3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs5
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: make mgs_setparam() handle fsname containing dash
18 Details : fsname containing a dash does not work with lctl conf_param
20 Severity : enhancement
22 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
24 Severity : enhancement
26 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
30 Frequency : rare, at shutdown
31 Description: access already free / zero obd_namespace.
32 Details : if client_disconnect_export was called without force flag set,
33 and exist connect request in flight, this can produce access to
34 NULL pointer (or already free pointer) when connect_interpret
35 store ocd flags in obd_namespace.
39 Frequency : only at startup
40 Description: not alloc memory with spinlock held.
41 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
44 Severity : enhancement
46 Description: make lustre randomly fail allocating memory
47 Details : Make lustre randomly failed allocating memory for testing purpose.
49 Severity : enhancement
51 Description: lost problems with lov objid file
52 Details : Fixes some scability and access to not inited memory problems
53 in work with lov objdid file.
58 Description: lfs find does not continue on file error
59 Details : Continue other files processing when a file/dir is absent.
63 Description: Inconsistent usage of lustre_pack_reply()
64 Details : Standardize the usage of lustre_pack_reply() such that it
65 always generate a CERROR on failure.
68 Frequency : occasional
70 Description: MDS hang or stay in waiting lock
71 Details : If client receive lock with CBPENDING flag ldlm need send lock
72 cancel as separate rpc, to avoid situation when cancel request
73 can't processed due all i/o threads stay in wait lock.
77 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
81 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
85 Description: Update to SLES9 kernel-2.6.5-7.287.3.
89 Description: Update to RHEL4 latest kernel.
91 Severity : enhancement
93 Description: Build SLES10 patchless client fails
94 Details : The configure was broken by run ./configure with
95 --with-linux-obj=.... argument for patchless client. When the
96 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
97 can't search header adequately. Use absolute path such as
98 -I($LINUX)/include instead.
102 Description: interrupt oig_wait produce painc on resend.
103 Details : brw_redo_request can be used for resend requests from ptlrpcd and
104 private set, and this produce situation when rq_ptlrpcd_data not
105 copyed to new allocated request and triggered LBUG on assert
106 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
107 ptlrpcd set if request is changed and can be safety changed to use
112 Description: LASSERT_{REQ,REP}SWAB macros are buggy
113 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
114 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
115 these macros and replace them with their difinitions instead.
119 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
120 Details : Update which_patch & target file for SLES10 latest kernel.
123 Description: Remove client patches from SLES 10 kernel.
124 Details : This causes SLES 10 clients to behave as patchless clients
125 even on a Lustre-patched (server) kernel.
127 Severity : enhancement
129 Description: self-adjustable client's lru lists
130 Details : use adaptive algorithm for managing client cached locks lru
131 lists according to current server load, other client's work
132 pattern, memory activities, etc. Both, server and client
133 side namespaces provide number of proc tunables for controlling
136 Severity : enhancement
138 Description: light-weight GSS support
139 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
140 and reduce performance overhead.
142 Severity : enhancement
144 Description: Linux keyring support
145 Details : Support using service of Linux keyring for Lustre GSS internal
146 context refresh/cache mechanism.
150 Description: Fix errors in lfs documentation
151 Details : Fixes man pages
155 Description: don't use GFP_* in generic Lustre code.
156 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
160 Description: obdclass is limited by single OBD_ALLOC(idarray)
161 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
165 Description: warnings with build patchless client with vanila 2.6.19 and up
166 Details : change old ctl_table style and replace ctl_table/ctl_table_header
167 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
171 Frequency : Only for SLES
172 Description: sanity_quota fail test_1
173 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
174 file, which makes TSTID[2] inunique.
178 Frequency : Only for SLES9
179 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
180 Details : Update target/ChangeLog/which_patch .
184 Frequency : Only for test_10 in sanity-quota.sh
185 Description: a bug in quota test code
186 Details : checking a test flag in a wrong place
190 Frequency : Only for test_13 in sanity-quota.sh
191 Description: a bug in quota test script
192 Details : shouldn't check overfull allocated quota
196 Description: jbd statistics
197 Details : Port older jbd statistics patch for sles10
199 Severity : enhancement
201 Description: Add printing inode info into message about error in writepage.
206 Description: reply_lock_interpret crash due to race with it and lock cancel.
207 Details : Do not replay locks that are being cancelled. Do not reference
208 locks by their address during replay, just by their handle.
212 Frequency : When flocks are used.
213 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
214 Details : Flock locks might destroy just granted lock if it could be merged
215 with another existing flock, this is done in completion handler,
216 so teach ldlm_cli_enquque_fini that this is a valid case for
222 Description: Oops in read and write path when failing to allocate lock.
223 Details : Check if lock allocation failed and return error back.
227 Description: lstripe command fails for valid OST index
228 Details : The stripe offset is compared to lov->desc.ld_tgt_count
229 instead of lov->desc.ld_active_tgt_count.
233 Description: sanity.sh failed test 103
234 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
239 Description: df doesn't work properly if diskfs blocksize != 4K
240 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
244 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
245 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
249 Description: obd_health_check_timeout too short
250 Details : set obd_health_check_timeout as 1.5x of obd_timeout
254 Description: llapi_file_create() does not allow some changes
255 Details : add llapi_file_open() that allows specifying the mode and
256 open flags, and also returns an open file handle.
260 Description: merge and cleanup kernel patches.
261 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
265 Description: Add journal checksum support.(Kernel part)
266 Details : The journal checksum feature adds two new flags i.e
267 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
268 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
269 indicates that the commit block contains the checksum for
270 the blocks described by the descriptor blocks. Now commit
271 record can be sent to disk without waiting for descriptor
272 blocks to be written to disk. This behavior is controlled
273 using JBD2_FEATURE_ASYNC_COMMIT flag.
277 Description: OSS needs mutliple precreate threads
278 Details : Add ability to start more than one create thread per OSS.
282 Description: Sanity.sh test_65k failed: setstripe should have succeeded
283 Details : execute lfs setstripe on client
287 Description: mds_obd_create error creating tmp object
288 Details : When the user sets quota on root, llog will be affected and can't
289 create files and write files.
292 Frequency : Always on ia64 patchless client, and possibly others.
294 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
295 Details : This allows the patchless client to be loaded on architectures
300 Description: RedHat Update kernel for RHEL5
301 Details : Add the kernel config file for RHEL5.
305 Description: RedHat Update kernel for RHEL5
306 Details : Modify the kernel config file more closer RHEL5.
310 Description: Build failure against Centos5 (RHEL5)
311 Details : Define PAGE_SIZE when it isn't present.
315 Description: test_7 on sanity-quota.sh can't be used on separate machines
316 Details : add facet to handle it
320 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
321 Details : with NFS, the anon dentry's parent was set to itself in
322 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
323 corresponding dentry other than use rec->ur_name.
325 Severity : enhancement
327 Description: lfs setstripe enhancement
328 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
332 Description: enable data checksumming by default
333 Details : enable checksum by default, allow --disable-checksum
334 configure option and "-o nochecksum" mount option. Checksums
335 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
339 Description: lustre support for RHEL5
340 Details : Add support for RHEL5.
344 Description: Client eviction due to failover config
345 Details : after a connection loss, the lustre client should attempt to
346 reconnect to the last active server first before trying the
347 other potential connections.
351 Description: when mds and osts use different quota unit(32bit and 64bit),
352 quota will be released repeatly.
353 Details : void sending multiple quota reqs to mds, which will keep the
354 status between the reqs.
358 Description: osts not allocated evenly to files
359 Details : change the condition to increase offset_idx
362 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
364 Description: Data corruption for OSTs that are formatted larger than 2TB
366 Details : When generating the bio request for lustre file writes the
367 sector number would overflow a temporary variable before being
368 used for the IO. The data reads correctly from Lustre (which
369 will overflow in a similar manner) but other file data or
370 filesystem metadata may be corrupted in some cases.
374 Description: Tune the kernel for good SCSI performance.
375 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
376 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
381 Description: rewrite ext2-derived code in obdclass/uuid.c
382 Details : rewrite inherited code (uuid parsing code from ext2 utils)
383 from scratch preserving functionality.
387 Description: "lfs find -obd UUID" prints directories
388 Details : "lfs find -obd UUID" will return all directory names instead
389 of just file names. It is incorrect because the directories
390 do not reside on the OSTs.
394 Description: use i_size_read and i_size_write in 2.6 port
395 Details : replace inode->i_size access with i_size_read/write()
398 Frequency : when removing large files
400 Description: scheduling issue during removal of large Lustre files
401 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
402 It causes scheduling issues when removing large files (17TB in the
406 Frequency : only with liblustre clients on XT3
408 Description: evictions taking too long
409 Details : allow llrd to evict clients directly on OSTs
412 Frequency : only on ppc
414 Description: /proc/fs/lustre/devices broken on ppc
415 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
416 We should make sure the type of 'index' is loff_t and then cast
417 explicitly as needed below. Do not assign an explicitly cast
421 Frequency : only for rhel5
423 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
424 Details : Modify the target file & which_kernel.
426 Severity : enhancement
428 Description: omit set fsid for export NFS
429 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
430 as FS_REQUIRES_DEV for avoid problems with generate fsid.
434 Frequency : Always, for kernels after 2.6.16
435 Description: Fix warning idr_remove called for id=.. which is not allocated.
436 Details : Last kernels save old s_dev before kill super and not allow
437 to restore from callback - restore it before call kill_anon_super.
441 Description: buffer overruns could theoretically occur
442 Details : llapi_semantic_traverse() modifies the "path" argument by
443 appending values to the end of the origin string, and a
444 overrun may occur. Adding buffer overrun check in liblustreapi.
448 Description: Fix error on 'ls .' at the top of the Lustre mount.
449 Details : Don't revalidate dentry if it is a root dentry.
453 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
454 Details : Modify vm-tunables-rhel4.patch.
458 Description: Kernel config for 2.6.18-vanilla.
459 Details : Modify targets/2.6-vanilla.target.in.
460 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
461 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
462 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
463 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
467 Description: improve handling recoverable errors
468 Details : if request processig with error which can be recoverable on server
469 request should be resend, otherwise page released from cache and
474 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
475 Details : Modify target file & which_patch.
476 A flaw was found in the IA32 system call emulation provided
477 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
478 value could be stored in the %RAX register, which could trigger an
479 out-of-bounds system call table access. An untrusted local user
480 could exploit this flaw to run code in the kernel
481 (ie a root privilege escalation). (CVE-2007-4573).
485 Description: O_DIRECT bypasses client statistics.
486 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
487 read_ahead_stats were updated. copied the stats section from
488 osc_send_oap_rpc() into async_internal().
492 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
496 Description: change order of libsysio includes
497 Details : '#include sysio.h' should always come before '#include xtio.h'
501 Description: handle "serious error: objid * already exists" more gracefully
502 Details : If LAST_ID value on disk is smaller than the objects existing in
503 the O/0/d* directories, it indicates disk corruption and causes an
504 LBUG(). If the object is 0-length, then we should use the existing
505 object. This will help to avoid a full fsck in most cases.
507 Severity : enhancement
509 Description: adapt the lustre_config script to support the upgrade case
510 Details : Add "-u" option for lustre_config script to support upgrading 1.4
511 server targets to 1.6 in parallel.
515 Description: To avoid grant space > avaible space when the disk is almost
516 full. Without this patch you might see the error "grant XXXX >
517 available" or some LBUG about grant, when the disk is almost
519 Details : In filter_check_grant, for non_grant cache write, we should
520 check the left space by if (*left > ungranted + bytes), instead
521 of (*left > ungranted), because only we are sure the left space
522 is enough for another "bytes", then the ungrant space should be
523 increase. In client, we should update cl_avail_grant only there
524 is OBD_MD_FLGRANT in the reply.
528 Description: Update RHEL 4 kernel to fix local root privilege escalation.
529 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
530 described in CVE-2007-4573. This problem could allow untrusted
531 local users to gain root access.
534 Frequency : when using O_DIRECT and quotas
536 Description: Incorrect file ownership on O_DIRECT output files
537 Details : block usage reported by 'lfs quota' does not take into account
538 files that have been written with O_DIRECT.
543 Description: touch file failed when fs is not full
544 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
545 otherwise we can get ENOSP while fs is not full.
549 Description: parallel lock callbacks
550 Details : Instead of sending blocking and completion callbacks as separated
551 requests, adding them to a set and sending in parallel.
554 Frequency : only for Cray XT3
555 Bugzilla : 12829/13455
556 Description: Changing primary group doesn't change the group lustre assigns to
558 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
559 overridden with the primary group provided by the group upcall,
560 whereas the supplied fsgid can be trusted if it is in the list of
561 supplementary groups returned by the group upcall.
563 --------------------------------------------------------------------------------
565 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
567 * Support for kernels:
568 2.6.5-7.283 (SLES 9),
569 2.6.9-55.EL (RHEL 4),
570 2.6.16.46-0.14 (SLES 10),
571 2.6.18.8 vanilla (kernel.org)
572 * Client support for unpatched kernels:
573 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
574 2.6.16 - 2.6.22 vanilla (kernel.org)
575 * Due to recently discovered recovery problems, we do not recommend
576 using patchless RHEL 4 clients with this or any earlier release.
577 * Recommended e2fsprogs version: 1.39.cfs8
578 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
579 * Starting with this release, the ldiskfs backing filesystem required
580 by Lustre is now in its own package, lustre-ldiskfs. This package
581 should be installed. It is versioned separately from Lustre and
582 may be released separately in future.
586 Description: block reactivating mgc import until all deactivates complete
587 Details : Fix race when failing back MDT/MGS to itself (testing)
590 Frequency : at statup only
592 Description: mds_lov_synchronize race leads to various problems
593 Details : simultaneous MDT->OST connections at startup can cause the
594 sync to abort, leaving the OSC in a bad state.
596 Severity : enhancement
598 Description: add optional extra BUILD_VERSION info
599 Details : add a new environment variable (namely LUSTRE_VERS) which allows
600 to override the lustre version.
603 Frequency : 2.6.18 servers only
605 Description: ll_kern_mount() doesn't release the module reference
606 Details : The ldiskfs module reference count never drops down to 0
607 because ll_kern_mount() doesn't release the module reference.
612 Description: server LBUG when using old ost_num_threads parameter
613 Details : Accept the old ost_num_threads parameter but warn that it
614 is deprecated, and fix an off-by-one error that caused an LBUG.
619 Description: Transient SCSI error results in persistent IO issue
620 Details : iobuf->dr_error is not reinitialized to 0 between two
624 Frequency : sometimes when underlying device returns I/O errors
626 Description: OSTs not going read-only during write failures
627 Details : OSTs are not remounted read-only when the journal commit threads
628 get I/O errors because fsfilt_ext3 calls journal_start/stop()
629 instead of the ext3 wrappers.
633 Description: poor connect scaling with increasing client count
634 Details : Don't run filter_grant_sanity_check for more than 100 exports
635 to improve scaling for large numbers of clients.
