used on a local file/directory
1624 Details : The case where a directory component was not specified wasn't
1625 handled correctly. Handle this properly.
1628 Frequency : always on ppc64
1630 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1631 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1632 perform endianness conversion.
1635 Frequency : rarely (truncate to non-zero file size after write under load)
1636 Bugzilla : 10730, 10687
1637 Description: Files padded with zeros to next 4K multiple
1638 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1639 it is possible that files that are truncated to a non-zero size
1640 immediately after being written are filled with zero bytes beyond
1641 the truncated size. No file data is lost.
1643 Severity : enhancement
1645 Description: Allow recovery/failover for liblustre clients.
1646 Details : liblustre clients were unaware of failover configurations until
1649 Severity : enhancement
1651 Description: user file locks should fail when not mounting with flock option
1652 Details : Set up an error-returning stub in ll_file_operations.lock field
1653 to prevent incorrect behaviour when client is mounted without
1654 flock option. Also, set up properly f_op->flock field for
1658 Frequency : always on ia64
1660 Description: "lfs df" loops on printing out MDS statfs information
1661 Details : The obd_ioctl_data was not initialized and in some systems
1662 this caused a failure during the ioctl that did not return
1663 an error. Initialize the struct and return an error on failure.
1666 Frequency : SLES 9 only
1668 Description: Error of copying files with lustre special EAs as root
1669 Details : Client side always return success for setxattr call for lustre
1670 special xattr (currently only "trusted.lov").
1673 Frequency : rarely on clusters with both ia64+i386 clients
1675 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1676 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1677 clients are doing concurrent writes to the same file it is
1678 possible that the ia64 clients may LASSERT because the OST
1679 extent locks are not PAGE_SIZE aligned. Ensure that grown
1680 locks are always aligned on the request boundary.
1683 Frequency : specific use, occasional
1685 Description: Overwriting in use executable truncates on-disk binary image
1686 Details : If one node attempts to overwrite an executable in use by
1687 another node, we now correctly return ETXTBSY instead of
1688 truncating the file.
1690 Severity : enhancement
1692 Description: Async OSC create to avoid the blocking unnecessarily.
1693 Details : If a OST has no remain object, system will block on the creating
1694 when need to create a new object on this OST. Now, ways use
1695 pre-created objects when available, instead of blocking on an
1696 empty osc while others are not empty. If we must block, we block
1697 for the shortest possible period of time.
1702 Description: chmod on Lustre root is propagated to other clients
1703 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1704 invalid by the follow_mount time.
1707 Frequency : liblustre clients only
1709 Description: Race in 'instant cancel' lock handling could lead to such locks
1710 never to be granted in case of SMP MDS
1711 Details : Do not destroy not yet granted but cbpending locks in
1715 Frequency : replay/resend of open
1717 Description: non null lock assetion failure in mds_intent_policy
1718 Details : Trying to replay/resend lockless open requests resulted in
1719 mds_open() returning 0 with no lock. Now it sets a flag if
1720 a lock is going to be returned.
1722 Severity : enhancement
1724 Description: Checksum enhancements
1725 Details : New checksum enhancements allow for resending RPCs that failed
1728 Severity : enhancement
1730 Description: Tunables on number of dirty pages in cacche
1731 Details : Allow to set limit on number of dirty pages cached.
1736 Description: client crash on unmount - lock still has references
1737 Details : In some error handling cases it was possible to leak a lock
1738 reference on a client while accessing a file. This was not
1739 harmful to the client during operation, but would cause the
1740 client to crash when the filesystem is unmounted.
1743 Frequency : specific case, rare
1745 Description: ETXTBSY on mds though file not in use by client
1746 Details : ETXTBSY is no longer incorrectly returned when attempting to
1747 chmod or chown a directory that the user previously tried to
1748 execute or a currently-executing binary.
1751 Frequency : extremely rare except on liblustre-based clients
1753 Description: Lustre space not freed when files are deleted
1754 Details : Clean up open-unlinked files after client eviction. Previously
1755 the unlink was skipped and the files remained as orphans.
1760 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1761 Details : In some cases it was possible to send a blocking callback to
1762 a client doing a glimpse, even though that client didn't get
1763 a lock granted. When the glimpse lock is cancelled on the OST
1764 the freed lock is left on the waiting list and corrupted the list.
1767 Frequency : all core dumps
1769 Description: Broke core dumps to lustre
1770 Details : Negative dentry may be unhashed if parent does not have UPDATE
1771 lock, but some callers, e.g. do_coredump, expect dentry to be
1772 hashed after successful create, hash it in ll_create_it.
1774 ------------------------------------------------------------------------------
1776 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1778 * Support for kernels:
1779 2.6.9-42.0.2.EL (RHEL 4)
1780 2.6.5-7.276 (SLES 9)
1781 2.4.21-40.EL (RHEL 3)
1782 2.6.12.6 vanilla (kernel.org)
1786 Frequency : always on RHEL 3
1788 Description: Number of open files grows over time
1789 Details : The number of open files grows over time, whether or not
1790 Lustre is started. This was due to a filp leak introduced
1791 by one of our kernel patches.
1793 ------------------------------------------------------------------------------
1795 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1797 * Support for kernels:
1798 2.6.9-42.EL (RHEL 4)
1799 2.6.5-7.276 (SLES 9)
1800 2.4.21-40.EL (RHEL 3)
1801 2.6.12.6 vanilla (kernel.org)
1806 Bugzilla : 5719, 9635, 9792, 9684
1807 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1808 Details : If a server is under heavy load and cannot reply to new
1809 connection requests before the client resends the (re)connect,
1810 the connection handling code can behave badly if two service
1811 threads are concurrently handing separate (re)connections from
1812 the same client. Add better locking to the connection handling
1813 code, and ensure that only a single connection will be processed
1814 for a given client UUID, even if the lock is dropped.
1816 Severity : enhancement
1818 Description: add TCP zero-copy support to kernel
1819 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1820 sends if the hardware supports scatter-gather and checksumming.
1821 This allows socklnd to do client-write and server-read more
1822 efficiently and reduce CPU utilization from skbuf copying.
1825 Frequency : only if NFS exporting from client
1827 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1828 Details : If setting attributes on a file created under NFS that had
1829 never been opened it would be possible to oops the client
1830 if the file had no objects.
1833 Frequency : always for liblustre
1835 Description: liblustre client does MDS+OSTs setattr RPC for each write
1836 Details : When doing a write from a liblustre client, the client
1837 incorrectly issued an RPC to the MDS and each OST the file was
1838 striped over in order to update the timestamps. When writing
1839 with small chunks and many clients this could overwhelm the MDS
1840 with RPCs. In all cases it would slow down the write because
1841 these RPCs are unnecessary.
1843 Severity : enhancement
1845 Description: allow number of MDS service threads to be changed at module load
1846 Details : It is now possible to change the number of MDS service threads
1847 running. Adding "options mds mds_num_threads={N}" to the MDS's
1848 /etc/modprobe.conf will set the number of threads for the next
1849 time Lustre is restarted (assuming the "mds" module is also
1850 reloaded at that time). The default number of threads will
1851 stay the same, 32 for most systems.
1856 Description: OST crash if filesystem is unformatted or corrupt
1857 Details : If an OST is started on a device that has never been formatted
1858 or if the filesystem is corrupt and cannot even mount then the
1859 error handling cleanup routines would dereference a NULL pointer.
1864 Description: NULL pointer deref in llap_from_page.
1865 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1866 incorrect) mapping if the page was truncated/reclaimed while it was
1867 searched for. Check for this condition and skip such pages when
1868 doing readahead. Introduce extra check to llap_from_page() to
1869 verify page->mapping->host is non-NULL (so page is not anonymous).
1872 Frequency : Sometimes when using sys_sendfile
1874 Description: "page not covered by a lock" warnings from ll_readpage
1875 Details : sendfile called ll_readpage without right page locks present.
1876 Now we introduced ll_file_sendfile that does necessary locking
1877 around call to generic_file_sendfile() much like we do in
1881 Frequency : with certain MDS communication failures at client mount time
1883 Description: NULL pointer deref after failed client mount
1884 Details : a client connection request may delayed by the network layer
1885 and not be sent until after the PTLRPC layer has timed out the
1886 request. If the client fails the mount immediately it will try
1887 to clean up before the network times out the request. Add a
1888 reference from the request import to the obd device and delay
1889 the cleanup until the network drops the request.
1892 Frequency : occasionally during client (re)connect
1894 Description: assertion failure during client (re)connect
1895 Details : processing a client connection request may be delayed by the
1896 client or server longer than the client connect timeout. This
1897 causes the client to resend the connection request. If the
1898 original connection request is replied in this interval, the
1899 client may trip an assertion failure in ptlrpc_connect_interpret()
1900 which thought it would be the only running connect process.
1903 Frequency : only with obd_echo servers and clients that are rebooted
1905 Description: kernel BUG accessing uninitialized data structure
1906 Details : When running an obd_echo server it did not start the ping_evictor
1907 thread, and when a client was evicted an uninitialized data
1908 structure was accessed. Start the ping_evictor in the RPC
1909 service startup instead of the OBD startup.
1911 Severity : enhancement
1912 Bugzilla : 10193 (patchless)
1913 Description: Remove dependency on various unexported kernel interfaces.
1914 Details : No longer need reparent_to_init, exit_mm, exit_files,
1915 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1918 Frequency : rare (only users of deprecated and unsupported LDAP config)
1920 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1921 Details : LDAP apparently contains 'lustreName' attributes instead of
1922 'name'. A simple remapping of the name is sufficient.
1925 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1927 Description: waiting_locks_callback trips kernel BUG if client is evicted
1928 Details : Running with the dump_on_timeout debug flag turned on makes
1929 it possible that the waiting_locks_callback() can try to dump
1930 the Lustre kernel debug logs from an interrupt handler. Defer
1931 this log dumping to the expired_lock_main() thread.
1933 Severity : enhancement
1935 Description: Support NFS exporting on 2.6 kernels.
1936 Details : Implement non-rawops metadata methods for NFS server to use without
1937 changing NFS server code.
1940 Frequency : very rare (synthetic metadata workload only)
1942 Description: two racing renames might cause an MDS thread to deadlock
1943 Details : Running the "racer" program may cause one MDS thread to rename
1944 a file from being the source of a rename to being the target of
1945 a rename at exactly the same time that another thread is doing
1946 so, and the second thread has already enqueued these locks after
1947 doing a lookup of the target and is trying to relock them in
1948 order. Ensure that we don't try to re-lock the same resource.
1951 Frequency : only very large systems with liblustre clients
1953 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1954 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1955 all OSTs in parallel. This allows the eviction of stale liblustre
1956 clients to proceed much faster than if they were done in series,
1957 and also offers similar improvements for other set_info RPCs.
1962 Description: excessive CPU usage during initial read phase on client
1963 Details : During the initial read phase on a client, it would agressively
1964 retry readahead on the file, consuming too much CPU and impacting
1965 performance (since 1.4.5.8). Improve the readahead algorithm
1966 to avoid this, and also improve some other common cases (read
1967 of small files in particular, where "small" is files smaller than
1968 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1973 Description: MDS crash when receiving packet with unknown intent.
1974 Details : Do not LBUG in unknown intent case, just return -EFAULT
1976 Severity : enhancement
1977 Bugzilla : 9293, 9385
1978 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1979 Details : Do not serialize getattr (non-intent version) and statfs.
1982 Frequency : occasional, when OST network is overloaded/intermittent
1984 Description: client evicted by OST after bulk IO timeout
1985 Details : If a client sends a bulk IO request (read or write) the OST
1986 may evict the client if it is unresposive to its data GET/PUT
1987 request. This is incorrect if the network is overloaded (takes
1988 too long to transfer the RPC data) or dropped the OST GET/PUT
1989 request. There is no need to evict the client at all, since
1990 the pinger and/or lock callbacks will handle this, and the
1991 client can restart the bulk request.