638 Frequency : SLES10 only
640 Description: sanity-quota.sh quotacheck failed: rc = -22
641 Details : Quotas cannot be enabled on SLES10.
644 Frequency : liblustre clients only
646 Description: getdirentries does not give error when run on compute nodes
647 Details : getdirentries does not fail when the size specified as an argument
648 is too small to contain at least one entry
650 Severity : enhancement
652 Description: Add LNET router traceability for debug purposes
653 Details : If a checksum failure occurs with a router as part of the
654 IO path, the NID of the last router that forwarded the bulk data
655 is printed so it can be identified.
660 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
661 Details : Due to a race condition, liblustre clients were occasionally
664 Severity : enhancement
666 Description: lfs setstripe use optional parameters instead of postional
669 Severity : enhancement
671 Description: Nanosecond timestamp support for ldiskfs
672 Details : The on-disk ldiskfs filesystem has added support for nanosecond
673 resolution timestamps. There is not yet support for this at
674 the Lustre filesystem level.
677 Frequency : during server recovery
679 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
680 Details : request with rq_no_resend flag not awake l_wait_event if they get a
684 Frequency : nfs export on patchless client
686 Description: connectathon hang when test nfs export over patchless client
687 Details : Disconnected dentry cannot be found with lookup, so we do not need
688 to unhash it or make it invalid
691 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
692 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
693 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
698 Description: open req refcounting wrong on reconnect
699 Details : If reconnect happened between getting open reply from server and
700 call to mdc_set_replay_data in ll_file_open, we will schedule
701 replay for unreferenced request that we are about to free.
702 Subsequent close will crash in variety of ways.
703 Check that request is still eligible for replay in
704 mdc_set_replay_data().
709 Description: disable writes to filesystem when reading health_check file
710 Details : the default for reading the health_check proc file has changed
711 to NOT do a journal transaction and write to disk, because this
712 can cause reads of the /proc file to hang and block HA state
713 checking on a healthy but otherwise heavily loaded system. It
714 is possible to return to the previous behaviour during configure
715 with --enable-health-write.
717 Severity : enhancement
719 Description: 64-bit inode version
720 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
721 to the inode. This will be required for version-based recovery.
726 Description: MDS fails to start if a duplicate client export is detected
727 Details : in some rare cases it was possible for a client to connect to
728 an MDS multiple times. Upon recovery the MDS would detect this
729 and fail during startup. Handle this more gracefully.
731 Severity : enhancement
733 Description: Add -o localflock option to simulate old noflock
735 Details : This will achieve local-only flock/fcntl locks
741 Description: log_commit_thread vs filter_destroy race leads to crash
742 Details : Take import reference before releasing llog record semaphore
747 Description: Wrong request locking in request set processing
748 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
749 imp_delayed_list, in this place should be used imp_lock.
752 Frequency : when reconnection
754 Description: Grant Leak when osc reconnect to OST
755 Details : When osc reconnect ost, OST(filter) should check whether it
756 should grant more space to client by comparing fed_grant and
757 cl_avail_grant, and return the granted space to client instead
758 of "new granted" space, because client will call osc_init_grant
759 to update the client grant space info.
762 Frequency : when client reconnect to OST
764 Description: Grant Leak when osc do resend and replay bulk write
765 Details : When osc reconnect to OST, OST(filter)should clear grant info of
766 bulk write request, because the grant info while be sync between
767 OSC and OST when reconnect, and we should ignore the grant info
768 these of resend/replay write req.
773 Description: Grant space more than avaiable left space sometimes.
774 Details : When then OST is about to be full, if two bulk writing from
775 different clients came to OST. Accord the avaliable space of the
776 OST, the first req should be permitted, and the second one
777 should be denied by ENOSPC. But if the seconde arrived before
778 the first one is commited. The OST might wrongly permit second
779 writing, which will cause grant space > avaiable space.
782 Frequency : when client is evicted
784 Description: Grant might be wrongly erased when osc is evicted by OST
785 Details : when the import is evicted by server, it will fork another
786 thread ptlrpc_invalidate_import_thread to invalidate the
787 import, where the grant will be set to 0. While the original
788 thread will update the grant it got when connecting. So if
789 the former happened latter, the grant will be wrongly errased
790 because of this race.
795 Description: Checking Stale with correct fid
796 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
797 to check whether it is stale, because sometimes, we want the
798 enqueue happened anyway, and op_data.fid2 will not be initialized.
800 Severity : enhancement
802 Description: update patchless client
803 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
806 Frequency : only with 2.4 kernel
808 Description: random memory corruption
809 Details : size of struct ll_inode_info is to big for union inode.u and this
810 can be cause of random memory corruption.
815 Description: Memory leak in recovery
816 Details : Lov_mds_md was not free in an error handler in mds_create_object.
817 It should also check obd_fail before fsfilt_start, otherwise if
818 fsfilt_start return -EROFS,(failover mds during mds recovery).
819 then the req will return with repmsg->transno = 0 and rc = EROFS.
820 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
821 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
822 be freed no matter whether fsfilt_commit success or not.
825 Frequency : only with huge count clients
827 Description: Prevents from taking the superblock lock in llap_from_page for
829 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
830 for llap_from_page call in ll_removepage prevents from taking the
831 superblock lock for a soon died page.
836 Description: Not check open intent error before release open handle
837 Details : in some rare cases, the open intent error is not checked before
838 release open handle, which may cause
839 ASSERTION(open_req->rq_transno != 0), because it tries to release
840 the failed open handle.
845 Description: Set cat log bitmap only after create log success.
846 Details : in some rare cases, the cat log bitmap is set too early. and it
847 should be set only after create log success.
851 Description: Accessing a block bevice can re-enable I/O when Lustre is
852 tearing down a device.
853 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
858 Description: service threads may hog cpus when there are a lot of requests
859 Details : Insert cond_resched to give other threads a chance to use some CPU
864 Description: the cat log was not initialized in recovery
865 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
866 unlink log on mds will not be initialized until mds post
867 recovery. And also in mds post recovery, the unlink log will
868 initialization will be done asynchronausly, so there will be race
869 between add unlink log and unlink log initialization.
873 Description: brw_stats were being printed incorrectly
874 Details : brw_stats were being printed as log2 but all of them were not
875 recorded as log2. Also remove some code duplication arising from
876 filter_tally_{read,write}.
880 Frequency : rare, only in recovery.
881 Description: ASSERTION(req->rq_type != LI_POISON) failed
882 Details : imp_lock should be held while iterating over imp_sending_list for
883 prevent destroy request after get timeout in ptlrpc_queue_wait.
887 Description: replay-single.sh test 52 fails
888 Details : A lock's skiplist need to be cleanup when it being unlinked
889 from its resource list.
893 Description: Short directio read returns full requested size rather than
895 Details : Direct I/O operations should return actual amount of bytes
896 transferred rather than requested size.
898 Severity : enhancement
900 Description: metadata RPC reduction (e.g. for rm performance)
901 Details : decrease the amount of synchronous RPC between clients and servers
902 by canceling conflicing lock before the operation on the client
903 and packing thier handles into the main operation RPC to server.
905 Severity : enhancement
907 Description: Async OSC create to avoid the blocking unnecessarily.
908 Details : If a OST has no remain object, system will block on the creating
909 when need to create a new object on this OST. Now, ways use
910 pre-created objects when available, instead of blocking on an
911 empty osc while others are not empty. If we must block, we block
912 for the shortest possible period of time.
914 Severity : enhancement
916 Description: refine locking for avoid write wrong info into lov_objid file
917 Details : fix possible races with add new target and write/update data in
920 --------------------------------------------------------------------------------
922 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
927 Frequency : on some architectures
929 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
930 Details : Uninitialized flags sometimes cause configuration commands to
934 Frequency : patchless clients only
936 Description: missing __iget() symbol export
937 Details : The __iget() symbol export is missing. To avoid the need for
938 this on patchless clients the deathrow inode reaper is turned
939 off, and we depend on the VM to clean up old inodes. This
940 dependency was during via the fix for bug 12181.
942 --------------------------------------------------------------------------------
944 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
946 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
947 INTEROPERATE with older versions automatically. In many cases a
948 special upgrade step is needed. Please read the
949 user documentation before upgrading any part of a live system.
950 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
951 version will not interoperate with 1.6 betas before beta5 (1.5.95).
952 * WARNING: Lustre configuration and startup changes are required with
953 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
959 Severity : enhancement
961 Description: MountConf
962 Details : Lustre configuration is now managed via mkfs and mount
963 commands instead of lmc and lconf. New obd types (MGS, MGC)
964 are added for dynamic configuration management. See
965 https://mail.clusterfs.com/wikis/lustre/MountConf for
968 Severity : enhancement
970 Description: dynamic OST addition
971 Details : OSTs can now be added to a live filesystem
973 Severity : enhancement
975 Description: startup order invariance
976 Details : MDTs and OSTs can be started in any order. Clients only
977 require the MDT to complete startup.
979 Severity : enhancement
981 Description: parallel, asynchronous orphan cleanup
982 Details : orphan cleanup is now performed in separate threads for each
983 OST, allowing parallel non-blocking operation.
985 Severity : enhancement
987 Description: optimized stripe assignment
988 Details : stripe assignments are now made based on ost space available,
989 ost previous usage, and OSS previous usage, in order to try
990 to optimize storage space and networking resources.
992 Severity : enhancement
994 Description: Permanently set tunables
995 Details : All writable /proc/fs/lustre tunables can now be permanently
996 set on a per-server basis, at mkfs time or on a live system.
998 Severity : enhancement
1000 Description: Lustre message v2
1001 Details : Add lustre message format v2.
1003 Severity : enhancement
1005 Description: client OST exclusion list
1006 Details : Clients can be started with a list of OSTs that should be
1007 declared "inactive" for known non-responsive OSTs.
1011 Description: SPEC SFS validation failure on NFS v2 over lustre.
1012 Details : Changes the blocksize for regular files to be 2x RPC size,
1013 and not depend on stripe size.
1015 Severity : enhancement
1017 Description: Multiple MD RPCs in flight.
1018 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1019 To avoid overly-overloading MDS, introduce a limit on number of
1020 MDS RPCs in flight for a single client and add /proc controls
1021 to adjust this limit.
1023 Severity : enhancement
1025 Description: client read/write statistics
1026 Details : Add client read/write call usage stats for performance
1027 analysis of user processes.
1028 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1029 file access. extents_stats shows chunk size distribution.
1030 extents_stats_per_process show chunk size distribution per
1033 Severity : enhancement
1035 Description: mds statistics
1036 Details : Add detailed mds operations statistics in
1037 /proc/fs/lustre/mds/*/stats.
1041 Description: Failure of copying files with lustre special EAs.
1042 Details : Client side always return success for setxattr call for lustre
1043 special xattr (currently only "trusted.lov").
1048 Description: Refcount LNET uuids
1049 Details : The global LNET uuid list grew linearly with every startup;
1050 refcount repeated list entries instead of always adding to
1053 Severity : enhancement
1055 Description: Dynamic service threads
1056 Details : Within a small range, start extra service threads
1057 automatically when the request queue builds up.
1060 Frequency : mixed-endian client/server environments
1062 Description: mixed-endian crashes
1063 Details : The new msg_v2 system had some failures in mixed-endian
1066 Severity : enhancement
1068 Description: Easy OST removal
1069 Details : OSTs can be permanently deactivated with e.g. 'lctl
1070 conf_param lustre-OST0001.osc.active=0'
1072 Severity : enhancement
1074 Description: MGS proc entries
1075 Details : Added basic proc entries for the MGS showing what filesystems
1078 Severity : enhancement
1080 Description: provide MGS failover
1081 Details : Added config lock reacquisition after MGS server failover.
1083 Severity : enhancement
1085 Description: add Linux 2.4 support
1086 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1090 Description: a large application tries to do I/O to the same resource and dies
1091 in the middle of it.
1092 Details : Check the req->rq_arrival time after the call to
1093 ost_brw_lock_get(), but before we do anything about
1094 processing it & sending the BULK transfer request. This
1095 should help move old stale pending locks off the queue as
1096 quickly as obd_timeout.
1099 Frequency : when an incorrect nid is specified during startup
1101 Description: ptlrpc connect to non-existant node causes kernel crash
1102 Details : LNET can't be re-entered from an event callback, which
1103 happened when we expire a message after the export has been
1104 cleaned up. Instead, hand the zombie cleanup off to another
1107 Severity : enhancement
1109 Description: plain/inodebits lock performance improvement
1110 Details : Grouping plain/inodebits in granted list by their request modes
1111 and bits policy, thus improving the performance of search through
1115 Frequency : only if OST filesystem is corrupted
1117 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1118 Details : for a short time RPCs with bulk IO are in the replay list,
1119 but replay of bulk IOs is unimplemented. If the OST filesystem
1120 is corrupted due to disk cache incoherency and then replay is
1121 started it is possible to trip an assertion. Avoid putting
1122 committed RPCs into the replay list at all to avoid this issue.
1125 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1127 Description: System hang on startup
1128 Details : This bug allowed the liblustre (e.g. catamount) client to
1129 return to the app before handling all startup RPCs. This
1130 could leave the node unresponsive to lustre network traffic
1131 and manifested as a server ptllnd timeout.
1133 Severity : enhancement
1135 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1136 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1137 boolean to control whether to print peer debug info when a
1138 client's RPC times out.
1141 Frequency : only for kernels with patches from Lustre below 1.4.3
1143 Description: Remove old rdonly API
1144 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1147 Frequency : only for devices with external journals
1149 Description: Set external device read-only also
1150 Details : During a commanded failover stop, we set the disk device
1151 read-only while the server shuts down. We now also set any
1152 external journal device read-only at the same time.
1155 Frequency : when upgrading from 1.4 while trying to change parameters
1157 Description: The wrong (new) MDC name was used when setting parameters for
1158 upgraded MDT's. Also allows changing of OSC (and MDC)
1159 parameters if --writeconf is specified at tunefs upgrade time.
1162 Frequency : when setting specific ost indicies
1164 Description: QOS code breaks on skipped indicies
1165 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1166 created with --index need not be sequential.
1168 Severity : enhancement
1170 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1171 Details : The uninit_groups feature works in conjunction with the kernel
1172 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1173 up the pass1 processing of e2fsck. This is a read-only feature
1174 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1175 mount filesystems that have had this feature enabled.
1177 Severity : enhancement
1179 Description: Improve multi-block allocation algorithm to avoid fragmentation
1180 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1181 allocation locality and avoid filesystem fragmentation.
1183 ------------------------------------------------------------------------------
1185 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1187 * Support for kernels:
1188 2.4.21-47.0.1.EL (RHEL 3)
1189 2.6.5-7.283 (SLES 9)
1190 2.6.9-42.0.10.EL (RHEL 4)
1191 2.6.12.6 vanilla (kernel.org)
1192 2.6.16.27-0.9 (SLES 10)
1193 * Recommended e2fsprogs version: 1.39.cfs5
1195 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1199 Frequency : occasional, depends on client load and configuration
1200 Bugzilla : 12181, 12203
1201 Description: data loss for recently-modified files
1203 Details : In some cases it is possible that recently written or created
1204 files may not be written to disk in a timely manner (this should
1205 normally be within 30s unless client IO load is very high).