1994 Frequency : Always when mmapping file with no objects
1996 Description: client crashes when mmapping file with no objects
1997 Details : Check that we actually have objects in a file before doing any
1998 operations on objects in ll_vm_open, ll_vm_close and
2004 Description: Request leak when working with deleted CWD
2005 Details : Introduce advanced request refcount tracking for requests
2006 referenced from lustre intent.
2008 Severity : Enhancement
2010 Description: Cache open file handles on client.
2011 Details : MDS now will return special lock along with openhandle, if
2012 requested and client is allowed to hold openhandle, even if unused,
2013 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2014 closing files for every read/write openration.
2016 Severity : Enhancement
2018 Description: Cache open negative dentries on client when possible.
2019 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2020 negative dentries on lock revocation.
2025 Description: Remounting a client read-only wasn't possible with a zconf mount
2026 Details : It wasn't possible to remount a client read-only with llmount.
2028 Severity : enhancement
2029 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2030 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2031 MPICH is included to add Lustre-specific ADIO interfaces.
2032 This is based closely on the UFS ADIO layer and only differs
2033 in file creation, in order to allow the OST striping to be set.
2034 This is user-contributed code and not supported by CFS.
2039 Description: extended inode attributes (immutable, append-only) work improperly
2040 when 2.4 and 2.6 kernels are used on client/server or vice versa
2041 Details : Introduce kernel-independent values for these flags.
2043 Severity : enhancement
2046 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2047 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2048 granularity. Now, Fractional MB granularity is be supported,
2049 this is very useful for low memory system.
2051 Severity : enhancement
2053 Description: Getattr by fid
2054 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2055 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2056 have locks and avoids getattr rpc requests for it on every stat.
2059 Frequency : Always, for filesystems larger than 2TB
2061 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2062 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2063 filesystems larger than 2TB to be mounted with mballoc enabled.
2066 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2068 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2069 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2070 filesystems larger than 2TB is mounted then the kernel will
2071 silently corrupt the start of the filesystem. CONFIG_LBD is
2072 enabled for all CFS-supported kernels, but the possibility of
2073 this happening with a modified kernel config exists.
2075 Severity : enhancement
2077 Description: add client O_DIRECT support for 2.6 kernels
2078 Details : It is now possible to do O_DIRECT reads and writes to files
2079 in the Lustre client mountpoint on 2.6 kernel clients.
2081 Severity : enhancement
2083 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2084 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2085 parallel, not waiting for response from every OST before sending
2086 a rpc to the next OST.
2091 Description: setattr vs write race when updating file timestamps
2092 Details : Client processes that update a file timestamp into the past
2093 right after writing to the file (e.g. tar) it is possible that
2094 the updated file modification time can be reset to the current
2095 time due to a race between processing the setattr and write RPC.
2097 Severity : enhancement
2099 Description: Bring 'lfs find' closer in line with regular Linux find.
2100 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2101 -print0 options and obtains all the needed info through the lustre
2104 Severity : enhancement
2106 Description: support up to 1024 configured devices on one node
2107 Details : change obd_dev array from statically allocated to dynamically
2108 allocated structs as they are first used to reduce memory usage
2113 Description: Flush dirty partially truncated pages during truncate
2114 Details : Immediatelly flush partially truncated pages in filter_setattr,
2115 this way we completely avoid having any pages in page cache on OST
2116 and can retire ugly workarounds during writes to flush such pages.
2121 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2122 Details : protect inode from truncation within vfs_unlink() context
2123 just take a reference before calling vfs_unlink() and release it
2124 when parent's i_sem is free.
2129 Description: last_id value checked outside lock on OST caused LASSERT failure
2130 Details : If there were multiple MDS->OST object precreate requests in
2131 flight, it was possible that the OST's last object id was checked
2132 outside a lock and incorrectly tripped an assertion. Move checks
2133 inside locks, and discard old precreate requests.
2136 Frequency : always, if extents are used on OSTs
2138 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2139 Details : OSTs using the extents format would not zero the high 16 bits of
2140 the index physical block number. This is not a problem for any
2141 OST filesystems smaller than 16TB, and no kernels support ext3
2142 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2143 new/modified files) and can be fixed for existing filesystems
2144 with e2fsprogs-1.39-cfs1.
2149 Description: import connection selection may be incorrect if timer wraps
2150 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2151 import connections to be ignored if the 32-bit jiffies counter
2152 wraps. Use a 64-bit jiffies counter.
2155 Frequency : very large clusters immediately after boot
2157 Description: LNET request buffers exhausted under heavy short-term load
2158 Details : If a large number of client requests are generated on a service
2159 that has previously never seen so many requests it is possible
2160 that the request buffer growth cannot keep up with the spike in
2161 demand. Instead of dropping incoming requests, they are held in
2162 the LND until the RPC service can accept more requests.
2165 Frequency : Sometimes during replay
2167 Description: Assertion failure in ll_local_open after replay.
2168 Details : If replay happened on an open request reply before we were able
2169 to set replay handler, reply will become not swabbed tripping the
2170 assertion in ll_local_open. Now we set the handler right after
2171 recognising of open request
2174 Frequency : very rare
2176 Description: kernel reports "badness in vsnprintf"
2177 Details : Reading from the "recovery_status" /proc file in small chunks
2178 may cause a negative length in lprocfs_obd_rd_recovery_status()
2179 call to vsnprintf() (which is otherwise harmless). Exit early
2180 if there is no more space in the output buffer.
2182 Severity : enhancement
2184 Description: clear OBD RPC statistics by writing to them
2185 Details : It is now possible to clear the OBD RPC statistics by writing
2186 to the "stats" file.
2191 Description: Client mtime is not the same on different clients after utimes
2192 Details : In some cases, the client was using the utimes() syscall on
2193 a file cached on another node. The clients now validate the
2194 ctime from the MDS + OSTs to determine which one is right.
2199 Description: Inability to activate failout mode
2200 Details : lconf script incorrectly assumed that in pythong string's numeric
2201 value is used in comparisons.
2204 Frequency : always with multiple stripes per file
2206 Description: Inefficient object allocation for mutli-stripe files
2207 Details : When selecting which OSTs to stripe files over, for files with
2208 a stripe count that divides evenly into the number of OSTs,
2209 the MDS is always picking the same starting OST for each file.
2210 Return the OST selection heuristic to the original design.
2215 Description: mount failures may take full timeout to return an error
2216 Details : Under some heavy load conditions it is possible that a
2217 failed mount can wait for the full obd_timeout interval,
2218 possibly several minutes, before reporting an error.
2219 Instead return an error as soon as the status is known.
2221 Frequency : quota enabled and large files being deleted
2223 Description: releasing more than 4GB of quota at once hangs OST
2224 Details : If a user deletes more than 4GB of files on a single OST it
2225 will cause the OST to spin in an infinite loop. Release
2226 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2231 Description: statfs data retrieved from /proc may be stale or zero
2232 Details : When reading per-device statfs data from /proc, in the
2233 {kbytes,files}_{total,free,avail} files, it may appear
2234 as zero or be out of date.
2237 Frequency : systems with MD RAID1 external journal devices
2239 Description: lconf's call to blkid is confused by RAID1 journal devices
2240 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2241 of returning all block devices that match the journal UUID.
2244 Frequency : always, for aggregate stripe size over 4GB
2246 Description: assertion fails when trying to use 4GB stripe size
2247 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2248 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2251 Frequency : always on ppc64
2253 Description: the first write on an ext3 filesystem with mballoc got stuck
2254 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2255 perform endianness conversion.
2257 ------------------------------------------------------------------------------
2259 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2261 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2262 INTEROPERATE with older versions automatically. Please read the
2263 user documentation before upgrading any part of a live system.
2264 * WARNING: Lustre networking configuration changes are required with
2265 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2268 * Support for kernels:
2269 2.6.9-22.0.2.EL (RHEL 4)
2270 2.6.5-7.244 (SLES 9)
2271 2.6.12.6 vanilla (kernel.org)
2274 Severity : enhancement
2275 Bugzilla : 7981/8208
2276 Description: Introduced Lustre Networking (LNET)
2277 Details : LNET is new networking infrastructure for Lustre, it includes
2278 a reorganized network configuration mode (see the user
2279 documentation for full details) as well as support for routing
2280 between different network fabrics. Lustre Networking Devices
2281 (LNDs) for the supported network fabrics have also been
2282 created for this new infrastructure.
2284 Severity : enhancement
2285 Description: Introduced Access control lists
2286 Details : clients can set ACLs on files and directories in order to have
2287 more fine-grained permissions than the standard Unix UGO+RWX.
2288 The MDS must be started with the "-o acl" mount option.
2290 Severity : enhancement
2291 Description: Introduced filesystem quotas
2292 Details : Administrators may now establish per-user quotas on the
2295 Severity : enhancement
2297 Description: Configuration change for the XT3
2298 The PTLLND is now used to run Lustre over Portals on the XT3
2299 The configure option(s) --with-cray-portals are no longer used.
2300 Rather --with-portals=<path-to-portals-includes> is used to
2301 enable building on the XT3. In addition to enable XT3 specific
2302 features the option --enable-cray-xt3 must be used.
2307 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2308 Details : A race between allocating a new llcd and re-getting the llcd_lock
2309 allowed another thread to grab newly-allocated llcd.
2311 Severity : enhancement
2313 Description: 2.6 OST async journal commit and locking fix to improve performance
2314 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2315 2.6 kernels are now async as they already were in 2.4 kernels so
2316 that they can commit concurrently with the network bulk transfer.
2317 For block-allocated files the filter allocation semaphore is held
2318 to avoid filesystem fragmentation during allocation. BKL lock
2319 removed for 2.6 xattr operations where it is no longer needed.
2324 Description: lconf incorrectly determined whether two IP networks could talk
2325 Details : In some more complicated routing and multiple-network
2326 configurations, lconf will avoid trying to make a network
2327 connection to a disjoint part of the IP space. It was doing the
2328 math incorrectly for one set of cases.
2333 Description: Fix for potential infinite loop processing records in an llog.
2334 Details : If an llog record is corrupted/zeroed, it is possible to loop
2335 forever in llog_process(). Validate the llog record length
2336 and skip the remainder of the block on error.
2339 Frequency : occasional (liblustre only)
2341 Description: liblustre could not open files whose last component is a symlink
2342 Details : sysio_path_walk() would incorrectly pass the open intent to
2343 intermediate path components.
2346 Frequency : rare (liblustre only with non-standard tuning)
2347 Bugzilla : 7201 (7350)
2348 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2349 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2350 after it has been released, only do this for referenced locks
2352 Severity : enhancement
2354 Description: specifying an (invalid) directory default stripe_size of -1
2355 would reset the directory default striping
2356 Details : stripe_size -1 was used internally to signal directory stripe
2357 removal, now use "all default" to signal dir stripe removal
2358 as a directory striping of "all default" is not useful
2361 Frequency : common for large clusters running liblustre clients
2363 Description: doing an ls when liblustre clients are running is slow
2364 Details : sending a glimpse AST to a liblustre client waits for every AST
2365 to time out, as liblustre clients will not respond. Since they
2366 cannot cache data we refresh the OST lock LVB from disk instead.
2368 Severity : enhancement
2370 Description: doing an ls at the same time as file IO can be slow
2371 Details : enqueue and other "small" requests can be blocked behind many
2372 large IO requests. Create a new OST IO portal for non-IO
2373 requests so they can be processed faster.
2376 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2378 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2379 created file with mode 000
2380 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2381 allow an ftruncate() as a non-root user to a file with mode 000.
2382 Lustre now respects this flag to disable mode checks when
2383 truncating a file owned by the user
2386 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2389 Description: Revoking locks from clients that went dead or catatonic might take
2391 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2392 cancellation of such locks instant on servers without waiting for
2393 any reply from clients. Clients drops these locks when cancel
2394 notification from server is received without replying.
2397 Frequency : liblustre-only, when liblustre client dies or becomes busy
2399 Description: Doing ls on Linux clients can take a long time with active
2401 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2402 granting such locks to it in the first place if possible. Locks
2403 are taken by proxy on the OST during the read or write and
2404 dropped immediately afterward. Add connect flags handling, do
2405 not grant locks to liblustre clients for glimpse ASTs.
2407 Severity : enhancement
2409 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2410 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2411 sequential reads and behaves badly with applications doing random
2412 reads. Improve it by reading ahead at least read region, and
2413 avoiding excessive large RPC for small reads.
2415 Severity : enhancement
2417 Description: Creating more than 1000 files for a single job may cause a load
2418 imbalance on the OSTs if there are also a large number of OSTs.
2419 Details : qos_prep_create() uses an OST index reseed value that is an
2420 even multiple of the number of available OSTs so that if the
2421 reseed happens in the middle of the object allocation it will
2422 still utilize the OSTs as uniformly as possible.
2427 Description: OST or MDS may oops in ping_evictor_main()
2428 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2429 but doesn't restart at beginning of obd_exports_timed list
2432 Severity : enhancement
2434 Description: improve by-nid export eviction on the MDS and OST
2435 Details : allow multiple exports with the same NID to be evicted at one
2436 time without re-searching the exports list.
2439 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2441 Description: MDS may oops in groups_free()
2442 Details : in rare race conditions a newly allocated group_info struct is
2443 freed again, and this can be NULL. The 2.4 compatibility code
2444 for groups_free() checked for a NULL pointer, but 2.6 did not.
2447 Frequency : common for liblustre clients doing little filesystem IO
2448 Bugzilla : 9352, 7313
2449 Description: server may evict liblustre clients accessing contended locks
2450 Details : if a client is granted a lock or receives a completion AST
2451 with a blocking AST already set it would not reply to the AST
2452 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2455 Frequency : lfs setstripe, only systems with more than 160 OSTs
2457 Description: unable to set striping with a starting offset beyond OST 160
2458 Details : llapi_create_file() incorrectly limited the starting stripe
2459 index to the maximum single-file stripe count.
2462 Frequency : LDAP users only
2464 Description: lconf did not handle in-kernel recovery with LDAP properly
2465 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2467 Severity : enhancement
2469 Description: bind OST threads to NUMA nodes to improve performance
2470 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2471 node and do their allocations there to localize memory access
2473 Severity : enhancement
2475 Description: llmount can determine client NID directly from Myrinet (GM)
2476 Details : the client NID code from gmnalnid was moved directly into
2477 llmount, removing the need to use this or specifying the
2478 client NID explicitly when mounting GM clients with zeroconf
2481 Frequency : if client is started with down MDS
2483 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2484 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2485 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2486 ptlrpc_import_delayed_req() because we didn't check if the
2487 request was interrupted, and we also didn't break out of the
2488 event loop if there was no timeout
2493 Description: data loss during non-page-aligned writes to a single file from
2494 both multiple nodes and multiple threads on one node at same time
2495 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2496 inconsistency led to false short-reads, that were cached and later
2497 used by ->prepare_write() to fill in partially written page,
2498 leading to data loss.
2501 Frequency : always, if lconf --abort_recovery used
2503 Description: lconf --abort_recovery fails with 'Operation not supported'
2504 Details : lconf was attempting to abort recovery on the MDT device and not
2507 Severity : enhancement
2509 Description: remove cleanup logs
2510 Details : replace lconf-generated cleanup logs with lustre internal
2511 cleanup routines. Eliminates the need for client-cleanup and
2514 Severity : enhancement
2516 Description: add support for EAs (user and system) on lustre filesystems
2517 Details : it is now possible to store extended attributes in the Lustre
2518 client filesystem, and with the user_xattr mount option it
2519 is possible to allow users to store EAs on their files also
2521 Severity : enhancement
2523 Description: Add possibility (config option) to show minimal available OST free
2525 Details : When compiled with --enable-mindf configure option, statfs(2)
2526 (and so, df) will return least minimal free space available from
2527 all OSTs as amount of free space on FS, instead of summary of
2528 free spaces of all OSTs.
2530 Severity : enhancement
2532 Description: do not expand extent locks acquired on OST-side
2533 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2534 by server: they are not cached anyway.
2537 Frequency : when mmap is used/binaries executed from Lustre
2539 Description: Unmmap pages before throwing them away from read cache.
2540 Details : llap_shrink cache now attempts to unmap pages before discarding
2541 them (if unmapping failed - do not discard). SLES9 kernel has
2542 extra checks that trigger if this unmapping is not done first.
2547 Description: lconf didn't resolve symlinks before checking to see whether a
2548 given mountpoint was already in use
2551 Frequency : when migrating failover services
2552 Bugzilla : 6395, 9514
2553 Description: When migrating a subset of services from a node (e.g. failback
2554 from a failover service node) the remaining services would
2555 time out and evict clients.
2556 Details : lconf --force (implied by --failover) sets the global obd_timeout
2557 to 5 seconds in order to quickly disconnect, but this caused
2558 other RPCs to time out too quickly. Do not change the global
2559 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2561 Severity : enhancement
2562 Frequency : if MDS is started with down OST
2563 Bugzilla : 9439,5706
2564 Description: Allow startup/shutdown of an MDS without depending on the
2565 availability of the OSTs.
2566 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2567 Add appropriate locking and lov-osc refcounts for safe
2568 cleaning. Add osc abort_inflight calls in case the
2569 synchronize never started.
2572 Frequency : occasional (Cray XT3 only)
2574 Description: root not authorized to access files in CRAY_PORTALS environment
2575 Details : The client process capabilities were not honoured on the MDS in
2576 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2577 been accessed by an authorized user then root was able to access
2578 the file on the local client also. The root user capabilities
2579 are now allowed on the MDS, as this environment has secure UID.
2582 Frequency : occasional
2584 Description: ldiskfs "too long searching" message happens too often
2585 Details : A debugging message (otherwise harmless) prints too often on
2586 the OST console. This has been reduced to only happen when
2587 there are fragmentation problems on the filesystem.
2592 Description: Division by zero in statfs when all OSCs are inactive
2593 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2594 lov_statfs divides by value returned by lov_get_stripecnt().
2598 Bugzilla : 9489, 3273
2599 Description: First write from each client to each OST was only 4kB in size,
2600 to initialize client writeback cache, which caused sub-optimal
2601 RPCs and poor layout on disk for the first writen file.
2602 Details : Clients now request an initial cache grant at (re)connect time
2603 and so that they can start streaming writes to the cache right
2604 away and always do full-sized RPCs if there is enough data.
2605 If the OST is rebooted the client also re-establishes its grant
2606 so that client cached writes will be honoured under the grant.
2611 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2612 Details : Now I/O RPCs go to different portal number and (presumably) fast
2613 lock requests (and glimses) and other RPCs get their own service
2614 threads pool that should be able to service those RPCs
2617 Severity : enhancement
2619 Description: Ability to exchange lustre version between client and servers and
2620 issue warnings at client side if client is too old. Also for
2621 liblustre clients there is ability to refuse connection of too old
2623 Details : New 'version' field is added to connect data structure that is
2624 filled with version info. That info is later checked by server and
2628 Frequency : rare, liblustre only.
2629 Bugzilla : 9296, 9581
2630 Description: Two simultaneous writes from liblustre at offset within same page
2631 might proceed at the same time overwriting eachother with stale
2633 Details : I/O lock withing llu_file_prwv was released too early, before data
2634 actually was hitting the wire. Extended lock-holding time until
2635 server acknowledges receiving data.
2638 Frequency : extremely rare. Never observed in practice.
2640 Description: avoid generating lustre_handle cookie of 0.
2641 Details : class_handle_hash() generates handle cookies by incrementing
2642 global counter, and can hit 0 occasionaly (this is unlikely, but
2643 not impossible, because initial value of cookie counter is
2644 selected randonly). Value of 0 is used as a sentinel meaning
2645 "unassigned handle" --- avoid it. Also coalesce two critical
2646 sections in this function into one.
2648 Severity : enhancement
2650 Description: allow liblustre clients to delegate truncate locking to OST
2651 Details : To avoid overhead of locking, liblustre client instructs OST to
2652 take extent lock in ost_punch() on client's behalf. New connection
2653 flag is added to handle backward compatibility.
2655 Severity : enhancement
2656 Bugzilla : 4928, 7341, 9758
2657 Description: allow number of OST service threads to be specified
2658 Details : a module parameter allows the number of OST service threads
2659 to be specified via "options ost ost_num_threads={N}" in the
2660 OSS's /etc/modules.conf or /etc/modprobe.conf.
2664 Bugzilla : 6146, 9635, 9895
2665 Description: servers crash with bad pointer in target_handle_connect()
2666 Details : In rare cases when a client is reconnecting it was possible that
2667 the connection request was the last reference for that export.
2668 We would temporarily drop the export reference and get a new
2669 one, but this may have been the last reference and the export
2670 was just destroyed. Get new reference before dropping old one.
2672 Severity : enhancement
2673 Frequency : if client is started with failover MDS
2675 Description: Allow multiple MDS hostnames in the mount command
2676 Details : Try to read the configuration from all specified MDS
2677 hostnames during a client mount in case the "primary"
2680 Severity : enhancement
2682 Description: Stop sending data to evicted clients as soon as possible.
2683 Details : Check if the client we are about to send or are sending data to
2684 was evicted already. (Check is done every second of waiting,
2685 for which l_wait_event interface was extended to allow checking
2686 of exit condition at specified intervals).
2689 Frequency : rare, normally only when NFS exporting is done from client
2691 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2693 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2696 Frequency : always, if srand() is called before liblustre initialization
2698 Description: Liblustre uses system PRNG disturbing its usage by user application
2699 Details : Introduce internal to lustre fast and high-quality PRNG for
2700 lustre usage and make liblustre and some other places in generic
2701 lustre code to use it.
2703 Severity : enhancement
2704 Bugzilla : 9477, 9557, 9870
2705 Description: Verify that the MDS configuration logs are updated when xml is
2706 Details : Check if the .xml configuration logs are newer than the config
2707 logs stored on the MDS and report an error if this is the case.
2708 Request --write-conf, or allow starting with --old_conf.
2710 Severity : enhancement
2712 Description: Handle symlinks in the path when checking if Lustre is mounted.
2713 Details : Resolve intermediate symlinks when checking if a client has
2714 mounted a filesystem to avoid duplicate client mounts.
2719 Description: lconf can hit an error exception but still return success.
2720 Details : The lconf command catches the Command error exception at the top
2721 level script context and will exit with the associated exit
2722 status, but doesn't ensure that this exit status is non-zero.
2727 Description: failure of ptlrpc thread startup can cause oops
2728 Details : Starting a ptlrpc service thread can fail if there are a large
2729 number of threads or the server memory is very fragmented.
2730 Handle this without oopsing.
2733 Frequency : always, only if liblustre and non-default acceptor port was used
2735 Description: liblustre cannot connect to servers with non-default acceptor port
2736 Details : tcpnal_set_default_params() was not called and was therefore
2737 ignoring the environment varaible TCPNAL_PORT, as well as other
2738 TCPNAL_ environment variables
2743 Description: two objects could be created on the same OST for a single file
2744 Details : If an OST is down, in some cases it was possible to create two
2745 objects on a single OST for a single file. No problems other
2746 than potential performance impact and spurious error messages.