1206 The problem appears as zero-length files or files that are a
1207 multiple of 1MB in size after a client crash or client eviction
1208 that are missing data at the end of the file.
1210 This problem is more likely to be hit on clients where files are
1211 repeatedly created and unlinked in the same directory, clients
1212 have a large amount of RAM, have many CPUs, the filesystem has
1213 many OSTs, the clients are rebooted frequently, and/or the files
1214 are not accessed by other nodes after being written.
1216 The presence of the problem can be detected by looking at
1217 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1218 smaller than the second (nr_unused) then dirty files will not
1219 be flushed automatically to disk. "sync; sleep 10" should be
1220 run several times on the node before unmounting it to update
1221 Lustre (this is also safe to run on nodes without this problem).
1223 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1224 that can cause this same problem, so customers using that kernel
1225 also need to update the kernel in addition to Lustre. In order
1226 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1228 It is normal that files written just before a client crash (less
1229 than 30s) may not yet have been flushed to disk, even for local
1233 Frequency : frequent on thin XT3 nodes
1235 Description: UUID collision on thin XT3 Linux nodes
1236 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1237 randomly, since we relied on an insufficiently-seeded PRNG.
1242 Description: OSS hangs after "All ost request buffers busy"
1243 Details : A deadlock between quota and journal operations caused OSS
1244 hangs after printing "All ost request buffers busy."
1247 Frequency : always on liblustre builds
1249 Description: Cleanup compiler warnings on liblustre
1252 Frequency : always on liblustre builds on XT3
1254 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1258 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1259 > 32000 subdirectories
1260 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1261 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1262 subdirectory count crosses 32000. This will aid e2fsck to
1263 correctly handle more than 32000 subdirectories.
1266 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1268 Description: System hang on startup
1269 Details : This bug allowed the liblustre (e.g. catamount) client to
1270 return to the app before handling all startup RPCs. This
1271 could leave the node unresponsive to lustre network traffic
1272 and manifested as a server ptllnd timeout.
1274 Severity : enhancement
1276 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1277 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1278 boolean to control whether to print peer debug info when a
1279 client's RPC times out.
1284 Description: make O_SYNC working on 2.6 kernels
1285 Details : 2.6 kernels use different method for mark pages for write,
1286 so need add a code to lustre for O_SYNC work.
1291 Description: Failure to close file and release space on NFS
1292 Details : Put inode details into lock acquired in ll_intent_file_open.
1293 Use mdc_intent_lock in ll_intent_open to properly
1294 detect all kind of errors unhandled by mdc_enqueue.
1299 Description: proc file read during shutdown sometimes raced obd removal,
1301 Details : Add lock to prevent obd access after proc file removal.
1304 Frequency : Only for files larger than 4GB on 32-bit clients.
1306 Description: improperly doing page alignment of locks
1307 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1308 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1313 Description: RPCs being resent when they shouldn't be
1314 Details : Some RPCs that should not be resent are being resent. This
1315 can cause inconsistencies in the RPC state machine. Do not
1316 resend such requests.
1319 Frequency : rare, only with NFS export
1321 Description: Crash on NFS re-export node
1322 Details : under very unusual load conditions an assertion is hit in
1323 ll_intent_file_open()
1326 Frequency : only if OST filesystem is corrupted
1328 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1329 Details : for a short time RPCs with bulk IO are in the replay list,
1330 but replay of bulk IOs is unimplemented. If the OST filesystem
1331 is corrupted due to disk cache incoherency and then replay is
1332 started it is possible to trip an assertion. Avoid putting
1333 committed RPCs into the replay list at all to avoid this issue.
1338 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1339 Details : Large single O_DIRECT read and write calls can fail to allocate
1340 a sufficiently large buffer to process the request. In case of
1341 allocation failure the allocation is retried with a smaller
1342 buffer and broken into smaller requests.
1344 Severity : enhancement
1346 Description: Add -o localflock option to simulate old noflock behaviour.
1347 Details : This will achieve local-only flock/fcntl locks coherentness.
1352 Description: versioning check is incomplete
1353 Details : Checking the version difference of client vs. server, report
1354 error if the gap is too big.
1359 Description: add support PG_writeback bit
1360 Details : add support for PG_writeback bit for Lustre, for more carefull
1361 work with page cache in 2.6 kernel. This also fix some deadlocks
1362 and remove hack for work O_SYNC with 2.6 kernel.
1364 Severity : enhancement
1366 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1367 Details : The uninit_groups feature works in conjunction with the kernel
1368 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1369 up the pass1 processing of e2fsck. This is a read-only feature
1370 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1371 mount filesystems that have had this feature enabled.
1373 Severity : enhancement
1375 Description: Improve multi-block allocation algorithm to avoid fragmentation
1376 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1377 allocation locality and avoid filesystem fragmentation.
1379 ------------------------------------------------------------------------------
1381 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1383 * Support for kernels:
1384 2.6.9-42.0.3.EL (RHEL 4)
1385 2.6.5-7.276 (SLES 9)
1386 2.4.21-47.0.1.EL (RHEL 3)
1387 2.6.12.6 vanilla (kernel.org)
1388 2.6.16.21-0.8 (SLES10)
1389 * Recommended e2fsprogs version: 1.39.cfs2-0
1391 * The backwards-compatible /proc/sys/portals symlink has been removed
1392 in this release. Before upgrading, please ensure that you change
1393 any configuration scripts or /etc/sysctl.conf files that access
1394 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1395 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1396 in advance of the upgrade on any system running Lustre 1.4.6 or
1397 newer, since /proc/sys/lnet was added in that version.
1398 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1402 Frequency : only when quota is used
1404 Description: avoid scanning export list for quota master
1405 Details : Change the algorithms to avoid scanning export list in order
1406 to improve the efficiency.
1409 Frequency : MDS failover only, very rarely
1411 Description: "went back in time" messages on mds failover
1412 Details : The greatest transno may be lost when the current operation
1413 finishes with an error (transno==0) and the client's last_rcvd
1414 record is over-written. Save the greatest transno in the
1415 mds_last_transno for this case.
1418 Frequency : always for specific kernels and striping counts
1420 Description: client may get "Matching packet too big" without ACL support
1421 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1422 when trying to access files in certain configurations. The
1423 clients should in fact be denied when mounting because they do
1424 not understand ACLs.
1427 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1429 Description: many clients connecting with IO in progress causes connect timeouts
1430 Details : Avoid synchronous journal commits to avoid delays caused by many
1431 clients connecting/disconnecting when bulk IO is in progress.
1432 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1433 OST_IO_PORTAL to avoid delays behind potentially many pending
1437 Frequency : occasionally with multiple writers to a single file
1439 Description: shared writes to file may result in wrong size reported by stat()
1440 Details : Allow growing of kms when extent lock is cancelled
1443 Frequency : always with random mmap IO to multi-striped file
1445 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1446 Details : Only if the hole is at the end of OST object so that kms is too
1447 small. Fix is to increase kms accordingly in ll_nopage.
1450 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1452 Description: writes to a missing object would leak memory on the OST
1453 Details : If there is an inconsistency between the MDS and OST filesystems,
1454 such that the MDS references an object that doesn't exist, writes
1455 to that object will leak memory due to incorrect cleanup in the
1456 error handling path, eventually running out of memory on the OST.
1461 Description: Creating too long symlink causes lustre errors
1462 Details : Check symlink and name lengths before sending requests to MDS.
1465 Frequency : only if flock is enabled (not on by default)
1467 Description: posix locks not released on fd closure on 2.6.9+
1468 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1469 kernels, this caused such locks not to be released on fd close and
1470 then assertions on fs unmount about still used locks.
1473 Frequency : MDS failover only, very rarely
1475 Description: clients may get ASSERTION(granted_lock != NULL)
1476 Details : When request was taking a long time, and a client was resending
1477 a getattr by name lock request. The were multiple lock requests
1478 with the same client lock handle and
1479 mds_getattr_name->fixup_handle_for_resent_request found one of the
1480 lock handles but later failed with ASSERTION(granted_lock != NULL).
1485 Description: handle->h_buffer_credits > 0, assertion failure
1486 Details : h_buffer_credits is zero after truncate, causing assertion
1487 failure. This patch extends the transaction or creates a new
1491 Frequency : NFS re-export or patchless client
1492 Bugzilla : 11179, 10796
1493 Description: Crash on NFS re-export node (__d_move)
1494 Details : We do not want to hash the dentry if we don't have a lock.
1495 But if this dentry is later used in d_move, we'd hit uninitialised
1496 list head d_hash, so we just do this to init d_hash field but
1497 leave dentry unhashed.
1500 Frequency : NFS re-export or patchless client
1502 Description: NFS exports has problem with symbolic link
1503 Details : lustre client didn't properly install dentry when re-exported
1504 to NFS or running patchless client.
1507 Frequency : NFS re-export or patchless client
1509 Description: Various nfs/patchless fixes.
1510 Details : fixes reuse disconected alias for lookup process - this fixes
1511 warning "find_exported_dentry: npd != pd",
1512 fix permission error with open files at nfs.
1513 fix apply umask when do revalidate.
1516 Frequency : occasional
1518 Description: Crash on NFS re-export node
1519 Details : calling clear_page() on the wrong pointer triggered oops in
1520 generic_mapping_read().
1523 Frequency : rarely, using O_DIRECT IO
1525 Description: unaligned directio crashes client with LASSERT
1526 Details : check for unaligned buffers before trying any requests.
1529 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1531 Description: stale data returned from RAID cache
1532 Details : If only a small amount of IO is done to the RAID device before
1533 reading it again it is possible to get stale data from the RAID
1534 cache instead of reading it from disk.
1537 Frequency : always for sles10 kernel
1539 Description: sles10 support
1540 Details : ll_follow_link: compile fixes and using of nd_set_link
1541 under newer kernels.
1544 Frequency : depends on arch, kernel and compiler version, always on sles10
1547 Description: recursive or deep enough symlinks cause stack overflow
1548 Details : getting rid of large stack-allocated variable in
1552 Frequency : depends on hardware
1554 Description: lustre write performance loss in the SLES10 kernel
1555 Details : the performance loss is caused by using of write barriers in the
1556 ext3 code. The SLES10 kernel turns barrier support on by
1557 default. The fix is to undo that change for ldiskfs.
1559 ------------------------------------------------------------------------------
1561 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1563 * Support for kernels:
1564 2.6.9-42.0.3EL (RHEL 4)
1565 2.6.5-7.276 (SLES 9)
1566 2.4.21-47.0.1.EL (RHEL 3)
1567 2.6.12.6 vanilla (kernel.org)
1571 Frequency : quota enabled and large files being deleted
1573 Description: releasing more than 4GB of quota at once hangs OST
1574 Details : If a user deletes more than 4GB of files on a single OST it
1575 will cause the OST to spin in an infinite loop. Release
1576 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1581 Description: statfs data retrieved from /proc may be stale or zero
1582 Details : When reading per-device statfs data from /proc, in the
1583 {kbytes,files}_{total,free,avail} files, it may appear
1584 as zero or be out of date.
1587 Frequency : systems with MD RAID1 external journal devices
1589 Description: lconf's call to blkid is confused by RAID1 journal devices
1590 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1591 of returning all block devices that match the journal UUID.
1594 Frequency : always, for aggregate stripe size over 4GB
1596 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1597 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1598 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1601 Frequency : always if "lfs find" used on a local file/directory
1603 Description: "lfs find" segfaults if used on a local file/directory
1604 Details : The case where a directory component was not specified wasn't
1605 handled correctly. Handle this properly.
1608 Frequency : always on ppc64
1610 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1611 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1612 perform endianness conversion.
1615 Frequency : rarely (truncate to non-zero file size after write under load)
1616 Bugzilla : 10730, 10687
1617 Description: Files padded with zeros to next 4K multiple
1618 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1619 it is possible that files that are truncated to a non-zero size
1620 immediately after being written are filled with zero bytes beyond
1621 the truncated size. No file data is lost.
1623 Severity : enhancement
1625 Description: Allow recovery/failover for liblustre clients.
1626 Details : liblustre clients were unaware of failover configurations until
1629 Severity : enhancement
1631 Description: user file locks should fail when not mounting with flock option
1632 Details : Set up an error-returning stub in ll_file_operations.lock field
1633 to prevent incorrect behaviour when client is mounted without
1634 flock option. Also, set up properly f_op->flock field for
1638 Frequency : always on ia64
1640 Description: "lfs df" loops on printing out MDS statfs information
1641 Details : The obd_ioctl_data was not initialized and in some systems
1642 this caused a failure during the ioctl that did not return
1643 an error. Initialize the struct and return an error on failure.
1646 Frequency : SLES 9 only
1648 Description: Error of copying files with lustre special EAs as root
1649 Details : Client side always return success for setxattr call for lustre
1650 special xattr (currently only "trusted.lov").
1653 Frequency : rarely on clusters with both ia64+i386 clients
1655 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1656 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1657 clients are doing concurrent writes to the same file it is
1658 possible that the ia64 clients may LASSERT because the OST
1659 extent locks are not PAGE_SIZE aligned. Ensure that grown
1660 locks are always aligned on the request boundary.
1663 Frequency : specific use, occasional
1665 Description: Overwriting in use executable truncates on-disk binary image
1666 Details : If one node attempts to overwrite an executable in use by
1667 another node, we now correctly return ETXTBSY instead of
1668 truncating the file.
1670 Severity : enhancement
1672 Description: Async OSC create to avoid the blocking unnecessarily.
1673 Details : If a OST has no remain object, system will block on the creating
1674 when need to create a new object on this OST. Now, ways use
1675 pre-created objects when available, instead of blocking on an
1676 empty osc while others are not empty. If we must block, we block
1677 for the shortest possible period of time.
1682 Description: chmod on Lustre root is propagated to other clients
1683 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1684 invalid by the follow_mount time.
1687 Frequency : liblustre clients only
1689 Description: Race in 'instant cancel' lock handling could lead to such locks
1690 never to be granted in case of SMP MDS
1691 Details : Do not destroy not yet granted but cbpending locks in
1695 Frequency : replay/resend of open
1697 Description: non null lock assetion failure in mds_intent_policy
1698 Details : Trying to replay/resend lockless open requests resulted in
1699 mds_open() returning 0 with no lock. Now it sets a flag if
1700 a lock is going to be returned.
1702 Severity : enhancement
1704 Description: Checksum enhancements
1705 Details : New checksum enhancements allow for resending RPCs that failed
1708 Severity : enhancement
1710 Description: Tunables on number of dirty pages in cacche
1711 Details : Allow to set limit on number of dirty pages cached.
1716 Description: client crash on unmount - lock still has references
1717 Details : In some error handling cases it was possible to leak a lock
1718 reference on a client while accessing a file. This was not
1719 harmful to the client during operation, but would cause the
1720 client to crash when the filesystem is unmounted.
1723 Frequency : specific case, rare
1725 Description: ETXTBSY on mds though file not in use by client
1726 Details : ETXTBSY is no longer incorrectly returned when attempting to
1727 chmod or chown a directory that the user previously tried to
1728 execute or a currently-executing binary.