2750 Bugzilla : 5681, 9562
2751 Description: Client may oops in ll_unhash_aliases
2752 Details : Client dcache may become inconsistent in race condition.
2753 In some cases "getcwd" can fail if the current directory is
2759 Description: Inode refcounting problems in NFS export code
2760 Details : link_raw functions used to call d_instantiate without obtaining
2761 extra inode reference first.
2765 Bugzilla : 9942, 9903
2766 Description: Referencing freed requests leading to crash, memleaks with NFS.
2767 Details : We used to require that call to ll_revalidate_it was always
2768 followed by ll_lookup_it. Also with revalidate_special() it is
2769 possible to call ll_revalidate_it() twice for the same dentry
2770 even if first occurence returned success. This fix changes semantic
2771 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2772 reference on a request referred from the intent.
2773 ll_intent_release() then releases such a request.
2776 Frequency : rare, normally benchmark loads only
2778 Description: unlinked inodes were kept in memory on the client
2779 Details : If a client is repeatedly creating and unlinking files it
2780 can accumulate a lot of stale inodes in the inode slab cache.
2781 If there is no other client load running this can cause the
2782 client node to run out of memory. Instead flush old inodes
2783 from client cache that have the same inode number as a new inode.
2786 Frequency : SLES9 2.6.5 kernel and long filenames only
2787 Bugzilla : 9969, 10379
2788 Description: utime reports stale NFS file handle
2789 Details : SLES9 uses out-of-dentry names in some cases, which confused
2790 the lustre dentry revalidation. Change it to always use the
2794 Frequency : rare, unless heavy write-truncate concurrency is continuous
2795 Bugzilla : 4180, 6984, 7171, 9963, 9331
2796 Description: OST becomes very slow and/or deadlocked during object unlink
2797 Details : filter_destroy() was holding onto the parent directory lock
2798 while truncating+unlinking objects. For very large objects this
2799 may block other threads for a long time and slow overall OST
2800 responsiveness. It may also be possible to get a lock ordering
2801 deadlock in this case, or run out of journal credits because of
2802 the combined truncate+unlink. Solution is to do object truncate
2803 first in one transaction without parent lock, and then do the
2804 final unlink in a new transaction with the parent lock. This
2805 reduces the lock hold time dramatically.
2808 Frequency : rare, 2.4 kernels only
2810 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2811 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2812 before unmounting their filesystem in order to avoid deadlock.
2813 The kernel_locked() function in 2.4 kernels only checks whether
2814 the kernel lock is held, not whether it is this process that is
2815 holding it as 2.6 kernels do.
2820 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2821 Details : The client ptlrpc code may be trying to reconnect to a down
2822 server before a previous connection attempt has timed out.
2823 Increase the reconnect interval to be longer than the connection
2824 timeout interval to avoid sending duplicate connections to
2828 Frequency : echo_client brw_test command
2830 Description: fix echo_client to work with OST preallocated code
2831 Details : OST preallocation code (5137) didn't take echo_client IO path
2832 into account: echo_client calls filter methods outside of any
2833 OST thread and, hence, there is no per-thread preallocated
2834 pages and buffers to use. Solution: hijack pga pages for IO. As
2835 a byproduct, this avoids unnecessary data copying.
2839 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2840 Description: Client can oops in mdc_commit_close() after open replay
2841 Details : It was possible for the MDS to return an open request with no
2842 transaction number in mds_finish_transno() if the client was
2843 evicted, but without actually returning an error. Clients
2844 would later try to replay that open and may trip an assertion
2845 Simplify the client close codepath, and always return an error
2846 from the MDS in case the open is not successful.
2849 Frequency : rare, 2.6 OSTs only
2851 Description: OST may deadlock under high load on fragmented files
2852 Details : If there was a heavy load and highly-fragmented OST filesystems
2853 it was possible to have all the OST threads deadlock waiting on
2854 allocation of biovecs, because the biovecs were not released
2855 until the entire RPC IO was completed. Instead, release biovecs
2856 as soon as they are complete to ensure forward IO progress.
2858 Severity : enhancement
2860 Description: Support for specifying external journal device at mount
2861 Details : If an OST or MDS device is formatted with an external journal
2862 device, this device major/minor is stored in the ext3 superblock
2863 and may not be valid for failover. Allow detecting and
2864 specifying the external journal at mount time.
2869 Description: Mounting an MDS with pending unlinked files may cause oops
2870 Details : target_finish_recovery() calls mds_postrecov() which returned
2871 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2872 considers this an error and immediately begins cleaning up the
2873 lov, just after starting the mds_lov process
2875 Severity : enhancement
2877 Description: Implement 'lfs df' to report actual free space on per-OST basis
2878 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2879 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2880 usage of objects; '-h' to report in human readable format.
2882 ------------------------------------------------------------------------------
2884 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2891 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2892 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2893 at startup during orphan unlinking. Instead, skip buddy update
2894 before setup, as it will be regenerated after recovery is complete.
2899 Description: If an OST is inactive, its locks might reference stale inodes.
2900 Details : lov_change_cbdata() must iterate over all namespaces, even if
2901 they are inactive to clear inode references from the lock.
2903 Severity : enhancement
2904 Frequency : occasional, if non-standard max_dirty_mb used
2906 Description: Client will block write RPCs if not enough grant
2907 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2908 then the client will block writes while waiting for another RPC
2909 to complete instead of consuming its dirty limit. With change
2910 we get improved performance when max_dirty_mb is small.
2912 Severity : enhancement
2913 Bugzilla : 3389, 6253
2914 Description: Add support for supplementary groups on the MDS.
2915 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2916 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2917 lookups for user supplementary groups into a cache.
2921 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2922 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2923 for liblustre O_EXCL is enforced by MDS.
2927 Description: readdir never returns NULL in liblustre.
2928 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2929 dentry in struct dirent.
2933 Frequency : liblustre only, depends on application IO pattern
2934 Description: liblustre clients evicted if not contacting servers
2935 Details : Don't put liblustre clients into the ping_evictor list, so
2936 they will not be evicted by the pinger ever.
2938 Severity : enhancement
2940 Description: Add ability to evict clients by NID from MDS.
2941 Details : By echoing "nid:$NID" string into
2942 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2943 $NID would be instantly evicted from this MDS and from all active
2944 OSTs connected to it.
2948 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2949 Details : lookup_it_finish() used to query file size from OSTs that was not
2954 Description: service threads change working directory to that of init
2955 Details : Starting lustre service threads may pin the working directory
2956 of the parent thread, making that filesystem busy. Threads
2957 now change to the working directory of init to avoid this.
2961 Frequency : during shutdown only
2962 Description: shutdown with a failed MDS or OST can cause unmount to hang
2963 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2968 Frequency : 2.6 only
2969 Description: chmod/chown may include an extra supplementary group
2970 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2971 supplementary group and if none is specified this is used.
2974 Bugzilla : 5479 (6816)
2976 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2977 Details : If lookup is in progress on a file that is unlinked we might try
2978 to revalidate the inode and fail in revalidate after lookup is
2979 complete and ll_file_open() enqueues the open again but
2980 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2983 Frequency : always, if lconf --abort_recovery used
2985 Description: lconf --abort_recovery fails with 'Operation not supported'
2986 Details : lconf was attempting to abort recovery on the MDT device and not
2989 ------------------------------------------------------------------------------
2991 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2996 Frequency : rare (only unsupported configurations with a node running as an
2998 Bugzilla : 6514, 5137
2999 Description: Mounting a Lustre file system on a node running as an OST could
3001 Details : OSTs now preallocates memory needed to write out data at
3002 startup, instead of when needed, to avoid having to
3003 allocate memory in possibly low memory situations.
3004 Specifically, if the file system is mounted on on OST,
3005 memory pressure could force it to try to write out data,
3006 which it needed to allocate memory to do. Due to the low
3007 memory, it would be unable to do so and the node would
3008 become unresponsive.
3010 Severity : enhancement
3012 Description: Addition of lconf --service command line option
3013 Details : lconf now accepts a '--service <arg>' option, which is
3014 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3016 Severity : enhancement
3018 Description: Failover mode is now the default for OSTs.
3019 Details : By default, OSTs will now run in failover mode. To return to
3020 the old behaviour, add '--failout' to the lmc line for OSTs.
3022 Severity : enhancement
3024 Description: Health checks are now provided for MDS and OSTs
3025 Details : Additional detailed health check information on MSD and OSTs
3026 is now provided through the procfs health_check value.
3029 Frequency : occasional, depends on IO load
3031 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3032 after numerous create/delete cycles
3033 Details : The ext3 inode allocation policy would not allocate new inodes
3034 very well on the OSTs because there are no new directories
3035 being created. Instead we look for groups with free space if
3036 the parent directories are nearly full.
3041 Description: Network or server problems during mount may cause partially
3042 mounted clients instead of returning an error.
3043 Details : The config llog parsing code may overwrite the error return
3044 code during mount error handling, returning success instead
3050 Description: MDS can fail to allocate large reply buffers
3051 Details : After long uptimes the MDS can fail to allocate large reply
3052 buffers (e.g. zconf client mount config records) due to memory
3053 fragmentation or consumption by the buffer cache. Preallocate
3054 some large reply buffers so that these replies can be sent even
3055 under memory pressure.
3059 Frequency : rare (liblustre)
3060 Description: fsx running with liblustre complained that using truncate() to
3061 extend the file doesn't work. This patch corrects that issue.
3062 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3063 ATTR_SIZE and lsm use in llu_setattr_raw.
3067 Frequency : rare, only 2.6 kernels
3068 Description: Unusual file access patterns on the MDS may result in inode
3069 data being lost in very rare circumstances.
3070 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3071 code in the 2.6 kernel caused the inode and/or EA data not to
3072 be read from disk, causing single-file corruption.
3076 Frequency : rare, only 2.6 filesystems using extents
3077 Description: Heavy concurrent write and delete load may cause data corruption.
3078 Details : It was possible under high-load situations to have an extent
3079 metadata block in the block device cache from a just-unlinked
3080 file overwrite a newly-allocated data block. We now unmap any
3081 metadata buffers that alias just-allocated data blocks.
3085 Frequency : filesystems with default stripe_count larger than 77
3086 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3087 Details : lconf specifies an inode size of 4096 bytes when the default
3088 stripe_count is larger than 77. This conflicts with the default
3089 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3090 this case to avoid pinning too much memory for large EAs.
3092 ------------------------------------------------------------------------------
3094 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3099 Frequency : rare (extremely heavy IO load with hundreds of clients)
3101 Description: Client is evicted, gets IO error writing to file
3102 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3103 caused clients to be evicted for AST timeouts. The fixes in
3104 bug 5192 mean we no longer need to have such short AST timeouts
3105 so ldlm_timeout has been increased.
3108 Frequency : occasional during --force or --failover shutdown under load
3109 Bugzilla : 5949, 4834
3110 Description: Server oops/LBUG if stopped with --force or --failover under load
3111 Details : a collection of import/export refcount and cleanup ordering
3112 issues fixed for safer force cleanup
3115 Frequency : only filesystems larger than 120 OSTs
3116 Bugzilla : 5990, 6223
3117 Description: lfs getstripe would oops on a very large filesystem
3118 Details : lov_getconfig used kfree on vmalloc'd memory
3121 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3122 Bugzilla : 6242, 6243
3123 Description: reading from files that had been truncated to a non-zero size
3124 but never opened returned no data
3125 Details : ll_file_read() reads zeros from no-object files to EOF
3130 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3131 Details : There's little harm in aborting MDS/OSS recovery and letting it
3132 try again, so I removed the LASSERT and return an error instead.