1731 Frequency : extremely rare except on liblustre-based clients
1733 Description: Lustre space not freed when files are deleted
1734 Details : Clean up open-unlinked files after client eviction. Previously
1735 the unlink was skipped and the files remained as orphans.
1740 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1741 Details : In some cases it was possible to send a blocking callback to
1742 a client doing a glimpse, even though that client didn't get
1743 a lock granted. When the glimpse lock is cancelled on the OST
1744 the freed lock is left on the waiting list and corrupted the list.
1747 Frequency : all core dumps
1749 Description: Broke core dumps to lustre
1750 Details : Negative dentry may be unhashed if parent does not have UPDATE
1751 lock, but some callers, e.g. do_coredump, expect dentry to be
1752 hashed after successful create, hash it in ll_create_it.
1754 ------------------------------------------------------------------------------
1756 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1758 * Support for kernels:
1759 2.6.9-42.0.2.EL (RHEL 4)
1760 2.6.5-7.276 (SLES 9)
1761 2.4.21-40.EL (RHEL 3)
1762 2.6.12.6 vanilla (kernel.org)
1766 Frequency : always on RHEL 3
1768 Description: Number of open files grows over time
1769 Details : The number of open files grows over time, whether or not
1770 Lustre is started. This was due to a filp leak introduced
1771 by one of our kernel patches.
1773 ------------------------------------------------------------------------------
1775 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1777 * Support for kernels:
1778 2.6.9-42.EL (RHEL 4)
1779 2.6.5-7.276 (SLES 9)
1780 2.4.21-40.EL (RHEL 3)
1781 2.6.12.6 vanilla (kernel.org)
1786 Bugzilla : 5719, 9635, 9792, 9684
1787 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1788 Details : If a server is under heavy load and cannot reply to new
1789 connection requests before the client resends the (re)connect,
1790 the connection handling code can behave badly if two service
1791 threads are concurrently handing separate (re)connections from
1792 the same client. Add better locking to the connection handling
1793 code, and ensure that only a single connection will be processed
1794 for a given client UUID, even if the lock is dropped.
1796 Severity : enhancement
1798 Description: add TCP zero-copy support to kernel
1799 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1800 sends if the hardware supports scatter-gather and checksumming.
1801 This allows socklnd to do client-write and server-read more
1802 efficiently and reduce CPU utilization from skbuf copying.
1805 Frequency : only if NFS exporting from client
1807 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1808 Details : If setting attributes on a file created under NFS that had
1809 never been opened it would be possible to oops the client
1810 if the file had no objects.
1813 Frequency : always for liblustre
1815 Description: liblustre client does MDS+OSTs setattr RPC for each write
1816 Details : When doing a write from a liblustre client, the client
1817 incorrectly issued an RPC to the MDS and each OST the file was
1818 striped over in order to update the timestamps. When writing
1819 with small chunks and many clients this could overwhelm the MDS
1820 with RPCs. In all cases it would slow down the write because
1821 these RPCs are unnecessary.
1823 Severity : enhancement
1825 Description: allow number of MDS service threads to be changed at module load
1826 Details : It is now possible to change the number of MDS service threads
1827 running. Adding "options mds mds_num_threads={N}" to the MDS's
1828 /etc/modprobe.conf will set the number of threads for the next
1829 time Lustre is restarted (assuming the "mds" module is also
1830 reloaded at that time). The default number of threads will
1831 stay the same, 32 for most systems.
1836 Description: OST crash if filesystem is unformatted or corrupt
1837 Details : If an OST is started on a device that has never been formatted
1838 or if the filesystem is corrupt and cannot even mount then the
1839 error handling cleanup routines would dereference a NULL pointer.
1844 Description: NULL pointer deref in llap_from_page.
1845 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1846 incorrect) mapping if the page was truncated/reclaimed while it was
1847 searched for. Check for this condition and skip such pages when
1848 doing readahead. Introduce extra check to llap_from_page() to
1849 verify page->mapping->host is non-NULL (so page is not anonymous).
1852 Frequency : Sometimes when using sys_sendfile
1854 Description: "page not covered by a lock" warnings from ll_readpage
1855 Details : sendfile called ll_readpage without right page locks present.
1856 Now we introduced ll_file_sendfile that does necessary locking
1857 around call to generic_file_sendfile() much like we do in
1861 Frequency : with certain MDS communication failures at client mount time
1863 Description: NULL pointer deref after failed client mount
1864 Details : a client connection request may delayed by the network layer
1865 and not be sent until after the PTLRPC layer has timed out the
1866 request. If the client fails the mount immediately it will try
1867 to clean up before the network times out the request. Add a
1868 reference from the request import to the obd device and delay
1869 the cleanup until the network drops the request.
1872 Frequency : occasionally during client (re)connect
1874 Description: assertion failure during client (re)connect
1875 Details : processing a client connection request may be delayed by the
1876 client or server longer than the client connect timeout. This
1877 causes the client to resend the connection request. If the
1878 original connection request is replied in this interval, the
1879 client may trip an assertion failure in ptlrpc_connect_interpret()
1880 which thought it would be the only running connect process.
1883 Frequency : only with obd_echo servers and clients that are rebooted
1885 Description: kernel BUG accessing uninitialized data structure
1886 Details : When running an obd_echo server it did not start the ping_evictor
1887 thread, and when a client was evicted an uninitialized data
1888 structure was accessed. Start the ping_evictor in the RPC
1889 service startup instead of the OBD startup.
1891 Severity : enhancement
1892 Bugzilla : 10193 (patchless)
1893 Description: Remove dependency on various unexported kernel interfaces.
1894 Details : No longer need reparent_to_init, exit_mm, exit_files,
1895 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1898 Frequency : rare (only users of deprecated and unsupported LDAP config)
1900 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1901 Details : LDAP apparently contains 'lustreName' attributes instead of
1902 'name'. A simple remapping of the name is sufficient.
1905 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1907 Description: waiting_locks_callback trips kernel BUG if client is evicted
1908 Details : Running with the dump_on_timeout debug flag turned on makes
1909 it possible that the waiting_locks_callback() can try to dump
1910 the Lustre kernel debug logs from an interrupt handler. Defer
1911 this log dumping to the expired_lock_main() thread.
1913 Severity : enhancement
1915 Description: Support NFS exporting on 2.6 kernels.
1916 Details : Implement non-rawops metadata methods for NFS server to use without
1917 changing NFS server code.
1920 Frequency : very rare (synthetic metadata workload only)
1922 Description: two racing renames might cause an MDS thread to deadlock
1923 Details : Running the "racer" program may cause one MDS thread to rename
1924 a file from being the source of a rename to being the target of
1925 a rename at exactly the same time that another thread is doing
1926 so, and the second thread has already enqueued these locks after
1927 doing a lookup of the target and is trying to relock them in
1928 order. Ensure that we don't try to re-lock the same resource.
1931 Frequency : only very large systems with liblustre clients
1933 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1934 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1935 all OSTs in parallel. This allows the eviction of stale liblustre
1936 clients to proceed much faster than if they were done in series,
1937 and also offers similar improvements for other set_info RPCs.
1942 Description: excessive CPU usage during initial read phase on client
1943 Details : During the initial read phase on a client, it would agressively
1944 retry readahead on the file, consuming too much CPU and impacting
1945 performance (since 1.4.5.8). Improve the readahead algorithm
1946 to avoid this, and also improve some other common cases (read
1947 of small files in particular, where "small" is files smaller than
1948 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1953 Description: MDS crash when receiving packet with unknown intent.
1954 Details : Do not LBUG in unknown intent case, just return -EFAULT
1956 Severity : enhancement
1957 Bugzilla : 9293, 9385
1958 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1959 Details : Do not serialize getattr (non-intent version) and statfs.
1962 Frequency : occasional, when OST network is overloaded/intermittent
1964 Description: client evicted by OST after bulk IO timeout
1965 Details : If a client sends a bulk IO request (read or write) the OST
1966 may evict the client if it is unresposive to its data GET/PUT
1967 request. This is incorrect if the network is overloaded (takes
1968 too long to transfer the RPC data) or dropped the OST GET/PUT
1969 request. There is no need to evict the client at all, since
1970 the pinger and/or lock callbacks will handle this, and the
1971 client can restart the bulk request.
1974 Frequency : Always when mmapping file with no objects
1976 Description: client crashes when mmapping file with no objects
1977 Details : Check that we actually have objects in a file before doing any
1978 operations on objects in ll_vm_open, ll_vm_close and
1984 Description: Request leak when working with deleted CWD
1985 Details : Introduce advanced request refcount tracking for requests
1986 referenced from lustre intent.
1988 Severity : Enhancement
1990 Description: Cache open file handles on client.
1991 Details : MDS now will return special lock along with openhandle, if
1992 requested and client is allowed to hold openhandle, even if unused,
1993 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1994 closing files for every read/write openration.
1996 Severity : Enhancement
1998 Description: Cache open negative dentries on client when possible.
1999 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2000 negative dentries on lock revocation.
2005 Description: Remounting a client read-only wasn't possible with a zconf mount
2006 Details : It wasn't possible to remount a client read-only with llmount.
2008 Severity : enhancement
2009 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2010 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2011 MPICH is included to add Lustre-specific ADIO interfaces.
2012 This is based closely on the UFS ADIO layer and only differs
2013 in file creation, in order to allow the OST striping to be set.
2014 This is user-contributed code and not supported by CFS.
2019 Description: extended inode attributes (immutable, append-only) work improperly
2020 when 2.4 and 2.6 kernels are used on client/server or vice versa
2021 Details : Introduce kernel-independent values for these flags.
2023 Severity : enhancement
2026 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2027 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2028 granularity. Now, Fractional MB granularity is be supported,
2029 this is very useful for low memory system.
2031 Severity : enhancement
2033 Description: Getattr by fid
2034 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2035 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2036 have locks and avoids getattr rpc requests for it on every stat.
2039 Frequency : Always, for filesystems larger than 2TB
2041 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2042 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2043 filesystems larger than 2TB to be mounted with mballoc enabled.
2046 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2048 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2049 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2050 filesystems larger than 2TB is mounted then the kernel will
2051 silently corrupt the start of the filesystem. CONFIG_LBD is
2052 enabled for all CFS-supported kernels, but the possibility of
2053 this happening with a modified kernel config exists.
2055 Severity : enhancement
2057 Description: add client O_DIRECT support for 2.6 kernels
2058 Details : It is now possible to do O_DIRECT reads and writes to files
2059 in the Lustre client mountpoint on 2.6 kernel clients.
2061 Severity : enhancement
2063 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2064 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2065 parallel, not waiting for response from every OST before sending
2066 a rpc to the next OST.
2071 Description: setattr vs write race when updating file timestamps
2072 Details : Client processes that update a file timestamp into the past
2073 right after writing to the file (e.g. tar) it is possible that
2074 the updated file modification time can be reset to the current
2075 time due to a race between processing the setattr and write RPC.
2077 Severity : enhancement
2079 Description: Bring 'lfs find' closer in line with regular Linux find.
2080 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2081 -print0 options and obtains all the needed info through the lustre
2084 Severity : enhancement
2086 Description: support up to 1024 configured devices on one node
2087 Details : change obd_dev array from statically allocated to dynamically
2088 allocated structs as they are first used to reduce memory usage
2093 Description: Flush dirty partially truncated pages during truncate
2094 Details : Immediatelly flush partially truncated pages in filter_setattr,
2095 this way we completely avoid having any pages in page cache on OST
2096 and can retire ugly workarounds during writes to flush such pages.
2101 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2102 Details : protect inode from truncation within vfs_unlink() context
2103 just take a reference before calling vfs_unlink() and release it
2104 when parent's i_sem is free.
2109 Description: last_id value checked outside lock on OST caused LASSERT failure
2110 Details : If there were multiple MDS->OST object precreate requests in
2111 flight, it was possible that the OST's last object id was checked
2112 outside a lock and incorrectly tripped an assertion. Move checks
2113 inside locks, and discard old precreate requests.
2116 Frequency : always, if extents are used on OSTs
2118 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2119 Details : OSTs using the extents format would not zero the high 16 bits of
2120 the index physical block number. This is not a problem for any
2121 OST filesystems smaller than 16TB, and no kernels support ext3
2122 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2123 new/modified files) and can be fixed for existing filesystems
2124 with e2fsprogs-1.39-cfs1.
2129 Description: import connection selection may be incorrect if timer wraps
2130 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2131 import connections to be ignored if the 32-bit jiffies counter
2132 wraps. Use a 64-bit jiffies counter.
2135 Frequency : very large clusters immediately after boot
2137 Description: LNET request buffers exhausted under heavy short-term load
2138 Details : If a large number of client requests are generated on a service
2139 that has previously never seen so many requests it is possible
2140 that the request buffer growth cannot keep up with the spike in
2141 demand. Instead of dropping incoming requests, they are held in
2142 the LND until the RPC service can accept more requests.
2145 Frequency : Sometimes during replay
2147 Description: Assertion failure in ll_local_open after replay.
2148 Details : If replay happened on an open request reply before we were able
2149 to set replay handler, reply will become not swabbed tripping the
2150 assertion in ll_local_open. Now we set the handler right after
2151 recognising of open request
2154 Frequency : very rare
2156 Description: kernel reports "badness in vsnprintf"
2157 Details : Reading from the "recovery_status" /proc file in small chunks
2158 may cause a negative length in lprocfs_obd_rd_recovery_status()
2159 call to vsnprintf() (which is otherwise harmless). Exit early
2160 if there is no more space in the output buffer.
2162 Severity : enhancement
2164 Description: clear OBD RPC statistics by writing to them
2165 Details : It is now possible to clear the OBD RPC statistics by writing
2166 to the "stats" file.
2171 Description: Client mtime is not the same on different clients after utimes
2172 Details : In some cases, the client was using the utimes() syscall on
2173 a file cached on another node. The clients now validate the
2174 ctime from the MDS + OSTs to determine which one is right.
2179 Description: Inability to activate failout mode
2180 Details : lconf script incorrectly assumed that in pythong string's numeric
2181 value is used in comparisons.
2184 Frequency : always with multiple stripes per file
2186 Description: Inefficient object allocation for mutli-stripe files
2187 Details : When selecting which OSTs to stripe files over, for files with
2188 a stripe count that divides evenly into the number of OSTs,
2189 the MDS is always picking the same starting OST for each file.
2190 Return the OST selection heuristic to the original design.
2195 Description: mount failures may take full timeout to return an error
2196 Details : Under some heavy load conditions it is possible that a
2197 failed mount can wait for the full obd_timeout interval,
2198 possibly several minutes, before reporting an error.
2199 Instead return an error as soon as the status is known.
2201 Frequency : quota enabled and large files being deleted
2203 Description: releasing more than 4GB of quota at once hangs OST
2204 Details : If a user deletes more than 4GB of files on a single OST it
2205 will cause the OST to spin in an infinite loop. Release
2206 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2211 Description: statfs data retrieved from /proc may be stale or zero
2212 Details : When reading per-device statfs data from /proc, in the
2213 {kbytes,files}_{total,free,avail} files, it may appear
2214 as zero or be out of date.
2217 Frequency : systems with MD RAID1 external journal devices
2219 Description: lconf's call to blkid is confused by RAID1 journal devices
2220 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2221 of returning all block devices that match the journal UUID.
2224 Frequency : always, for aggregate stripe size over 4GB
2226 Description: assertion fails when trying to use 4GB stripe size
2227 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2228 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2231 Frequency : always on ppc64
2233 Description: the first write on an ext3 filesystem with mballoc got stuck
2234 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2235 perform endianness conversion.
2237 ------------------------------------------------------------------------------
2239 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2241 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2242 INTEROPERATE with older versions automatically. Please read the
2243 user documentation before upgrading any part of a live system.
2244 * WARNING: Lustre networking configuration changes are required with
2245 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2248 * Support for kernels:
2249 2.6.9-22.0.2.EL (RHEL 4)
2250 2.6.5-7.244 (SLES 9)
2251 2.6.12.6 vanilla (kernel.org)
2254 Severity : enhancement
2255 Bugzilla : 7981/8208
2256 Description: Introduced Lustre Networking (LNET)
2257 Details : LNET is new networking infrastructure for Lustre, it includes
2258 a reorganized network configuration mode (see the user
2259 documentation for full details) as well as support for routing
2260 between different network fabrics. Lustre Networking Devices
2261 (LNDs) for the supported network fabrics have also been
2262 created for this new infrastructure.
2264 Severity : enhancement
2265 Description: Introduced Access control lists
2266 Details : clients can set ACLs on files and directories in order to have
2267 more fine-grained permissions than the standard Unix UGO+RWX.
2268 The MDS must be started with the "-o acl" mount option.
2270 Severity : enhancement
2271 Description: Introduced filesystem quotas
2272 Details : Administrators may now establish per-user quotas on the
2275 Severity : enhancement
2277 Description: Configuration change for the XT3
2278 The PTLLND is now used to run Lustre over Portals on the XT3
2279 The configure option(s) --with-cray-portals are no longer used.
2280 Rather --with-portals=<path-to-portals-includes> is used to
2281 enable building on the XT3. In addition to enable XT3 specific
2282 features the option --enable-cray-xt3 must be used.
2287 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2288 Details : A race between allocating a new llcd and re-getting the llcd_lock
2289 allowed another thread to grab newly-allocated llcd.
2291 Severity : enhancement
2293 Description: 2.6 OST async journal commit and locking fix to improve performance
2294 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2295 2.6 kernels are now async as they already were in 2.4 kernels so
2296 that they can commit concurrently with the network bulk transfer.
2297 For block-allocated files the filter allocation semaphore is held
2298 to avoid filesystem fragmentation during allocation. BKL lock
2299 removed for 2.6 xattr operations where it is no longer needed.
2304 Description: lconf incorrectly determined whether two IP networks could talk
2305 Details : In some more complicated routing and multiple-network
2306 configurations, lconf will avoid trying to make a network
2307 connection to a disjoint part of the IP space. It was doing the
2308 math incorrectly for one set of cases.
2313 Description: Fix for potential infinite loop processing records in an llog.
2314 Details : If an llog record is corrupted/zeroed, it is possible to loop
2315 forever in llog_process(). Validate the llog record length
2316 and skip the remainder of the block on error.
2319 Frequency : occasional (liblustre only)
2321 Description: liblustre could not open files whose last component is a symlink
2322 Details : sysio_path_walk() would incorrectly pass the open intent to
2323 intermediate path components.
2326 Frequency : rare (liblustre only with non-standard tuning)
2327 Bugzilla : 7201 (7350)
2328 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2329 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2330 after it has been released, only do this for referenced locks
2332 Severity : enhancement
2334 Description: specifying an (invalid) directory default stripe_size of -1
2335 would reset the directory default striping
2336 Details : stripe_size -1 was used internally to signal directory stripe
2337 removal, now use "all default" to signal dir stripe removal
2338 as a directory striping of "all default" is not useful
2341 Frequency : common for large clusters running liblustre clients
2343 Description: doing an ls when liblustre clients are running is slow
2344 Details : sending a glimpse AST to a liblustre client waits for every AST
2345 to time out, as liblustre clients will not respond. Since they
2346 cannot cache data we refresh the OST lock LVB from disk instead.
2348 Severity : enhancement
2350 Description: doing an ls at the same time as file IO can be slow
2351 Details : enqueue and other "small" requests can be blocked behind many
2352 large IO requests. Create a new OST IO portal for non-IO
2353 requests so they can be processed faster.
2356 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2358 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2359 created file with mode 000
2360 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2361 allow an ftruncate() as a non-root user to a file with mode 000.
2362 Lustre now respects this flag to disable mode checks when
2363 truncating a file owned by the user
2366 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2369 Description: Revoking locks from clients that went dead or catatonic might take
2371 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2372 cancellation of such locks instant on servers without waiting for
2373 any reply from clients. Clients drops these locks when cancel
2374 notification from server is received without replying.
2377 Frequency : liblustre-only, when liblustre client dies or becomes busy
2379 Description: Doing ls on Linux clients can take a long time with active
2381 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2382 granting such locks to it in the first place if possible. Locks
2383 are taken by proxy on the OST during the read or write and
2384 dropped immediately afterward. Add connect flags handling, do
2385 not grant locks to liblustre clients for glimpse ASTs.
2387 Severity : enhancement
2389 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2390 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2391 sequential reads and behaves badly with applications doing random
2392 reads. Improve it by reading ahead at least read region, and
2393 avoiding excessive large RPC for small reads.
2395 Severity : enhancement
2397 Description: Creating more than 1000 files for a single job may cause a load
2398 imbalance on the OSTs if there are also a large number of OSTs.
2399 Details : qos_prep_create() uses an OST index reseed value that is an
2400 even multiple of the number of available OSTs so that if the
2401 reseed happens in the middle of the object allocation it will
2402 still utilize the OSTs as uniformly as possible.
2407 Description: OST or MDS may oops in ping_evictor_main()
2408 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2409 but doesn't restart at beginning of obd_exports_timed list
2412 Severity : enhancement
2414 Description: improve by-nid export eviction on the MDS and OST
2415 Details : allow multiple exports with the same NID to be evicted at one
2416 time without re-searching the exports list.
2419 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2421 Description: MDS may oops in groups_free()
2422 Details : in rare race conditions a newly allocated group_info struct is
2423 freed again, and this can be NULL. The 2.4 compatibility code
2424 for groups_free() checked for a NULL pointer, but 2.6 did not.
2427 Frequency : common for liblustre clients doing little filesystem IO
2428 Bugzilla : 9352, 7313
2429 Description: server may evict liblustre clients accessing contended locks
2430 Details : if a client is granted a lock or receives a completion AST
2431 with a blocking AST already set it would not reply to the AST
2432 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2435 Frequency : lfs setstripe, only systems with more than 160 OSTs
2437 Description: unable to set striping with a starting offset beyond OST 160
2438 Details : llapi_create_file() incorrectly limited the starting stripe
2439 index to the maximum single-file stripe count.
2442 Frequency : LDAP users only
2444 Description: lconf did not handle in-kernel recovery with LDAP properly
2445 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2447 Severity : enhancement
2449 Description: bind OST threads to NUMA nodes to improve performance
2450 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2451 node and do their allocations there to localize memory access
2453 Severity : enhancement
2455 Description: llmount can determine client NID directly from Myrinet (GM)
2456 Details : the client NID code from gmnalnid was moved directly into
2457 llmount, removing the need to use this or specifying the
2458 client NID explicitly when mounting GM clients with zeroconf
2461 Frequency : if client is started with down MDS
2463 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2464 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2465 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2466 ptlrpc_import_delayed_req() because we didn't check if the
2467 request was interrupted, and we also didn't break out of the
2468 event loop if there was no timeout
2473 Description: data loss during non-page-aligned writes to a single file from
2474 both multiple nodes and multiple threads on one node at same time
2475 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2476 inconsistency led to false short-reads, that were cached and later
2477 used by ->prepare_write() to fill in partially written page,
2478 leading to data loss.
2481 Frequency : always, if lconf --abort_recovery used
2483 Description: lconf --abort_recovery fails with 'Operation not supported'
2484 Details : lconf was attempting to abort recovery on the MDT device and not
2487 Severity : enhancement
2489 Description: remove cleanup logs
2490 Details : replace lconf-generated cleanup logs with lustre internal
2491 cleanup routines. Eliminates the need for client-cleanup and
2494 Severity : enhancement
2496 Description: add support for EAs (user and system) on lustre filesystems
2497 Details : it is now possible to store extended attributes in the Lustre
2498 client filesystem, and with the user_xattr mount option it
2499 is possible to allow users to store EAs on their files also
2501 Severity : enhancement
2503 Description: Add possibility (config option) to show minimal available OST free
2505 Details : When compiled with --enable-mindf configure option, statfs(2)
2506 (and so, df) will return least minimal free space available from
2507 all OSTs as amount of free space on FS, instead of summary of
2508 free spaces of all OSTs.
2510 Severity : enhancement
2512 Description: do not expand extent locks acquired on OST-side
2513 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2514 by server: they are not cached anyway.
2517 Frequency : when mmap is used/binaries executed from Lustre
2519 Description: Unmmap pages before throwing them away from read cache.
2520 Details : llap_shrink cache now attempts to unmap pages before discarding
2521 them (if unmapping failed - do not discard). SLES9 kernel has
2522 extra checks that trigger if this unmapping is not done first.
2527 Description: lconf didn't resolve symlinks before checking to see whether a
2528 given mountpoint was already in use
2531 Frequency : when migrating failover services
2532 Bugzilla : 6395, 9514
2533 Description: When migrating a subset of services from a node (e.g. failback
2534 from a failover service node) the remaining services would
2535 time out and evict clients.
2536 Details : lconf --force (implied by --failover) sets the global obd_timeout
2537 to 5 seconds in order to quickly disconnect, but this caused
2538 other RPCs to time out too quickly. Do not change the global
2539 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2541 Severity : enhancement
2542 Frequency : if MDS is started with down OST
2543 Bugzilla : 9439,5706
2544 Description: Allow startup/shutdown of an MDS without depending on the
2545 availability of the OSTs.
2546 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2547 Add appropriate locking and lov-osc refcounts for safe
2548 cleaning. Add osc abort_inflight calls in case the
2549 synchronize never started.
2552 Frequency : occasional (Cray XT3 only)
2554 Description: root not authorized to access files in CRAY_PORTALS environment
2555 Details : The client process capabilities were not honoured on the MDS in
2556 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2557 been accessed by an authorized user then root was able to access
2558 the file on the local client also. The root user capabilities
2559 are now allowed on the MDS, as this environment has secure UID.
2562 Frequency : occasional
2564 Description: ldiskfs "too long searching" message happens too often
2565 Details : A debugging message (otherwise harmless) prints too often on
2566 the OST console. This has been reduced to only happen when
2567 there are fragmentation problems on the filesystem.
2572 Description: Division by zero in statfs when all OSCs are inactive
2573 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2574 lov_statfs divides by value returned by lov_get_stripecnt().
2578 Bugzilla : 9489, 3273
2579 Description: First write from each client to each OST was only 4kB in size,
2580 to initialize client writeback cache, which caused sub-optimal
2581 RPCs and poor layout on disk for the first writen file.
2582 Details : Clients now request an initial cache grant at (re)connect time
2583 and so that they can start streaming writes to the cache right
2584 away and always do full-sized RPCs if there is enough data.
2585 If the OST is rebooted the client also re-establishes its grant
2586 so that client cached writes will be honoured under the grant.
2591 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2592 Details : Now I/O RPCs go to different portal number and (presumably) fast
2593 lock requests (and glimses) and other RPCs get their own service
2594 threads pool that should be able to service those RPCs
2597 Severity : enhancement
2599 Description: Ability to exchange lustre version between client and servers and
2600 issue warnings at client side if client is too old. Also for
2601 liblustre clients there is ability to refuse connection of too old
2603 Details : New 'version' field is added to connect data structure that is
2604 filled with version info. That info is later checked by server and
2608 Frequency : rare, liblustre only.
2609 Bugzilla : 9296, 9581
2610 Description: Two simultaneous writes from liblustre at offset within same page
2611 might proceed at the same time overwriting eachother with stale
2613 Details : I/O lock withing llu_file_prwv was released too early, before data
2614 actually was hitting the wire. Extended lock-holding time until
2615 server acknowledges receiving data.
2618 Frequency : extremely rare. Never observed in practice.
2620 Description: avoid generating lustre_handle cookie of 0.
2621 Details : class_handle_hash() generates handle cookies by incrementing
2622 global counter, and can hit 0 occasionaly (this is unlikely, but
2623 not impossible, because initial value of cookie counter is
2624 selected randonly). Value of 0 is used as a sentinel meaning
2625 "unassigned handle" --- avoid it. Also coalesce two critical
2626 sections in this function into one.
2628 Severity : enhancement
2630 Description: allow liblustre clients to delegate truncate locking to OST
2631 Details : To avoid overhead of locking, liblustre client instructs OST to
2632 take extent lock in ost_punch() on client's behalf. New connection
2633 flag is added to handle backward compatibility.
2635 Severity : enhancement
2636 Bugzilla : 4928, 7341, 9758
2637 Description: allow number of OST service threads to be specified
2638 Details : a module parameter allows the number of OST service threads
2639 to be specified via "options ost ost_num_threads={N}" in the
2640 OSS's /etc/modules.conf or /etc/modprobe.conf.
2644 Bugzilla : 6146, 9635, 9895
2645 Description: servers crash with bad pointer in target_handle_connect()
2646 Details : In rare cases when a client is reconnecting it was possible that
2647 the connection request was the last reference for that export.
2648 We would temporarily drop the export reference and get a new
2649 one, but this may have been the last reference and the export
2650 was just destroyed. Get new reference before dropping old one.
2652 Severity : enhancement
2653 Frequency : if client is started with failover MDS
2655 Description: Allow multiple MDS hostnames in the mount command
2656 Details : Try to read the configuration from all specified MDS
2657 hostnames during a client mount in case the "primary"
2660 Severity : enhancement
2662 Description: Stop sending data to evicted clients as soon as possible.
2663 Details : Check if the client we are about to send or are sending data to
2664 was evicted already. (Check is done every second of waiting,
2665 for which l_wait_event interface was extended to allow checking
2666 of exit condition at specified intervals).
2669 Frequency : rare, normally only when NFS exporting is done from client
2671 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2673 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2676 Frequency : always, if srand() is called before liblustre initialization
2678 Description: Liblustre uses system PRNG disturbing its usage by user application
2679 Details : Introduce internal to lustre fast and high-quality PRNG for
2680 lustre usage and make liblustre and some other places in generic
2681 lustre code to use it.
2683 Severity : enhancement
2684 Bugzilla : 9477, 9557, 9870
2685 Description: Verify that the MDS configuration logs are updated when xml is
2686 Details : Check if the .xml configuration logs are newer than the config
2687 logs stored on the MDS and report an error if this is the case.