3134 Severity : enhancement
3136 Description: New debugging infrastructure for tracking down data corruption
3137 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3138 (b) cover more of the client-side code path, and (c) try to narrow
3139 down where problems occurred
3143 Bugzilla : 3819, 4364, 4397, 6313
3144 Description: Racing close and eviction MDS could cause assertion in mds_close
3145 Details : It was possible to get multiple mfd references during close and
3146 client eviction, leading to one thread referencing a freed mfd.
3148 Severity: : enhancement
3149 Bugzilla : 3262, 6359
3150 Description: Attempts to reconnect to servers are now more aggressive.
3151 Details : This builds on the enhanced upcall-less recovery that was added
3152 in 1.4.2. When trying to reconnect to servers, clients will
3153 now try each server in the failover group every 10 seconds. By
3154 default, clients would previously try one server every 25 seconds.
3159 Description: After recovery, certain operations trigger a failed
3160 assertion on a client.
3161 Details : Failing over an mds, using lconf -d --failover, while a
3162 client was doing a readdir() call would cause the client to
3163 LBUG after recovery completed and the readdir() was resent.
3165 Severity : enhancement
3167 Description: Default groups are now added by lconf
3168 Details : You can now run lconf --group <servicename> without having to
3169 manually add groups with lmc.
3172 Frequency : occasional
3174 Description: Nodes with an elan id of 0 trigger a failed assertion
3177 Frequency : always when accessing e.g. tty/console device nodes
3179 Description: tty and some other devices nodes cannot be used on lustre
3180 Details : file's private_data field is used by device data and lustre
3181 values in there got lost. New field was added to struct file to
3182 store fs-specific private data.
3185 Frequency : when exporting Lustre via NFS
3187 Description: NFSD failed occasionally when looking up a path component
3188 Details : NFSD is looking up ".." which was broken in ext3 directories
3189 that had grown large enough to become hashed.
3192 Frequency : Clusters with multiple interfaces not on the same subnet
3194 Description: Nodes will repeatedly try to reconnect to an interface which it
3195 cannot reach and report an error to the log.
3196 Details : Extra peer list entries will be created by lconf with some peers
3197 unreachable. lconf now validates the peer before adding it.
3200 Frequency : Only if a default stripe is set on the filesystem root.
3202 Description: Setting a default stripe on the filesystem root prevented the
3203 filesystem from being remounted.
3204 Details : The client was sending extra request flags in the root getattr
3205 request and did not allocate a reply buffer for the dir EA.
3208 Frequency : occasional, higher if lots of files are accessed by one client
3209 Bugzilla : 6159, 6097
3210 Description: Client trips assertion regarding lsm mismatch/magic
3211 Details : While revalidating inodes the VFS looks up inodes with ifind()
3212 and in rare cases can find an inode that is being freed.
3213 The ll_test_inode() code will free the lsm during ifind()
3214 when it finds an existing inode and then the VFS later attaches
3215 this free lsm to a new inode.
3219 Bugzilla : 6422, 7030
3220 Description: MDS deadlock between mkdir and client eviction
3221 Details : Creating a new file via mkdir or mknod (starting a transaction
3222 and getting the ns lock) can deadlock with client eviction
3223 (gets ns lock and trying to finish a synchronous transaction).
3226 Frequency : occasional
3227 Description: While starting a server, the fsfilt_ext3 module could not be
3229 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3230 kernels. Previously, lconf would still use the ext3 name
3231 when trying to load modules. Now, it will correctly use
3232 ext3 on 2.4 and ldiskfs on 2.6.
3234 Severity : enhancement
3235 Description: The default stripe count has been changed to 1
3236 Details : The interpretation of the default stripe count (0, to lfs
3237 or lmc) has been changed to mean striping across a single
3238 OST, rather than all available. For general usage we have
3239 found a stripe count of 1 or 2 works best.
3241 Severity : enhancement
3242 Description: Add support for compiling against Cray portals.
3243 Details : Conditional compiling for some areas that are different
3247 Frequency : occasional
3248 Bugzilla : 6409, 6834
3249 Description: Creating files with an explicit stripe count may lead to
3250 a failed assertion on the MDS
3251 Details : If some OSTs are full or unavailable, creating files may
3252 trigger a failed assertion on the MDS. Now, Lustre will
3253 try to use other servers or return an error to the
3257 Frequency : occasional
3259 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3260 caused lock timeout during readahead (since 1.4.2).
3261 Details : Processes doing readahead might match a lock that hasn't been
3262 granted yet if there are overlapping and conflicting lock
3263 requests. The readahead process waits on ungranted lock
3264 (original lock is CBPENDING), while OST waits for that process
3265 to cancel CBPENDING read lock and eventually evicts client.
3267 Severity : enhancement
3269 Description: Initial enabling of flock support for clients
3270 Details : Implements fcntl advisory locking and file status functions.
3271 This feature is provided as an optional mount flag (default
3272 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3273 locking are implemented yet, and those that are are not guaranteed
3274 to be completely correct in production environments.
3275 mount -t lustre -o [flock|noflock] ...
3278 Frequency : occasional
3280 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3281 assertion in the ext3 JBD (journaling) layer.
3282 Details : The b_committed_data struct is protected by the big kernel lock
3283 in 2.4 kernels, serializing journal_commit_transaction() and
3284 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3285 access to this struct. In 2.6 kernels there is finer grained
3286 locking to improve SMP performance of the JBD layer.
3290 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3295 Description: merge and cleanup kernel patches.
3298 -----------------------------------------------------------------------------
3300 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3302 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3303 versions of Lustre. Please update all servers and clients to
3304 version 1.4.2 or later at the same time. You must also run
3305 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3306 to update the configuration logs.
3308 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3309 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3310 - swab llog records of type '0' so we get proper header size/idx (5861)
3311 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3312 - fix rename of one directory over another leaking an inode (5953)
3313 - avoid SetPageDirty on 2.6 (5981)
3314 - don't re-add just-being-destroyed locks to the waiting list (5653)
3315 - when creating new directories, inherit the parent's custom
3316 striping settings if present parent (3048)
3317 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3318 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3319 - don't hold client locks on temporary worklist from l_lru (5666)
3320 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3321 - automatically evict dead clients (5921)
3322 - Update file size properly in create+truncate+fstat case (6196)
3323 - Do not unhash mountpoint dentries, do not allow removal of
3325 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3326 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3327 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3328 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3329 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3330 - Correctly handle reads of files with no objects (6243)
3331 - lctl recover will also mark a device active if deactivate used (5933)
3333 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3334 - allow --write-conf on an MDS with different nettype than client (5619)
3335 - don't write config llogs to MDS for mounts not from that MDS (5617)
3336 - lconf should create multiple TCP connections from a client (5201)
3337 - init scripts are now turned off by default; run chkconfig --on
3338 lustre and chkconfig --on lustrefs to use them
3339 - upcalls are no longer needed for clients to recover to failover
3341 - add --abort-recovery option to lconf to abort recovery on device
3343 - add support for an arbitrary number of OSTs (3026)
3344 - Quota support protocol changes.
3345 - forward compatibility changes to wire structs (6007)
3346 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3347 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3348 - improved "lustre status" script
3349 - initialize blocksize for non-regular files (6062)
3350 - added --disable-server and --disable-client configure options (5782)
3351 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3352 - Vanilla 2.4.29 support
3353 - increase maximum number of obd devices to 520 (6242)
3354 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3355 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3356 SLES 9 2.6.5 kernels
3357 - limit stripes per file to 160 (the maximum EA size) (6093)
3359 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3362 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3363 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3364 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3365 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3366 - small ext3 extents cleanups and fixes (5733)
3367 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3368 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3369 - avoid needless client->OST connect, fix handle mismatch (5317)
3370 - fix DLM error path that led to out-of-sync client, long delays (5779)
3371 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3372 - fix several locking issues related to i_size (5492,5624,5654,5672)
3373 - don't move pending lock onto export if it is already evicted (5683)
3374 - fix kernel oops when creating .foo in unlinked directory (5548)
3375 - fix deadlock in obdfilter statistics vs. object create (5811)
3376 - use time_{before,after} to avoid timer jiffies wrap (5882)
3377 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3378 - Do not leak request if server was not able to process it (5154)
3379 - If mds_open unable to find parent dir, make that negative lookup(5154)
3380 - don't create new directories with extent-mapping (5909, 5936)
3382 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3383 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3384 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3385 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3386 - support for RHEL4 kernel on i686 (5773)
3387 - provide error messages when incompatible logs are encountered (5898)
3389 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3390 * version 1.4.0.10 (1.4.1 release candidate 1)
3392 - don't keep a lock reference when lock is not granted (4238)
3393 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3394 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3395 - drop import inflight refcount on signal_completed_replay error (5255)
3396 - unlock page after async write error during send (3677)
3397 - handle missing objects in filter_preprw_read properly (5265)
3398 - no transno return for symlink open, don't save no-trasno open (3440)
3399 - don't try to complete elan receive that already failed (4012)
3400 - free RPC server reply state on error (5406)
3401 - clean up thread from ptlrpc_start_thread() on error (5160)
3402 - readahead could read extra page into cache that wasn't ejected (5388)
3403 - prevent races in class_attach/setup/cleanup/detach (5260)
3404 - don't dereference de->d_inode after l_dput of de (5458)
3405 - use "int" for stripe value returned from lock_to_stripe (5544)
3406 - mballoc allocation and error-checking fixes in 2.6 (5504)
3407 - block device patches to fix I/O request sizes in 2.6 (5482)
3408 - look up hostnames for IB nals (5602)
3409 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3410 - don't start multiple acceptors for the same port (5277)
3411 - fix incorrect LASSERT in mds_getattr_name (5635)
3412 - export a proc file for general "ping" checking (5628)
3413 - fix "lfs check" to not block when the MDS is down (5628)
3415 - service request history (4965)
3416 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3417 - create an "evict_client" /proc entry on OSTs, like the MDS has
3418 - fix mount usage message, return errors per mount(8) (5168)
3419 - change grep [] to grep "[]" in tests so they work in more UMLs
3420 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3421 - remove ancient LOV_MAGIC_V0 EA support (5047)
3422 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3423 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3424 - allow readahead allocations to fail when low on memory (5383)
3425 - mmap locking landed again, after considerable improvement (2828)
3426 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3428 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3431 - send OST transaction number in read/write reply to free req (4966)
3432 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3433 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3434 - the watchdog thread now runs as interruptible (5246)
3435 - flock/lockf fixes (but it's still disabled, pending 5135)
3436 - don't use EXT3 constants in llite code (5094)
3437 - memory shortage at startup could cause assertion (5176)
3439 - reorganization of lov code
3440 - single portals codebase
3442 - add extents/mballoc support (5025)
3443 - direct I/O reads in the obdfilter (4048)
3444 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3446 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3449 - send OST transaction number in read/write reply to free req (4966)
3450 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3451 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3452 - the watchdog thread now runs as interruptible (5246)
3453 - handle missing objects in filter_preprw_read properly (5265)
3454 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3455 - drop import inflight refcount on signal_completed_replay error (5255)
3456 - unlock page after async write error during send (3677)
3457 - return original error code on reconstructed replies (3761)
3458 - no transno return for symlink open, don't save no-trasno open (3440)