2688 Request --write-conf, or allow starting with --old_conf.
2690 Severity : enhancement
2692 Description: Handle symlinks in the path when checking if Lustre is mounted.
2693 Details : Resolve intermediate symlinks when checking if a client has
2694 mounted a filesystem to avoid duplicate client mounts.
2699 Description: lconf can hit an error exception but still return success.
2700 Details : The lconf command catches the Command error exception at the top
2701 level script context and will exit with the associated exit
2702 status, but doesn't ensure that this exit status is non-zero.
2707 Description: failure of ptlrpc thread startup can cause oops
2708 Details : Starting a ptlrpc service thread can fail if there are a large
2709 number of threads or the server memory is very fragmented.
2710 Handle this without oopsing.
2713 Frequency : always, only if liblustre and non-default acceptor port was used
2715 Description: liblustre cannot connect to servers with non-default acceptor port
2716 Details : tcpnal_set_default_params() was not called and was therefore
2717 ignoring the environment varaible TCPNAL_PORT, as well as other
2718 TCPNAL_ environment variables
2723 Description: two objects could be created on the same OST for a single file
2724 Details : If an OST is down, in some cases it was possible to create two
2725 objects on a single OST for a single file. No problems other
2726 than potential performance impact and spurious error messages.
2730 Bugzilla : 5681, 9562
2731 Description: Client may oops in ll_unhash_aliases
2732 Details : Client dcache may become inconsistent in race condition.
2733 In some cases "getcwd" can fail if the current directory is
2739 Description: Inode refcounting problems in NFS export code
2740 Details : link_raw functions used to call d_instantiate without obtaining
2741 extra inode reference first.
2745 Bugzilla : 9942, 9903
2746 Description: Referencing freed requests leading to crash, memleaks with NFS.
2747 Details : We used to require that call to ll_revalidate_it was always
2748 followed by ll_lookup_it. Also with revalidate_special() it is
2749 possible to call ll_revalidate_it() twice for the same dentry
2750 even if first occurence returned success. This fix changes semantic
2751 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2752 reference on a request referred from the intent.
2753 ll_intent_release() then releases such a request.
2756 Frequency : rare, normally benchmark loads only
2758 Description: unlinked inodes were kept in memory on the client
2759 Details : If a client is repeatedly creating and unlinking files it
2760 can accumulate a lot of stale inodes in the inode slab cache.
2761 If there is no other client load running this can cause the
2762 client node to run out of memory. Instead flush old inodes
2763 from client cache that have the same inode number as a new inode.
2766 Frequency : SLES9 2.6.5 kernel and long filenames only
2767 Bugzilla : 9969, 10379
2768 Description: utime reports stale NFS file handle
2769 Details : SLES9 uses out-of-dentry names in some cases, which confused
2770 the lustre dentry revalidation. Change it to always use the
2774 Frequency : rare, unless heavy write-truncate concurrency is continuous
2775 Bugzilla : 4180, 6984, 7171, 9963, 9331
2776 Description: OST becomes very slow and/or deadlocked during object unlink
2777 Details : filter_destroy() was holding onto the parent directory lock
2778 while truncating+unlinking objects. For very large objects this
2779 may block other threads for a long time and slow overall OST
2780 responsiveness. It may also be possible to get a lock ordering
2781 deadlock in this case, or run out of journal credits because of
2782 the combined truncate+unlink. Solution is to do object truncate
2783 first in one transaction without parent lock, and then do the
2784 final unlink in a new transaction with the parent lock. This
2785 reduces the lock hold time dramatically.
2788 Frequency : rare, 2.4 kernels only
2790 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2791 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2792 before unmounting their filesystem in order to avoid deadlock.
2793 The kernel_locked() function in 2.4 kernels only checks whether
2794 the kernel lock is held, not whether it is this process that is
2795 holding it as 2.6 kernels do.
2800 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2801 Details : The client ptlrpc code may be trying to reconnect to a down
2802 server before a previous connection attempt has timed out.
2803 Increase the reconnect interval to be longer than the connection
2804 timeout interval to avoid sending duplicate connections to
2808 Frequency : echo_client brw_test command
2810 Description: fix echo_client to work with OST preallocated code
2811 Details : OST preallocation code (5137) didn't take echo_client IO path
2812 into account: echo_client calls filter methods outside of any
2813 OST thread and, hence, there is no per-thread preallocated
2814 pages and buffers to use. Solution: hijack pga pages for IO. As
2815 a byproduct, this avoids unnecessary data copying.
2819 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2820 Description: Client can oops in mdc_commit_close() after open replay
2821 Details : It was possible for the MDS to return an open request with no
2822 transaction number in mds_finish_transno() if the client was
2823 evicted, but without actually returning an error. Clients
2824 would later try to replay that open and may trip an assertion
2825 Simplify the client close codepath, and always return an error
2826 from the MDS in case the open is not successful.
2829 Frequency : rare, 2.6 OSTs only
2831 Description: OST may deadlock under high load on fragmented files
2832 Details : If there was a heavy load and highly-fragmented OST filesystems
2833 it was possible to have all the OST threads deadlock waiting on
2834 allocation of biovecs, because the biovecs were not released
2835 until the entire RPC IO was completed. Instead, release biovecs
2836 as soon as they are complete to ensure forward IO progress.
2838 Severity : enhancement
2840 Description: Support for specifying external journal device at mount
2841 Details : If an OST or MDS device is formatted with an external journal
2842 device, this device major/minor is stored in the ext3 superblock
2843 and may not be valid for failover. Allow detecting and
2844 specifying the external journal at mount time.
2849 Description: Mounting an MDS with pending unlinked files may cause oops
2850 Details : target_finish_recovery() calls mds_postrecov() which returned
2851 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2852 considers this an error and immediately begins cleaning up the
2853 lov, just after starting the mds_lov process
2855 Severity : enhancement
2857 Description: Implement 'lfs df' to report actual free space on per-OST basis
2858 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2859 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2860 usage of objects; '-h' to report in human readable format.
2862 ------------------------------------------------------------------------------
2864 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2871 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2872 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2873 at startup during orphan unlinking. Instead, skip buddy update
2874 before setup, as it will be regenerated after recovery is complete.
2879 Description: If an OST is inactive, its locks might reference stale inodes.
2880 Details : lov_change_cbdata() must iterate over all namespaces, even if
2881 they are inactive to clear inode references from the lock.
2883 Severity : enhancement
2884 Frequency : occasional, if non-standard max_dirty_mb used
2886 Description: Client will block write RPCs if not enough grant
2887 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2888 then the client will block writes while waiting for another RPC
2889 to complete instead of consuming its dirty limit. With change
2890 we get improved performance when max_dirty_mb is small.
2892 Severity : enhancement
2893 Bugzilla : 3389, 6253
2894 Description: Add support for supplementary groups on the MDS.
2895 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2896 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2897 lookups for user supplementary groups into a cache.
2901 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2902 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2903 for liblustre O_EXCL is enforced by MDS.
2907 Description: readdir never returns NULL in liblustre.
2908 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2909 dentry in struct dirent.
2913 Frequency : liblustre only, depends on application IO pattern
2914 Description: liblustre clients evicted if not contacting servers
2915 Details : Don't put liblustre clients into the ping_evictor list, so
2916 they will not be evicted by the pinger ever.
2918 Severity : enhancement
2920 Description: Add ability to evict clients by NID from MDS.
2921 Details : By echoing "nid:$NID" string into
2922 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2923 $NID would be instantly evicted from this MDS and from all active
2924 OSTs connected to it.
2928 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2929 Details : lookup_it_finish() used to query file size from OSTs that was not
2934 Description: service threads change working directory to that of init
2935 Details : Starting lustre service threads may pin the working directory
2936 of the parent thread, making that filesystem busy. Threads
2937 now change to the working directory of init to avoid this.
2941 Frequency : during shutdown only
2942 Description: shutdown with a failed MDS or OST can cause unmount to hang
2943 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2948 Frequency : 2.6 only
2949 Description: chmod/chown may include an extra supplementary group
2950 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2951 supplementary group and if none is specified this is used.
2954 Bugzilla : 5479 (6816)
2956 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2957 Details : If lookup is in progress on a file that is unlinked we might try
2958 to revalidate the inode and fail in revalidate after lookup is
2959 complete and ll_file_open() enqueues the open again but
2960 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2963 Frequency : always, if lconf --abort_recovery used
2965 Description: lconf --abort_recovery fails with 'Operation not supported'
2966 Details : lconf was attempting to abort recovery on the MDT device and not
2969 ------------------------------------------------------------------------------
2971 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2976 Frequency : rare (only unsupported configurations with a node running as an
2978 Bugzilla : 6514, 5137
2979 Description: Mounting a Lustre file system on a node running as an OST could
2981 Details : OSTs now preallocates memory needed to write out data at
2982 startup, instead of when needed, to avoid having to
2983 allocate memory in possibly low memory situations.
2984 Specifically, if the file system is mounted on on OST,
2985 memory pressure could force it to try to write out data,
2986 which it needed to allocate memory to do. Due to the low
2987 memory, it would be unable to do so and the node would
2988 become unresponsive.
2990 Severity : enhancement
2992 Description: Addition of lconf --service command line option
2993 Details : lconf now accepts a '--service <arg>' option, which is
2994 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2996 Severity : enhancement
2998 Description: Failover mode is now the default for OSTs.
2999 Details : By default, OSTs will now run in failover mode. To return to
3000 the old behaviour, add '--failout' to the lmc line for OSTs.
3002 Severity : enhancement
3004 Description: Health checks are now provided for MDS and OSTs
3005 Details : Additional detailed health check information on MSD and OSTs
3006 is now provided through the procfs health_check value.
3009 Frequency : occasional, depends on IO load
3011 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3012 after numerous create/delete cycles
3013 Details : The ext3 inode allocation policy would not allocate new inodes
3014 very well on the OSTs because there are no new directories
3015 being created. Instead we look for groups with free space if
3016 the parent directories are nearly full.
3021 Description: Network or server problems during mount may cause partially
3022 mounted clients instead of returning an error.
3023 Details : The config llog parsing code may overwrite the error return
3024 code during mount error handling, returning success instead
3030 Description: MDS can fail to allocate large reply buffers
3031 Details : After long uptimes the MDS can fail to allocate large reply
3032 buffers (e.g. zconf client mount config records) due to memory
3033 fragmentation or consumption by the buffer cache. Preallocate
3034 some large reply buffers so that these replies can be sent even
3035 under memory pressure.
3039 Frequency : rare (liblustre)
3040 Description: fsx running with liblustre complained that using truncate() to
3041 extend the file doesn't work. This patch corrects that issue.
3042 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3043 ATTR_SIZE and lsm use in llu_setattr_raw.
3047 Frequency : rare, only 2.6 kernels
3048 Description: Unusual file access patterns on the MDS may result in inode
3049 data being lost in very rare circumstances.
3050 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3051 code in the 2.6 kernel caused the inode and/or EA data not to
3052 be read from disk, causing single-file corruption.
3056 Frequency : rare, only 2.6 filesystems using extents
3057 Description: Heavy concurrent write and delete load may cause data corruption.
3058 Details : It was possible under high-load situations to have an extent
3059 metadata block in the block device cache from a just-unlinked
3060 file overwrite a newly-allocated data block. We now unmap any
3061 metadata buffers that alias just-allocated data blocks.
3065 Frequency : filesystems with default stripe_count larger than 77
3066 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3067 Details : lconf specifies an inode size of 4096 bytes when the default
3068 stripe_count is larger than 77. This conflicts with the default
3069 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3070 this case to avoid pinning too much memory for large EAs.
3072 ------------------------------------------------------------------------------
3074 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3079 Frequency : rare (extremely heavy IO load with hundreds of clients)
3081 Description: Client is evicted, gets IO error writing to file
3082 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3083 caused clients to be evicted for AST timeouts. The fixes in
3084 bug 5192 mean we no longer need to have such short AST timeouts
3085 so ldlm_timeout has been increased.
3088 Frequency : occasional during --force or --failover shutdown under load
3089 Bugzilla : 5949, 4834
3090 Description: Server oops/LBUG if stopped with --force or --failover under load
3091 Details : a collection of import/export refcount and cleanup ordering
3092 issues fixed for safer force cleanup
3095 Frequency : only filesystems larger than 120 OSTs
3096 Bugzilla : 5990, 6223
3097 Description: lfs getstripe would oops on a very large filesystem
3098 Details : lov_getconfig used kfree on vmalloc'd memory
3101 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3102 Bugzilla : 6242, 6243
3103 Description: reading from files that had been truncated to a non-zero size
3104 but never opened returned no data
3105 Details : ll_file_read() reads zeros from no-object files to EOF
3110 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3111 Details : There's little harm in aborting MDS/OSS recovery and letting it
3112 try again, so I removed the LASSERT and return an error instead.
3114 Severity : enhancement
3116 Description: New debugging infrastructure for tracking down data corruption
3117 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3118 (b) cover more of the client-side code path, and (c) try to narrow
3119 down where problems occurred
3123 Bugzilla : 3819, 4364, 4397, 6313
3124 Description: Racing close and eviction MDS could cause assertion in mds_close
3125 Details : It was possible to get multiple mfd references during close and
3126 client eviction, leading to one thread referencing a freed mfd.
3128 Severity: : enhancement
3129 Bugzilla : 3262, 6359
3130 Description: Attempts to reconnect to servers are now more aggressive.
3131 Details : This builds on the enhanced upcall-less recovery that was added
3132 in 1.4.2. When trying to reconnect to servers, clients will
3133 now try each server in the failover group every 10 seconds. By
3134 default, clients would previously try one server every 25 seconds.
3139 Description: After recovery, certain operations trigger a failed
3140 assertion on a client.
3141 Details : Failing over an mds, using lconf -d --failover, while a
3142 client was doing a readdir() call would cause the client to
3143 LBUG after recovery completed and the readdir() was resent.
3145 Severity : enhancement
3147 Description: Default groups are now added by lconf
3148 Details : You can now run lconf --group <servicename> without having to
3149 manually add groups with lmc.
3152 Frequency : occasional
3154 Description: Nodes with an elan id of 0 trigger a failed assertion
3157 Frequency : always when accessing e.g. tty/console device nodes
3159 Description: tty and some other devices nodes cannot be used on lustre
3160 Details : file's private_data field is used by device data and lustre
3161 values in there got lost. New field was added to struct file to
3162 store fs-specific private data.
3165 Frequency : when exporting Lustre via NFS
3167 Description: NFSD failed occasionally when looking up a path component
3168 Details : NFSD is looking up ".." which was broken in ext3 directories
3169 that had grown large enough to become hashed.
3172 Frequency : Clusters with multiple interfaces not on the same subnet
3174 Description: Nodes will repeatedly try to reconnect to an interface which it
3175 cannot reach and report an error to the log.
3176 Details : Extra peer list entries will be created by lconf with some peers
3177 unreachable. lconf now validates the peer before adding it.
3180 Frequency : Only if a default stripe is set on the filesystem root.
3182 Description: Setting a default stripe on the filesystem root prevented the
3183 filesystem from being remounted.
3184 Details : The client was sending extra request flags in the root getattr
3185 request and did not allocate a reply buffer for the dir EA.
3188 Frequency : occasional, higher if lots of files are accessed by one client
3189 Bugzilla : 6159, 6097
3190 Description: Client trips assertion regarding lsm mismatch/magic
3191 Details : While revalidating inodes the VFS looks up inodes with ifind()
3192 and in rare cases can find an inode that is being freed.
3193 The ll_test_inode() code will free the lsm during ifind()
3194 when it finds an existing inode and then the VFS later attaches
3195 this free lsm to a new inode.
3199 Bugzilla : 6422, 7030
3200 Description: MDS deadlock between mkdir and client eviction
3201 Details : Creating a new file via mkdir or mknod (starting a transaction
3202 and getting the ns lock) can deadlock with client eviction
3203 (gets ns lock and trying to finish a synchronous transaction).
3206 Frequency : occasional
3207 Description: While starting a server, the fsfilt_ext3 module could not be
3209 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3210 kernels. Previously, lconf would still use the ext3 name
3211 when trying to load modules. Now, it will correctly use
3212 ext3 on 2.4 and ldiskfs on 2.6.
3214 Severity : enhancement
3215 Description: The default stripe count has been changed to 1
3216 Details : The interpretation of the default stripe count (0, to lfs
3217 or lmc) has been changed to mean striping across a single
3218 OST, rather than all available. For general usage we have
3219 found a stripe count of 1 or 2 works best.
3221 Severity : enhancement
3222 Description: Add support for compiling against Cray portals.
3223 Details : Conditional compiling for some areas that are different
3227 Frequency : occasional
3228 Bugzilla : 6409, 6834
3229 Description: Creating files with an explicit stripe count may lead to
3230 a failed assertion on the MDS
3231 Details : If some OSTs are full or unavailable, creating files may
3232 trigger a failed assertion on the MDS. Now, Lustre will
3233 try to use other servers or return an error to the
3237 Frequency : occasional
3239 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3240 caused lock timeout during readahead (since 1.4.2).
3241 Details : Processes doing readahead might match a lock that hasn't been
3242 granted yet if there are overlapping and conflicting lock
3243 requests. The readahead process waits on ungranted lock
3244 (original lock is CBPENDING), while OST waits for that process
3245 to cancel CBPENDING read lock and eventually evicts client.
3247 Severity : enhancement
3249 Description: Initial enabling of flock support for clients
3250 Details : Implements fcntl advisory locking and file status functions.
3251 This feature is provided as an optional mount flag (default
3252 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3253 locking are implemented yet, and those that are are not guaranteed
3254 to be completely correct in production environments.
3255 mount -t lustre -o [flock|noflock] ...
3258 Frequency : occasional
3260 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3261 assertion in the ext3 JBD (journaling) layer.
3262 Details : The b_committed_data struct is protected by the big kernel lock
3263 in 2.4 kernels, serializing journal_commit_transaction() and
3264 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3265 access to this struct. In 2.6 kernels there is finer grained
3266 locking to improve SMP performance of the JBD layer.
3270 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3275 Description: merge and cleanup kernel patches.
3278 -----------------------------------------------------------------------------
3280 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3282 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3283 versions of Lustre. Please update all servers and clients to
3284 version 1.4.2 or later at the same time. You must also run
3285 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3286 to update the configuration logs.
3288 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3289 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3290 - swab llog records of type '0' so we get proper header size/idx (5861)
3291 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3292 - fix rename of one directory over another leaking an inode (5953)
3293 - avoid SetPageDirty on 2.6 (5981)
3294 - don't re-add just-being-destroyed locks to the waiting list (5653)
3295 - when creating new directories, inherit the parent's custom
3296 striping settings if present parent (3048)
3297 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3298 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3299 - don't hold client locks on temporary worklist from l_lru (5666)
3300 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3301 - automatically evict dead clients (5921)
3302 - Update file size properly in create+truncate+fstat case (6196)
3303 - Do not unhash mountpoint dentries, do not allow removal of
3305 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3306 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3307 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3308 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3309 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3310 - Correctly handle reads of files with no objects (6243)
3311 - lctl recover will also mark a device active if deactivate used (5933)
3313 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3314 - allow --write-conf on an MDS with different nettype than client (5619)
3315 - don't write config llogs to MDS for mounts not from that MDS (5617)
3316 - lconf should create multiple TCP connections from a client (5201)
3317 - init scripts are now turned off by default; run chkconfig --on
3318 lustre and chkconfig --on lustrefs to use them
3319 - upcalls are no longer needed for clients to recover to failover
3321 - add --abort-recovery option to lconf to abort recovery on device
3323 - add support for an arbitrary number of OSTs (3026)
3324 - Quota support protocol changes.
3325 - forward compatibility changes to wire structs (6007)
3326 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3327 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3328 - improved "lustre status" script
3329 - initialize blocksize for non-regular files (6062)
3330 - added --disable-server and --disable-client configure options (5782)
3331 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3332 - Vanilla 2.4.29 support
3333 - increase maximum number of obd devices to 520 (6242)
3334 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3335 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3336 SLES 9 2.6.5 kernels
3337 - limit stripes per file to 160 (the maximum EA size) (6093)
3339 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3342 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3343 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3344 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3345 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3346 - small ext3 extents cleanups and fixes (5733)
3347 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3348 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3349 - avoid needless client->OST connect, fix handle mismatch (5317)
3350 - fix DLM error path that led to out-of-sync client, long delays (5779)
3351 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3352 - fix several locking issues related to i_size (5492,5624,5654,5672)
3353 - don't move pending lock onto export if it is already evicted (5683)
3354 - fix kernel oops when creating .foo in unlinked directory (5548)
3355 - fix deadlock in obdfilter statistics vs. object create (5811)
3356 - use time_{before,after} to avoid timer jiffies wrap (5882)
3357 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3358 - Do not leak request if server was not able to process it (5154)
3359 - If mds_open unable to find parent dir, make that negative lookup(5154)
3360 - don't create new directories with extent-mapping (5909, 5936)
3362 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3363 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3364 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3365 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3366 - support for RHEL4 kernel on i686 (5773)
3367 - provide error messages when incompatible logs are encountered (5898)
3369 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3370 * version 1.4.0.10 (1.4.1 release candidate 1)
3372 - don't keep a lock reference when lock is not granted (4238)
3373 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3374 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3375 - drop import inflight refcount on signal_completed_replay error (5255)
3376 - unlock page after async write error during send (3677)
3377 - handle missing objects in filter_preprw_read properly (5265)
3378 - no transno return for symlink open, don't save no-trasno open (3440)
3379 - don't try to complete elan receive that already failed (4012)
3380 - free RPC server reply state on error (5406)
3381 - clean up thread from ptlrpc_start_thread() on error (5160)
3382 - readahead could read extra page into cache that wasn't ejected (5388)
3383 - prevent races in class_attach/setup/cleanup/detach (5260)
3384 - don't dereference de->d_inode after l_dput of de (5458)
3385 - use "int" for stripe value returned from lock_to_stripe (5544)
3386 - mballoc allocation and error-checking fixes in 2.6 (5504)
3387 - block device patches to fix I/O request sizes in 2.6 (5482)
3388 - look up hostnames for IB nals (5602)
3389 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3390 - don't start multiple acceptors for the same port (5277)
3391 - fix incorrect LASSERT in mds_getattr_name (5635)
3392 - export a proc file for general "ping" checking (5628)
3393 - fix "lfs check" to not block when the MDS is down (5628)
3395 - service request history (4965)
3396 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3397 - create an "evict_client" /proc entry on OSTs, like the MDS has
3398 - fix mount usage message, return errors per mount(8) (5168)
3399 - change grep [] to grep "[]" in tests so they work in more UMLs
3400 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3401 - remove ancient LOV_MAGIC_V0 EA support (5047)
3402 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3403 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3404 - allow readahead allocations to fail when low on memory (5383)
3405 - mmap locking landed again, after considerable improvement (2828)
3406 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3408 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3411 - send OST transaction number in read/write reply to free req (4966)
3412 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3413 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3414 - the watchdog thread now runs as interruptible (5246)
3415 - flock/lockf fixes (but it's still disabled, pending 5135)
3416 - don't use EXT3 constants in llite code (5094)
3417 - memory shortage at startup could cause assertion (5176)
3419 - reorganization of lov code
3420 - single portals codebase
3422 - add extents/mballoc support (5025)
3423 - direct I/O reads in the obdfilter (4048)
3424 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3426 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3429 - send OST transaction number in read/write reply to free req (4966)
3430 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3431 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3432 - the watchdog thread now runs as interruptible (5246)
3433 - handle missing objects in filter_preprw_read properly (5265)
3434 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3435 - drop import inflight refcount on signal_completed_replay error (5255)
3436 - unlock page after async write error during send (3677)
3437 - return original error code on reconstructed replies (3761)
3438 - no transno return for symlink open, don't save no-trasno open (3440)
3440 - add pid to ldlm debugging output (4922)
3441 - bump the watchdog timeouts -- we can't handle 30sec yet
3442 - extra debugging for orphan dentry/inode bug (5259)
3444 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3447 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3448 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3449 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3450 - fix NULL dereference in /proc/sys/portals/routes (4827)
3451 - allow failed mdc_close() operations to be interrupted (4561)
3452 - stop precreate on OST before MDS would time out on it (4778)
3453 - don't send partial-page writes before EOF from client (4410)
3454 - discard client grant for sub-page writes on large-page clients (4520)
3455 - don't free dentries not owned by NFS code, check generation (4806)
3456 - fix lsm leak if mds_create_objects() fails (4801)
3457 - limit debug_daemon file size, always print CERROR messages (4789)
3458 - use transno after validating reply (3892)
3459 - process timed out requests if import state changes (3754)
3460 - update mtime on OST during writes, return in glimpse (4829)
3461 - add mkfsoptions to LDAP (4679)
3462 - use ->max_readahead method instead of zapping global ra (5039)
3463 - don't interrupt __l_wait_event() during strace
3465 - add software watchdogs to catch hung threads quickly (4941)
3466 - make lustrefs init script start after nfs is mounted
3467 - fix CWARN/ERROR duplication (4930)
3468 - return async write errors to application if possible (2248)
3469 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3470 - print NAL number in %x format (4645)
3471 - update barely-supported suse-2.4.21-171 series (4842)
3472 - support for sles 9 %post scripts
3473 - support for building 2.6 kernel-source packages
3474 - support for sles km_* packages
3476 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3479 - ignore -ENOENT errors in osc_destroy (3639)
3480 - notify osc create thread that OSC is being cleaned up (4600)
3481 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3482 - reconstruct ost_handle() like mds_handle() (4657)
3483 - create a new thread to do import eviction to avoid deadlock (3969)
3484 - let lconf resolve symlinked-to devices (4629)
3485 - don't unlink "objects" from directory with default EA (4554)
3486 - hold socknal file ref over connect in case target is down (4394)
3487 - allow more than 32000 subdirectories in a single directory (3244)
3488 - fix blocks count for O_DIRECT writes (3751)
3489 - OST returns ENOSPC from object create when no space left (4539)
3490 - don't send truncate RPC if file size isn't changing (4410)
3491 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3492 - bind to privileged port in socknal and tcpnal (3689)
3494 - rate limit CERROR/CWARN console message to avoid overload (4519)
3495 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3496 - basic mmap support (3918)
3497 - kernel patch series update from b1_4 (4711)
3499 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3502 - avoid crash during MDS cleanup with OST shut down (2775)
3503 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3504 - don't use bad inodes on the MDS (3744)
3505 - dynamic object preallocation to improve recovery speed (4236)
3506 - don't hold spinlock over lock dumping or change debug flags (4401)
3507 - don't zero obd_dev when it is force cleaned (3651)
3508 - print grants to console if they go negative (4431)
3509 - "lctl deactivate" will stop automatic recovery attempts (3406)
3510 - look for existing locks in ldlm_handle_enqueue() (3764)
3511 - don't resolve lock handle twice in recovery avoiding race (4401)
3512 - revalidate should check working dir is a directory (4134)
3514 - don't always mark "slow" obdfilter messages as errors (4418)
3516 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3519 - don't close LustreDB during write_conf until it is done (3860)
3520 - fix typo in lconf for_each_profile (3821)
3521 - allow dumping logs from multiple threads at one time (3820)
3522 - don't allow multiple threads in OSC recovery (3812)
3523 - fix debug_size parameters (3864)
3524 - fix mds_postrecov to initialize import for llog ctxt (3121)
3525 - replace config semaphore with spinlock (3306)
3526 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3527 - don't allow enqueue to complete on a destroyed export (3822)
3528 - down write_lock before checking llog header bitmap (3825)
3529 - recover from lock replay timeout (3764)
3530 - up llog sem before sending rpc (3652)
3531 - reduce ns lock hold times when setting kms (3267)
3532 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3533 - fix NULL deref and obd_dev leak on setup error (3312)
3534 - replace some LBUG about llog ops with error handling (3841)
3535 - don't match INVALID dentries from d_lookup and spin (3784)
3536 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3537 - fix invalid assertion in ptlrpc_set_wait (3880)
3539 - add libwrap support for the TCP acceptor (3996)
3540 - add /proc/sys/portals/routes for non-root route listing (3994)
3541 - allow setting MDS UUID in .xml (2580)
3542 - print the stack of a process that LBUGs (4228)
3544 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3547 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3548 - make sure to unset replay flag from failed open requests (3440)
3549 - if default stripe count is 0, use OST count for inode size (3636)
3550 - update parent mtime/ctime on client for create/unlink (2611)
3551 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3552 - free recovery state on server during a forced cleanup (3571)
3553 - unregister_reply for resent reqs (3063)
3554 - loop back devices mounting and status check on 2.6 (3563)
3555 - fix resource-creation race that can provoke i_size == 0 (3513)
3556 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3557 - more debugging for page-accounting assertion (3746)
3558 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3559 - avoid deadlock after precreation failure (3758)
3560 - fix race and lock order deadlock in orphan handling (3450, 3750)
3561 - add validity checks when grabbing inodes from l_ast_data (3599)
3563 - add /proc/.../recovery_status to obdfilter (3428)
3564 - lightweight CDEBUG infrastructure, debug daemon (3668)
3565 - change default OSC RPC parameters to be better on small clusters
3566 - turn off OST read cache for files smaller than 32MB
3567 - install man pages and include them in rpms (3100)
3568 - add new init script for (un)mounting lustre filesystems (2593)
3569 - run chkconfig in %post for init scripts (3701)
3570 - drop scimac NAL (unmaintained)
3572 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3575 - clean kiobufs before and after use (3485)
3576 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3577 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3578 - change init script to fail more clearly if not run as root (1528)
3579 - allow clients to reconnect during replay (1742)
3580 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3581 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3582 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3583 - fix cond_resched (3554)
3584 - only evict unfinished clients after recovery (3515)