3460 - add pid to ldlm debugging output (4922)
3461 - bump the watchdog timeouts -- we can't handle 30sec yet
3462 - extra debugging for orphan dentry/inode bug (5259)
3464 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3467 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3468 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3469 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3470 - fix NULL dereference in /proc/sys/portals/routes (4827)
3471 - allow failed mdc_close() operations to be interrupted (4561)
3472 - stop precreate on OST before MDS would time out on it (4778)
3473 - don't send partial-page writes before EOF from client (4410)
3474 - discard client grant for sub-page writes on large-page clients (4520)
3475 - don't free dentries not owned by NFS code, check generation (4806)
3476 - fix lsm leak if mds_create_objects() fails (4801)
3477 - limit debug_daemon file size, always print CERROR messages (4789)
3478 - use transno after validating reply (3892)
3479 - process timed out requests if import state changes (3754)
3480 - update mtime on OST during writes, return in glimpse (4829)
3481 - add mkfsoptions to LDAP (4679)
3482 - use ->max_readahead method instead of zapping global ra (5039)
3483 - don't interrupt __l_wait_event() during strace
3485 - add software watchdogs to catch hung threads quickly (4941)
3486 - make lustrefs init script start after nfs is mounted
3487 - fix CWARN/ERROR duplication (4930)
3488 - return async write errors to application if possible (2248)
3489 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3490 - print NAL number in %x format (4645)
3491 - update barely-supported suse-2.4.21-171 series (4842)
3492 - support for sles 9 %post scripts
3493 - support for building 2.6 kernel-source packages
3494 - support for sles km_* packages
3496 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3499 - ignore -ENOENT errors in osc_destroy (3639)
3500 - notify osc create thread that OSC is being cleaned up (4600)
3501 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3502 - reconstruct ost_handle() like mds_handle() (4657)
3503 - create a new thread to do import eviction to avoid deadlock (3969)
3504 - let lconf resolve symlinked-to devices (4629)
3505 - don't unlink "objects" from directory with default EA (4554)
3506 - hold socknal file ref over connect in case target is down (4394)
3507 - allow more than 32000 subdirectories in a single directory (3244)
3508 - fix blocks count for O_DIRECT writes (3751)
3509 - OST returns ENOSPC from object create when no space left (4539)
3510 - don't send truncate RPC if file size isn't changing (4410)
3511 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3512 - bind to privileged port in socknal and tcpnal (3689)
3514 - rate limit CERROR/CWARN console message to avoid overload (4519)
3515 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3516 - basic mmap support (3918)
3517 - kernel patch series update from b1_4 (4711)
3519 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3522 - avoid crash during MDS cleanup with OST shut down (2775)
3523 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3524 - don't use bad inodes on the MDS (3744)
3525 - dynamic object preallocation to improve recovery speed (4236)
3526 - don't hold spinlock over lock dumping or change debug flags (4401)
3527 - don't zero obd_dev when it is force cleaned (3651)
3528 - print grants to console if they go negative (4431)
3529 - "lctl deactivate" will stop automatic recovery attempts (3406)
3530 - look for existing locks in ldlm_handle_enqueue() (3764)
3531 - don't resolve lock handle twice in recovery avoiding race (4401)
3532 - revalidate should check working dir is a directory (4134)
3534 - don't always mark "slow" obdfilter messages as errors (4418)
3536 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3539 - don't close LustreDB during write_conf until it is done (3860)
3540 - fix typo in lconf for_each_profile (3821)
3541 - allow dumping logs from multiple threads at one time (3820)
3542 - don't allow multiple threads in OSC recovery (3812)
3543 - fix debug_size parameters (3864)
3544 - fix mds_postrecov to initialize import for llog ctxt (3121)
3545 - replace config semaphore with spinlock (3306)
3546 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3547 - don't allow enqueue to complete on a destroyed export (3822)
3548 - down write_lock before checking llog header bitmap (3825)
3549 - recover from lock replay timeout (3764)
3550 - up llog sem before sending rpc (3652)
3551 - reduce ns lock hold times when setting kms (3267)
3552 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3553 - fix NULL deref and obd_dev leak on setup error (3312)
3554 - replace some LBUG about llog ops with error handling (3841)
3555 - don't match INVALID dentries from d_lookup and spin (3784)
3556 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3557 - fix invalid assertion in ptlrpc_set_wait (3880)
3559 - add libwrap support for the TCP acceptor (3996)
3560 - add /proc/sys/portals/routes for non-root route listing (3994)
3561 - allow setting MDS UUID in .xml (2580)
3562 - print the stack of a process that LBUGs (4228)
3564 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3567 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3568 - make sure to unset replay flag from failed open requests (3440)
3569 - if default stripe count is 0, use OST count for inode size (3636)
3570 - update parent mtime/ctime on client for create/unlink (2611)
3571 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3572 - free recovery state on server during a forced cleanup (3571)
3573 - unregister_reply for resent reqs (3063)
3574 - loop back devices mounting and status check on 2.6 (3563)
3575 - fix resource-creation race that can provoke i_size == 0 (3513)
3576 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3577 - more debugging for page-accounting assertion (3746)
3578 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3579 - avoid deadlock after precreation failure (3758)
3580 - fix race and lock order deadlock in orphan handling (3450, 3750)
3581 - add validity checks when grabbing inodes from l_ast_data (3599)
3583 - add /proc/.../recovery_status to obdfilter (3428)
3584 - lightweight CDEBUG infrastructure, debug daemon (3668)
3585 - change default OSC RPC parameters to be better on small clusters
3586 - turn off OST read cache for files smaller than 32MB
3587 - install man pages and include them in rpms (3100)
3588 - add new init script for (un)mounting lustre filesystems (2593)
3589 - run chkconfig in %post for init scripts (3701)
3590 - drop scimac NAL (unmaintained)
3592 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3595 - clean kiobufs before and after use (3485)
3596 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3597 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3598 - change init script to fail more clearly if not run as root (1528)
3599 - allow clients to reconnect during replay (1742)
3600 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3601 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3602 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3603 - fix cond_resched (3554)
3604 - only evict unfinished clients after recovery (3515)
3605 - allow bulk resend, prevent data loss (3570)
3606 - dynamic ptlrpc request buffer allocation (2102)
3607 - don't allow unlinking open directory if it isn't empty (2904)
3608 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3609 - remove extraneous obd dereference causing LASSERT failure (3334)
3610 - don't use get_cycles() when creating temp. files on the mds (3156)
3611 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3612 - handle EEXIST for set-stripe, set proper directory name (3336)
3614 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3615 - fix ksocknal_fmb_callback() error messages (2918)
3617 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3620 - don't copy lvb into (possibly NULL) reply on error (2983)
3621 - don't deref dentry after dput, don't free lvb on error (2922)
3622 - use the kms to determine writeback rpc length (2947)
3623 - increment oti_logcookies when osc is inactive (2948)
3624 - update client's i_blocks count via lvb messages (2543)
3625 - handle intent open/close of special files properly (1557)
3626 - mount MDS with errors=remount-ro, like obdfilter (2009)
3627 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3628 - don't use cancelling-locks' kms values (2947)
3629 - use highest lock extent for kms, not last one (2925)
3630 - don't dereference ERR_PTR() dentry in error handling path (3107)
3631 - fix thread race in portals_debug_dumplog() (3122)
3632 - create lprocfs device entries at setup instead of at attach (1519)
3633 - common AST error handler, don't evict client on completion race (3145)
3634 - zero nameidata in detach_mnt in 2.6 (3118)
3635 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3636 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3637 - initialize RPC timeout timer earlier for 2.6 (3219)
3638 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3639 - print nal/nid for unknown nid (3258)
3640 - additional checks for oscc recovery before doing precreate (3284)
3641 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3642 - don't crash in mdc_close for bad permissions on open (3285)
3643 - zero i_rdev for non-device files (3147)
3644 - clear page->private before handing to FS, better assertion (3119)
3645 - tune the read pipeline (3236)
3646 - fix incorrect decref of invalidated dentry (2350)
3647 - provide read-ahead stats and refine rpc in flight stats (3328)
3648 - don't hold journal transaction open across create RPC (3313)
3649 - update atime on MDS at close time (3265)
3650 - close LDAP connection when recovering to avoid server load (3315)
3651 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3652 - don't leak open file on MDS after open resend (3325)
3653 - serialize filter_precreate and filter_destroy_precreated (3329)
3654 - loop device shouldn't call sync_dev() for nul device (3092)
3655 - clear page cache after eviction (2766)
3656 - resynchronize MDS->OST in background (2824)
3657 - refuse to mount the same filesystem twice on same mountpoint (3394)
3658 - allow llmount to create routes for mounting behind routers (3320)
3659 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3660 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3661 - fix rare problem with rename on htree directories (3417)
3663 - allow default OST striping configuration per directory (1414)
3664 - fix compilation for qswnal for 2.6 kernels (3125)
3665 - increase maximum number of MDS request buffers for large systems
3666 - change liblustreapi to be useful for external progs like lfsck (3098)
3667 - increase local configuration timeout for slow disks (3353)
3668 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3670 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3673 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3674 - don't overwrite extent policy data in reply if lock was blocked (2901)
3675 - drop filter export grants atomically with removal from device (2663)
3676 - del obd_self_export from work_list in class_disconnect_exports (2908)
3677 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3678 - swab reply message in mdc_close, other PPC fixes (2464)
3679 - fix destroying of named logs (2325)
3680 - overwrite old logs when running lconf --write_conf (2264)
3681 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3682 - fix race in target_handle_connect (2898)
3683 - mds_reint_create() should take same inode create lock (2926)
3684 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3685 - don't close files for self_export to avoid uninitialized obd (2936)
3686 - allow MDS with the same name as client node (2939)
3687 - hold dentry reference for closed log files for unlink (2325)
3688 - reserve space for all logs during transactions (2059)
3689 - don't evict page beyond end of stripe extent (2925)
3690 - don't oops on a deleted current working directory (2399)
3691 - handle hard links to targets without a parent properly (2517)
3692 - don't dereference NULL lock when racing during eviction (2867)
3693 - don't grow lock extents when lots of conflicting locks (2919)
3695 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3698 - account for cache space usage on clients to avoid data loss (974)
3699 - lfsck support in lustre kernel code (2349)
3700 - reduce journal credits needed for BRW writes (2370)
3701 - orphan handling to avoid losing space on client/server crashes
3702 - ptlrpcd can be blocked, stopping ALL progress (2477)
3703 - use lock value blocks to assist in proper KMS, faster stat (1021)
3704 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3705 - recovery for initial connections (2355)
3706 - fixes for mds_cleanup_orphans (1934)
3707 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3708 - block all file creations until orphan recovery completes (1901)
3709 - client remove rq_connection from request struct (2423)
3710 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3711 - recovery timer race (2670)
3712 - mdc_close recovey bug (2532)
3713 - ptlrpc cleanup bug (2710)
3714 - mds timeout on local locks (2588)
3715 - namespace lock held during RPCs (2431)
3716 - handle interrupted sync write properly (2503)
3717 - don't try to handle a message that hasn't been replied to (2699)
3718 - client assert failure during cleanup after abort recovery (2701)
3719 - leak mdc device after failed mount (2712)
3720 - ptlrpc_check_set allows timedout requests to complete (2714)
3721 - wait for inflight reqs when ptlrpcd finishes (2710)
3722 - make sure unregistered services are removed from the srv_list
3723 - reset bulk XID's when resending them (caught by 1138 test)
3724 - unregister_bulk after timeout
3725 - fix lconf error (2694)
3726 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3727 - readahead locks pages, leaves pending causing memory pressure (2673)
3728 - increase OST request buffers to 4096 on large machines (2729)
3729 - fix up permission of existing directories in simple_mkdir (2661)
3730 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3731 - don't assign transno to errored transactions (2742)
3732 - don't delete objects on OST if given a bogus objid from MDS (2751)
3733 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3734 - if rq_no_resend, then timeout request after recovery (2432)
3735 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3736 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3737 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3738 - MDS recovery shouldn't delete objects that it has given out (2730)
3739 - if enqueue arrives after completion, don't clobber LVB (2819)
3740 - don't unlock pages twice when trigger_group_io returns error (2814)
3741 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3742 - don't write pages to disk if there was an error (1450)
3743 - don't ping imports that have recovery disabled (2676)
3744 - take buffered bytes into account when balancing socknal conn (2817)
3745 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3746 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3747 - remove little-endian swabbing of llog records (1987)
3748 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3749 - retry reposting request buffers if they fail (1191)
3750 - grow extent at grant time to avoid granting a revoked lock (2809)
3751 - lock revoke doesn't evict page if covered by a second lock (2765)
3752 - disable VM readahead to avoid reading outside lock extents (2805)
3754 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3755 - updated kernel patches for hp-2.4.20 kernel (2681)
3757 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3760 - fix truncated write corruption (2366)
3761 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3763 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3764 - ensure len(uuid) < 37 in lmc (1171)
3765 - fix ia64 OOPS in llog_test (2255)
3766 - zero end of page at obdfilter for partial page writes (2648)
3767 - don't leave stale dentries around after renames (bug 2428)
3768 - fix timeouts when evicting a client with a single lock held (2642)
3769 - set deadline for the initial HELLO message to drain (2634)
3770 - print out dotted-quad IP addresses in the socknal (2302)
3772 - additional debugging for MDS client eviction problem (2443)
3773 - fix mkfsoptions support for osts (2603, 2604)
3775 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3778 - add series for the vanilla 2.6.0 kernel
3779 - add series for the vanilla 2.4.24 kernel
3780 - add series for a cray x86/64 UL kernel drop
3781 - fix xattr patches for the vanilla 2.4.19 series
3783 - generate true UUIDs in lmc (1171)
3784 - have portals stack dumping break in UML (2466)
3785 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3786 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3787 - wake callback waiting threads on client eviction (2460)
3788 - Add --ptldebug and --subsystem to lmc (1719)
3789 - update assertion to allow safe interrupt allocation
3790 - set rq_no_resend for cancel requests (2432)
3791 - recalculate ptlrpcd timeout after resend (2494)
3792 - call vfs_rmdir when removing pending directories (2368)
3793 - fix renaming a file to itself (2429)
3794 - lmc creates a default one-stripe lov (2454)
3795 - expand procfs space to handle large clusters (2326)
3796 - increase UML stack to avoid overflow
3797 - update lconf's list of debug and subsystem masks
3798 - fix lfs find --obd (2510)
3799 - /proc tunable for disabling filter read caching (2591)
3800 - stop rpm packages from altering slapd.conf (2301)
3801 - disable nagle in the socknal under 0conf (2578)
3802 - choose mds inode size based on stripe count (2572)
3803 - fix kernel-source rpm problems (2516)
3805 - add --disable-doc to avoid pdf generation (2421)
3806 - update documentation, tests, type-os, comments
3807 - avoid format warnings on ia64
3808 - remove the TOE NAL
3809 - tiny code cleanups by removing unused fields
3811 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3814 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3815 - remove the most blatant lies from BUILDING file (2371)
3816 - change default debug level to reasonable production setting
3817 - reduce client side cache size to reduce cache flush time
3818 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3819 - make TCP zerocopy and pinger support enabled by default (2476)
3820 - sync writes completed after process exits caused crashes (2319)
3821 - maintain correct mount count on the MDS (2356)
3822 - backout 1557, because 2316 wasn't really fixed
3823 - better file I/O statistics gathering in /proc
3824 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3825 - another kernel patch to fix zero-copy TCP function export
3826 - don't take duplicate lock when processing re-sent getattr (2420)
3827 - lctl uses obd_self_export instead of creating new conn (2353)
3828 - MDS/OST recovery case which requires object creation asserted (2425)
3829 - move lfs from /usr/sbin to /usr/bin in packages
3830 - fix race between mds_client_add and mds_client_free (2417)
3831 - use kmalloc instead of slabs in portals (2430)
3832 - don't create duplicate records when a failover MDS is present (2442)
3833 - remove unnecessary mount age check (2332)
3834 - don't remove directory inodes from locks prematurely (2451)
3835 - don't break if MDS service name is the same as hostname (2103)
3836 - fix races in client write RPC generation when cache full (2482)
3838 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3841 - remove now-unused request->rq_obd (278)
3842 - if an allocation fails, print out how much memory we've used (1933)
3843 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3844 - add the "configurable stack size" patch to most series files (1256)
3845 - ability to write large log records, for 100+ OST configs (2306)
3846 - fix NULL deref when filter_prep fails (2314)
3847 - fix operator precedence error in filter_sync
3848 - dynamic allocation of socknal TX descriptors (2315)
3849 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3850 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3851 - ensure that CWARN messages in Portals always get to the syslog
3852 - __init/__exit are not for prototype decls (ldlm_init/exit)
3853 - x86-64 compile warning fixes
3854 - fix gateway LMC keyword conflict (2318)
3855 - fix MDS lock inversions in getattr/reint paths (1844)
3856 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3857 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3858 - DLM race condition prevented some lock evictions (2328)
3859 - ENOMEM detection and retry on socknal sends (2230)
3860 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3861 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3862 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3863 - better error messages when a client is rejected during recovery (1505)
3864 - avoid cancelling locks which were never granted, after failure (2330)
3865 - fix i_sem/journal inversion in mds_client_add (2333)
3866 - fix truncate/getattr lock cycle deadlock (2334)
3867 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3868 - fix two rare exit paths which could leak an l_lock() ref (2321)
3869 - fencepost error in MDS/OST orphan recovery (2226)
3870 - make log record alignment 8 bytes (1988)
3871 - lstripe now fails when requested offset > ost_count (2237)
3872 - ensure that all kernel series have a complete list.h (1607)
3873 - fix crashes in special-file operations (2316)
3874 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3876 - allow configurable automake binary, for testing new versions
3877 - small update to the lfs documentation
3879 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3881 * fix negative export reference count in fsfilt_sync (2312)
3883 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3884 * release candidate 0.9.1
3886 - orphans are moved into the PENDING directory for possible recovery
3887 - replayed opens now open by fid for orphan/rename safety (1042)
3888 - last close of an orphan inode generates a transno (683)
3889 - chdir() and mount() now pin the directory entry (1020)
3890 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3891 - discard very old requests without processing them (1502)
3892 - remove some common, well-understood CERRORs (1505)
3893 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3894 - clear "grant" flags in OST replies until OST grant code lands (1644)
3895 - fix read performance by not clobbering i_blksize on client (1598)
3896 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3897 - make LRU size a /proc tunable, clears locks when reduced (707)
3898 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3899 - server locks take references on exports now (1558)
3900 - build fixes for 2.4.20-rh trees (1663)
3901 - return an error from lov_create if all OSCs are inactive (1751)
3902 - fix import levels when a reconnect happens without a timeout (1597)
3903 - exit early from mds_open if we get a lookup error (1749)
3904 - partial page read at EOF wouldn't wait for disk before sending (1642)
3905 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3906 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3907 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3908 - write/truncate lock inversion (1639)
3909 - Don't auto-load obdclass, portals modules during cleanup (1495)
3910 - fix timestamps from jumping to "now" (1763)
3911 - extra journal assertions (1648)
3912 - add an extra multiunlink test (1771)
3913 - fix read_record/write_record API (1776)
3914 - fix leak of offset_extent, possible incorrect i_size later (1772)
3915 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3916 - Debugging for the kqswnal_get_idle_tx problems (1820)
3917 - Allow recovery to be attempted multiple times (1536)
3918 - Write out MDS last_rcvd file after it is first created (1600)
3919 - Fix tx_descriptor leak in failed transmit situations (1827)
3920 - ext3 journaling fixes for assertion failure after IO error (1871)
3921 - class_export_put() on freed export after completion AST error (1896)
3922 - Fix revalidate looping in VFS (1322)
3923 - Don't access a freed export during MDS_REINT timeout (1521)
3924 - Add open-unlink recovery support on the MDS (1673,1764)
3925 - Return an error if no MDS data was read from last_rcvd (1946)
3926 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3927 - Don't setup a disk device that doesn't match exported UUID (317)
3928 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3929 - avoid committing NULL handle in force close
3930 - local.sh is now a one-stripe LOV configuration
3931 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3932 - fix ext3 htree duplicate directory entry corruption (1516)
3933 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3934 - update to new LOV EA format (2097)
3935 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3937 2003-06-15 Phil Schwan <phil@clusterfs.com>
3940 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3941 - per-import/export recovery handling (958, 931, 959)
3942 - multiple last-rcvd slots, for serving multiple FSes (949)
3943 - connections are again shared between multiple imp/exports (963, 964)
3944 - "umount -f" would hang if any requests needed to be sent (393, 978)
3945 - avoid pinning large req buffer by copying for queued messages (989)
3946 - add "uuid" to "lctl device" command to help upcalls (991)
3947 - "open" RPCs with transnos would confuse recovery counters (1037)
3948 - do proper endian conversion of all wire messages (288, 340, 891)
3949 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3950 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3951 - fix ptlrpc_connection leak in target_handle_connect (1174)
3952 - fix import refcounting bug in OST and MDS cleanup (1134)
3953 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3954 - fix very unlikely obd_types race condition (501)
3955 - remove osc_open hack for echo_client (1187)
3956 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3957 - a failure in read_inode2 leads to deadlock (1139)
3958 - cancel ack-locks as soon as transaction is committed (1072)
3959 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3960 - make sure to commitrw after any preprw to avoid deadlock (1162)
3961 - failing to execute a file in a lustre FS would lock inode (1203)
3962 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3963 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3964 - obd_uuid2tgt would walk off the end of the list (1255)
3965 - on IA64 the osc would give portals incorrect bulk size (1258)
3966 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3967 - fix lock inversion caused by new llite matching code (1282)
3968 - limit the number of dirty pages on a client to 10MB (1286)
3969 - timed out locks were not being corrected cancelled (1289)
3970 - fix O_DIRECT above 4GB on IA-32 (1292)
3971 * major user-visible changes
3972 - fail out/fail over policy now controlled by the upcall (993)
3974 - add OBD_PING to check server availability and failure (954)
3975 - lustre messages are now sent in sending host order (288, 340, 891)
3976 - add eadatalen to MDS getattr reply (340)
3977 - OST read replies may contain second buffer, with per-page status (593)
3979 2003-03-11 Phil Schwan <phil@clusterfs.com>
3982 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3983 - failed open()s could cause deadlock; fixed (867, 869)
3984 - stop cancelling OST locks when files are closed (481)
3985 - overlapping XID spaces caused network corruption (851, 853)
3986 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3987 - setattr_raw added, to do single-RPC, server-side setattrs
3988 - lmc/lconf syntax change for OST UUIDs
3989 - fix crashy race condition between ptlrpc_free_req and osc_close
3990 - don't use request in mdc_enqueue if we hit a timeout (889)
3991 - don't set the inode i_size for regular files from the MDS (896)
3992 - handle out of order completion AST (842)
3993 - don't LBUG if a lock request times out after receiving AST (913)
3994 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3995 - if a bad lock AST arrives, send an error instead of dropping entirely
3996 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3997 - fix leak in bulk IO when only partially completed (899, 900, 926)
3998 - fix O_DIRECT for ia64 (55)
3999 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4000 - C-z after timeout could hang a process forever; fixed (977)
4002 - client-side I/O cache (678, 924, 929, 941, 970)
4004 - READPAGE and SETATTRs which don't take server-side locks get
4007 2003-02-11 Phil Schwan <phil@clusterfs.com>
4010 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4011 - Fix fsfilt_extN_readpage() to read a full page of directory
4012 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4013 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4014 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4015 - Most intent operations were removed, in favour of a new RPC mode
4016 that does a single RPC to the server and bypasses most of the VFS
4017 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4018 - Aggressively cancel local locks on DLM servers
4019 - mds_reint_unlink sends EA to the client if it's the last nlink.
4020 client uses that EA to unlink OST objects.
4021 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4022 - recursive symlinks were fixed (439)
4023 - fixed NULL deref in DEBUG_REQ
4024 - filter_update_lastobjid no longer calls sync, which annoyed extN
4025 - fixed multi-client small-writes to a single file problem (445)
4026 - fixed mtime updates during file writes (607)
4027 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4028 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4029 - fixed Linux OST size reporting problem (444, 656)
4030 - OST now updates object mtime with writes or setattr (607, 619)
4031 - client verifies file size before zeroing page past EOF (445)
4032 - OST now writes last allocated objid to disk with allocation (108)
4033 - LOV on echo now works (409)
4035 - mds_reint_unlink sends a new buffer, with the EA included. this
4036 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4037 set if a regular file was being unlinked, and it was the last link
4038 - use PtlGet from the target for bulk writes (315)
4039 - OST now updates object mtime with writes or setattr (607, 619)
4040 - LDLM now has a grant-time callback to revalidate locked items, if
4042 - Many MDS operations were reorganized to combat race conditions
4044 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4045 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4047 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4050 - Fully reactivate OST imports after reconnection (512, others)
4051 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4052 - More graceful error handling for truncating on dead OST (515)
4053 - Don't error out unless we're actually accessing dead stripes (474)
4054 - Fix garbage sizes when stripes are missing (410)
4055 - LRU counters were broken, causing constant lock purge (433, 432)
4056 - garbage on read from stripes with failed OSTs (441)
4057 - mark OSCs as active before reconnecting during recovery (438)
4058 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4059 - lfind did not preserve OST order in output (443)
4060 - symlinks cause hung clients, incorrect data (439)
4061 - stop dereferencing request after dropping refcount (457)
4062 - don't LASSERT(spin_is_locked) on non-SMP (455)
4063 - fixes for many rename() bugs
4064 - fstat didn't correctly synchronize attributes (399)
4065 - server must handle lock cancellation during blocking AST prep (487)
4066 - bulk descriptors were free()d too soon (511)
4067 - fix paths in lconf, which would load incorrect modules (451, 507)
4068 - fix confusing lconf 'host not found' error message (386)
4069 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4070 - fix race condition in mdc_blocking_ast() for inode access (526)
4071 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4072 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4073 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4074 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4075 - fix lstripe in several ways (kernel side) (527)
4076 - fix request leak in ldlm_cli_enqueue (262)
4077 - incorrect OSC was marked inactive after OST failure
4078 - call mds_fs_cleanup before unmounting filesystem (524)
4079 - fix races between taking ns_lock and ldlm_lock_change_resource
4080 - fix races updating LOV export open file list
4081 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4082 - fix recovery NULL deref in ldlm_cli_cancel_unused
4083 - fix some DLM races by using new hash table for lock handles (419)
4084 - permit the client to specify desired inodes, at replay
4085 - duplicate requests when we queue them for replay reintegration
4086 - fix last_rcvd offset calculation
4087 - sync after each recovered transaction, so we always make progress
4088 - never, not always, ERESTART requests without transnos
4089 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4091 - skip replay if the MDS says that the client is already connected
4092 - don't check for a recovery-enabled export to match lctl's UUID
4093 - don't INC_USE_COUNT for phantom exports
4094 - don't crash when cleaning up phantom exports (567)
4095 - don't double-finish or set replay data for errored mdc_open requests
4096 - abort requests when they time out, so we don't get old replies
4097 - send/receive replies for AST messages again
4098 - if the client says that it doesn't have the lock, cancel it on the
4100 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4101 mark it as destroyed, it will all work out when decref is called
4102 - fix module use counts (22, 581)
4104 - ASTs now expect a reply (server cancels lock on error reply)
4106 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4109 - fix many simultaneous client startup (392)
4110 - fix dentry->d_it clobbering
4111 - credentials weren't being shipped for readdir/getattr operations
4112 - remove invalid assertions triggered during some concurrent MD
4114 - proper Lustre versions added (336, 389)
4115 - fix memory leak for create error case (398)
4116 - fix LOV locking bug that would get cli/srv out of sync
4117 - fix echo client over LOV (409)
4118 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4119 - fix double-O_EXCL intent crash (424)
4120 - avoid sending multiple lock CANCELs (352)
4122 - MDS can do multi-client recovery (modulo bugs in new code)
4124 - many updates, edits, cleanups
4126 2002-11-18 Phil Schwan <phil@clusterfs.com>
4129 - fix null d_it dereference (346)
4130 - fix full OST/dbench hang (333)
4131 - fix permission problem with file removal (286)
4132 - fix removal of OSCs from LOV when they fail
4133 - fix NULL deref during bulk timeout (214)
4134 - fix problems related to multiple filesystems on one MDS (241)
4135 - fixed serious subtle metadata locking bugs
4136 - free locks on clients when inodes are removed due to memory
4138 - fix inode pointer in lock data (285)
4139 - partial support for multiple MDS on a single host (241)
4140 - data locks weren't cancelled at clear_inode time (290, 311)
4141 - intent locks could lead to unbounded lock growth (205)
4142 - added a maximum lock count, an LRU list, and a flusher
4143 - fix multiple rename (365)
4144 - properly abstracted the echo client
4145 - OSC locked 1 byte too many; fixed
4146 - rewrote brw callback code:
4147 - fixed recovery bugs related to LOVs (306)
4148 - fixed too-many-pages-in-one-write crash (191)
4149 - fixed (again) crash in sync_io_timeout (214)
4150 - probably fixed callback-related race (385)
4152 - Add capability to MDS protocol
4153 - LDLM cancellations and callbacks on different portals
4155 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4158 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4159 - defer open object destruction to close time (601981, 138)
4160 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4161 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4162 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4163 - don't call obd_setattr in ll_file_release for destroyed objects
4164 * protocol change to lustre_msg: move |version| and add |flags|
4165 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4166 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4167 * added replay of create, unlink, link and rename operations during
4168 MDS failover; recovery should be much more robust now
4169 * remove failed OSCs from LOVs (only lov_create uses this so far)
4170 * the lustre-HOWTO was brought (more) up to date (582544)
4172 2002-10-23 Phil Schwan <phil@clusterfs.com>
4175 - in-use dentries weren't being reused properly (617851)
4176 - prevent multiple LDLM setup (599178)
4177 - fix LOV size calculations for truncate (617853)
4178 - fix client handling of MDS intent errors (POSIX)
4179 - fix permission bug in lovstripe.c test (624321)
4180 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4181 - truncate past end of file could corrupt data
4182 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4183 - a race in recovery could return ETIMEDOUT to apps (623947)
4184 - building outside the source directory was fixed
4185 * the lustre-HOWTO was brought (more) up to date (582544)
4186 * major progress was made on recovery functionality
4188 2002-10-10 Phil Schwan <phil@clusterfs.com>
4191 - recovery deadlock fix
4192 - rm -rf causes LBUG fix (617817)
4193 - file open by multiple tasks fix (618962)
4194 - directory permissions bugs (602707 and 620007)
4195 - journal_stop fixed with locking (611313)
4196 - O_APPEND failures resolved (618273, perhaps 614459)
4197 - lconf PATH fix (619770)
4198 - IA64 build fix (621450)
4199 - RPC buffer sizes scale with amount of memory
4201 2002-10-01 Phil Schwan <phil@clusterfs.com>
4204 - locks would be cancelled without throwing away data pages,
4205 resulting in inconsistent data (605627)
4206 - inode attributes were not always being refreshed (605627, 612449)
4207 - lconf now continues to cleanup after lctl reports an error
4208 - MDS now enforces user permissions (602707)
4209 - lprocfs cleanup fixed, but not yet enabled (614157)
4210 - fixed infinite server hang, should a client not respond to an AST
4211 - avoid going into recovery if user calls readlink() with a buffer
4212 that's too small (613941)
4213 - AST RPCs no longer require replies (614867) -- this may be changed
4214 - don't crash server if client sends an IOV that's too big (611336)
4215 - fixed lock conversion deadlock (611892)
4216 - fixed the following of symlinks (614622)
4217 * recovery: the server can remove locks from a client that dies, other
4218 clients can make progress
4219 * more extN patch fixes
4220 * compile-time configurable ptlrpc buffer allocations
4222 - collaborative read cache document
4223 - Lustre Lite Performance CDR document-in-progress
4225 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4228 - fix typo in patch-2.4.18
4230 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4233 - clear ptlrpc request each time in handle_incoming_request()
4234 - unlink of files now destroys the object on the OST
4236 2002-09-19 Peter Braam <braam@clusterfs.com>
4238 * add hard link support
4239 * change obdfile creation method
4240 * kernel patch changed
4242 2002-09-19 Peter Braam <braam@clusterfs.com>
4245 - stack overflow bug in extN fixed
4247 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4249 * documentation updates
4250 - add man pages for config tools
4251 - update tests/README to describe testing with new config tools
4252 - finish metadata API descriptions
4253 * bug fixes and cleanups
4254 - statfs workaround for 16TB limit
4255 - LOV stripe allocation improved, can stripe on subset of OSTs
4256 - LOV file size/IO offset was wrong for files > 4GB in size
4257 - object EA data was being dropped, caused files to be unreadable
4258 - memory overflow with non-LOV OST caused memory corruption
4259 - fixed regression tests to work with new config tools, obdfilter
4260 - fixed bug when directory size became larger than 1 block
4261 - fixed bug (for single client case) when PWD was deleted
4262 - invalidate local directory pages when doing intent-based ops
4263 - avoid LDLM oops when lock callback contained bad data
4265 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4267 * documentation updates
4268 * bug fixes and cleanups
4269 - configuration tools
4272 - 64-bit compile warnings
4273 - 64-bit internal statfs data
4275 * test_brw on persistent OST devices
4277 * lprocfs (disabled)
4279 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4281 * documentation updates
4282 * bug fixes and cleanups
4283 * configuration tools
4285 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4288 * many small fixes to 0_5_4
4289 * io/network handling
4290 * thinkos in MDS operations
4292 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4295 * crucial basic fixes to 0.5.3
4296 * IOR, Iozone work over Elan
4299 2002-08-07 Phil Schwan <phil@clusterfs.com>
4300 * version 0_5_3, our first alpha
4301 * we use the new Portals iovs
4302 * documentation updates
4303 * bug fixes and cleanups
4304 * small changes in the DLM wire protocol
4306 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4307 * version 0_5_1 with some initial stability,
4308 * locking on MD and file I/O.
4309 * documentation updates
4310 * several bug fixes since 0.5.0
4311 * small changes in wire protocol
4313 2002-07-18 Phil Schwan <phil@clusterfs.com>
4315 * delivered as Lustre Light Alpha
4316 * fixed a crash after handling invalid MDS requests
4317 * fixed directory pages for architectures with non-4k pages sizes
4319 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4320 * release version v0_4_4
4321 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4323 * Fixes a number of interruption problems with OST operations.
4324 * Update documentation for portals header changes
4325 * Move all wire protocol structs/defines to lustre_idl.h
4326 * Fixes symlink length bug.
4327 * Add tcpdump to repository.
4329 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4330 * release version v0_4_3
4331 * Fixes statfs for inodes on extN.
4332 * Fixes bug in runtests which would delete /etc/hosts.
4333 * Use 64-bit object IDs wherever possible (not into VFS though)
4334 Remove ost_get_info, which is unused by lustre, and out of date.
4336 2002-07-03 Peter Braam <braam@clusterfs.com>
4337 * release version v0_4_2 Fixes a lookup error (type not passed)
4338 * move forward to head of Portals
4339 * move forward to latest Lustre kernel
4341 2002-06-25 Peter Braam <braam@clusterfs.com>
4342 * release version v0_4_1. Hopefully stable on single node use.