3585 - allow bulk resend, prevent data loss (3570)
3586 - dynamic ptlrpc request buffer allocation (2102)
3587 - don't allow unlinking open directory if it isn't empty (2904)
3588 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3589 - remove extraneous obd dereference causing LASSERT failure (3334)
3590 - don't use get_cycles() when creating temp. files on the mds (3156)
3591 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3592 - handle EEXIST for set-stripe, set proper directory name (3336)
3594 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3595 - fix ksocknal_fmb_callback() error messages (2918)
3597 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3600 - don't copy lvb into (possibly NULL) reply on error (2983)
3601 - don't deref dentry after dput, don't free lvb on error (2922)
3602 - use the kms to determine writeback rpc length (2947)
3603 - increment oti_logcookies when osc is inactive (2948)
3604 - update client's i_blocks count via lvb messages (2543)
3605 - handle intent open/close of special files properly (1557)
3606 - mount MDS with errors=remount-ro, like obdfilter (2009)
3607 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3608 - don't use cancelling-locks' kms values (2947)
3609 - use highest lock extent for kms, not last one (2925)
3610 - don't dereference ERR_PTR() dentry in error handling path (3107)
3611 - fix thread race in portals_debug_dumplog() (3122)
3612 - create lprocfs device entries at setup instead of at attach (1519)
3613 - common AST error handler, don't evict client on completion race (3145)
3614 - zero nameidata in detach_mnt in 2.6 (3118)
3615 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3616 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3617 - initialize RPC timeout timer earlier for 2.6 (3219)
3618 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3619 - print nal/nid for unknown nid (3258)
3620 - additional checks for oscc recovery before doing precreate (3284)
3621 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3622 - don't crash in mdc_close for bad permissions on open (3285)
3623 - zero i_rdev for non-device files (3147)
3624 - clear page->private before handing to FS, better assertion (3119)
3625 - tune the read pipeline (3236)
3626 - fix incorrect decref of invalidated dentry (2350)
3627 - provide read-ahead stats and refine rpc in flight stats (3328)
3628 - don't hold journal transaction open across create RPC (3313)
3629 - update atime on MDS at close time (3265)
3630 - close LDAP connection when recovering to avoid server load (3315)
3631 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3632 - don't leak open file on MDS after open resend (3325)
3633 - serialize filter_precreate and filter_destroy_precreated (3329)
3634 - loop device shouldn't call sync_dev() for nul device (3092)
3635 - clear page cache after eviction (2766)
3636 - resynchronize MDS->OST in background (2824)
3637 - refuse to mount the same filesystem twice on same mountpoint (3394)
3638 - allow llmount to create routes for mounting behind routers (3320)
3639 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3640 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3641 - fix rare problem with rename on htree directories (3417)
3643 - allow default OST striping configuration per directory (1414)
3644 - fix compilation for qswnal for 2.6 kernels (3125)
3645 - increase maximum number of MDS request buffers for large systems
3646 - change liblustreapi to be useful for external progs like lfsck (3098)
3647 - increase local configuration timeout for slow disks (3353)
3648 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3650 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3653 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3654 - don't overwrite extent policy data in reply if lock was blocked (2901)
3655 - drop filter export grants atomically with removal from device (2663)
3656 - del obd_self_export from work_list in class_disconnect_exports (2908)
3657 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3658 - swab reply message in mdc_close, other PPC fixes (2464)
3659 - fix destroying of named logs (2325)
3660 - overwrite old logs when running lconf --write_conf (2264)
3661 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3662 - fix race in target_handle_connect (2898)
3663 - mds_reint_create() should take same inode create lock (2926)
3664 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3665 - don't close files for self_export to avoid uninitialized obd (2936)
3666 - allow MDS with the same name as client node (2939)
3667 - hold dentry reference for closed log files for unlink (2325)
3668 - reserve space for all logs during transactions (2059)
3669 - don't evict page beyond end of stripe extent (2925)
3670 - don't oops on a deleted current working directory (2399)
3671 - handle hard links to targets without a parent properly (2517)
3672 - don't dereference NULL lock when racing during eviction (2867)
3673 - don't grow lock extents when lots of conflicting locks (2919)
3675 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3678 - account for cache space usage on clients to avoid data loss (974)
3679 - lfsck support in lustre kernel code (2349)
3680 - reduce journal credits needed for BRW writes (2370)
3681 - orphan handling to avoid losing space on client/server crashes
3682 - ptlrpcd can be blocked, stopping ALL progress (2477)
3683 - use lock value blocks to assist in proper KMS, faster stat (1021)
3684 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3685 - recovery for initial connections (2355)
3686 - fixes for mds_cleanup_orphans (1934)
3687 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3688 - block all file creations until orphan recovery completes (1901)
3689 - client remove rq_connection from request struct (2423)
3690 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3691 - recovery timer race (2670)
3692 - mdc_close recovey bug (2532)
3693 - ptlrpc cleanup bug (2710)
3694 - mds timeout on local locks (2588)
3695 - namespace lock held during RPCs (2431)
3696 - handle interrupted sync write properly (2503)
3697 - don't try to handle a message that hasn't been replied to (2699)
3698 - client assert failure during cleanup after abort recovery (2701)
3699 - leak mdc device after failed mount (2712)
3700 - ptlrpc_check_set allows timedout requests to complete (2714)
3701 - wait for inflight reqs when ptlrpcd finishes (2710)
3702 - make sure unregistered services are removed from the srv_list
3703 - reset bulk XID's when resending them (caught by 1138 test)
3704 - unregister_bulk after timeout
3705 - fix lconf error (2694)
3706 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3707 - readahead locks pages, leaves pending causing memory pressure (2673)
3708 - increase OST request buffers to 4096 on large machines (2729)
3709 - fix up permission of existing directories in simple_mkdir (2661)
3710 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3711 - don't assign transno to errored transactions (2742)
3712 - don't delete objects on OST if given a bogus objid from MDS (2751)
3713 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3714 - if rq_no_resend, then timeout request after recovery (2432)
3715 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3716 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3717 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3718 - MDS recovery shouldn't delete objects that it has given out (2730)
3719 - if enqueue arrives after completion, don't clobber LVB (2819)
3720 - don't unlock pages twice when trigger_group_io returns error (2814)
3721 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3722 - don't write pages to disk if there was an error (1450)
3723 - don't ping imports that have recovery disabled (2676)
3724 - take buffered bytes into account when balancing socknal conn (2817)
3725 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3726 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3727 - remove little-endian swabbing of llog records (1987)
3728 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3729 - retry reposting request buffers if they fail (1191)
3730 - grow extent at grant time to avoid granting a revoked lock (2809)
3731 - lock revoke doesn't evict page if covered by a second lock (2765)
3732 - disable VM readahead to avoid reading outside lock extents (2805)
3734 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3735 - updated kernel patches for hp-2.4.20 kernel (2681)
3737 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3740 - fix truncated write corruption (2366)
3741 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3743 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3744 - ensure len(uuid) < 37 in lmc (1171)
3745 - fix ia64 OOPS in llog_test (2255)
3746 - zero end of page at obdfilter for partial page writes (2648)
3747 - don't leave stale dentries around after renames (bug 2428)
3748 - fix timeouts when evicting a client with a single lock held (2642)
3749 - set deadline for the initial HELLO message to drain (2634)
3750 - print out dotted-quad IP addresses in the socknal (2302)
3752 - additional debugging for MDS client eviction problem (2443)
3753 - fix mkfsoptions support for osts (2603, 2604)
3755 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3758 - add series for the vanilla 2.6.0 kernel
3759 - add series for the vanilla 2.4.24 kernel
3760 - add series for a cray x86/64 UL kernel drop
3761 - fix xattr patches for the vanilla 2.4.19 series
3763 - generate true UUIDs in lmc (1171)
3764 - have portals stack dumping break in UML (2466)
3765 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3766 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3767 - wake callback waiting threads on client eviction (2460)
3768 - Add --ptldebug and --subsystem to lmc (1719)
3769 - update assertion to allow safe interrupt allocation
3770 - set rq_no_resend for cancel requests (2432)
3771 - recalculate ptlrpcd timeout after resend (2494)
3772 - call vfs_rmdir when removing pending directories (2368)
3773 - fix renaming a file to itself (2429)
3774 - lmc creates a default one-stripe lov (2454)
3775 - expand procfs space to handle large clusters (2326)
3776 - increase UML stack to avoid overflow
3777 - update lconf's list of debug and subsystem masks
3778 - fix lfs find --obd (2510)
3779 - /proc tunable for disabling filter read caching (2591)
3780 - stop rpm packages from altering slapd.conf (2301)
3781 - disable nagle in the socknal under 0conf (2578)
3782 - choose mds inode size based on stripe count (2572)
3783 - fix kernel-source rpm problems (2516)
3785 - add --disable-doc to avoid pdf generation (2421)
3786 - update documentation, tests, type-os, comments
3787 - avoid format warnings on ia64
3788 - remove the TOE NAL
3789 - tiny code cleanups by removing unused fields
3791 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3794 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3795 - remove the most blatant lies from BUILDING file (2371)
3796 - change default debug level to reasonable production setting
3797 - reduce client side cache size to reduce cache flush time
3798 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3799 - make TCP zerocopy and pinger support enabled by default (2476)
3800 - sync writes completed after process exits caused crashes (2319)
3801 - maintain correct mount count on the MDS (2356)
3802 - backout 1557, because 2316 wasn't really fixed
3803 - better file I/O statistics gathering in /proc
3804 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3805 - another kernel patch to fix zero-copy TCP function export
3806 - don't take duplicate lock when processing re-sent getattr (2420)
3807 - lctl uses obd_self_export instead of creating new conn (2353)
3808 - MDS/OST recovery case which requires object creation asserted (2425)
3809 - move lfs from /usr/sbin to /usr/bin in packages
3810 - fix race between mds_client_add and mds_client_free (2417)
3811 - use kmalloc instead of slabs in portals (2430)
3812 - don't create duplicate records when a failover MDS is present (2442)
3813 - remove unnecessary mount age check (2332)
3814 - don't remove directory inodes from locks prematurely (2451)
3815 - don't break if MDS service name is the same as hostname (2103)
3816 - fix races in client write RPC generation when cache full (2482)
3818 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3821 - remove now-unused request->rq_obd (278)
3822 - if an allocation fails, print out how much memory we've used (1933)
3823 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3824 - add the "configurable stack size" patch to most series files (1256)
3825 - ability to write large log records, for 100+ OST configs (2306)
3826 - fix NULL deref when filter_prep fails (2314)
3827 - fix operator precedence error in filter_sync
3828 - dynamic allocation of socknal TX descriptors (2315)
3829 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3830 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3831 - ensure that CWARN messages in Portals always get to the syslog
3832 - __init/__exit are not for prototype decls (ldlm_init/exit)
3833 - x86-64 compile warning fixes
3834 - fix gateway LMC keyword conflict (2318)
3835 - fix MDS lock inversions in getattr/reint paths (1844)
3836 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3837 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3838 - DLM race condition prevented some lock evictions (2328)
3839 - ENOMEM detection and retry on socknal sends (2230)
3840 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3841 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3842 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3843 - better error messages when a client is rejected during recovery (1505)
3844 - avoid cancelling locks which were never granted, after failure (2330)
3845 - fix i_sem/journal inversion in mds_client_add (2333)
3846 - fix truncate/getattr lock cycle deadlock (2334)
3847 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3848 - fix two rare exit paths which could leak an l_lock() ref (2321)
3849 - fencepost error in MDS/OST orphan recovery (2226)
3850 - make log record alignment 8 bytes (1988)
3851 - lstripe now fails when requested offset > ost_count (2237)
3852 - ensure that all kernel series have a complete list.h (1607)
3853 - fix crashes in special-file operations (2316)
3854 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3856 - allow configurable automake binary, for testing new versions
3857 - small update to the lfs documentation
3859 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3861 * fix negative export reference count in fsfilt_sync (2312)
3863 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3864 * release candidate 0.9.1
3866 - orphans are moved into the PENDING directory for possible recovery
3867 - replayed opens now open by fid for orphan/rename safety (1042)
3868 - last close of an orphan inode generates a transno (683)
3869 - chdir() and mount() now pin the directory entry (1020)
3870 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3871 - discard very old requests without processing them (1502)
3872 - remove some common, well-understood CERRORs (1505)
3873 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3874 - clear "grant" flags in OST replies until OST grant code lands (1644)
3875 - fix read performance by not clobbering i_blksize on client (1598)
3876 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3877 - make LRU size a /proc tunable, clears locks when reduced (707)
3878 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3879 - server locks take references on exports now (1558)
3880 - build fixes for 2.4.20-rh trees (1663)
3881 - return an error from lov_create if all OSCs are inactive (1751)
3882 - fix import levels when a reconnect happens without a timeout (1597)
3883 - exit early from mds_open if we get a lookup error (1749)
3884 - partial page read at EOF wouldn't wait for disk before sending (1642)
3885 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3886 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3887 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3888 - write/truncate lock inversion (1639)
3889 - Don't auto-load obdclass, portals modules during cleanup (1495)
3890 - fix timestamps from jumping to "now" (1763)
3891 - extra journal assertions (1648)
3892 - add an extra multiunlink test (1771)
3893 - fix read_record/write_record API (1776)
3894 - fix leak of offset_extent, possible incorrect i_size later (1772)
3895 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3896 - Debugging for the kqswnal_get_idle_tx problems (1820)
3897 - Allow recovery to be attempted multiple times (1536)
3898 - Write out MDS last_rcvd file after it is first created (1600)
3899 - Fix tx_descriptor leak in failed transmit situations (1827)
3900 - ext3 journaling fixes for assertion failure after IO error (1871)
3901 - class_export_put() on freed export after completion AST error (1896)
3902 - Fix revalidate looping in VFS (1322)
3903 - Don't access a freed export during MDS_REINT timeout (1521)
3904 - Add open-unlink recovery support on the MDS (1673,1764)
3905 - Return an error if no MDS data was read from last_rcvd (1946)
3906 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3907 - Don't setup a disk device that doesn't match exported UUID (317)
3908 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3909 - avoid committing NULL handle in force close
3910 - local.sh is now a one-stripe LOV configuration
3911 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3912 - fix ext3 htree duplicate directory entry corruption (1516)
3913 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3914 - update to new LOV EA format (2097)
3915 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3917 2003-06-15 Phil Schwan <phil@clusterfs.com>
3920 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3921 - per-import/export recovery handling (958, 931, 959)
3922 - multiple last-rcvd slots, for serving multiple FSes (949)
3923 - connections are again shared between multiple imp/exports (963, 964)
3924 - "umount -f" would hang if any requests needed to be sent (393, 978)
3925 - avoid pinning large req buffer by copying for queued messages (989)
3926 - add "uuid" to "lctl device" command to help upcalls (991)
3927 - "open" RPCs with transnos would confuse recovery counters (1037)
3928 - do proper endian conversion of all wire messages (288, 340, 891)
3929 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3930 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3931 - fix ptlrpc_connection leak in target_handle_connect (1174)
3932 - fix import refcounting bug in OST and MDS cleanup (1134)
3933 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3934 - fix very unlikely obd_types race condition (501)
3935 - remove osc_open hack for echo_client (1187)
3936 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3937 - a failure in read_inode2 leads to deadlock (1139)
3938 - cancel ack-locks as soon as transaction is committed (1072)
3939 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3940 - make sure to commitrw after any preprw to avoid deadlock (1162)
3941 - failing to execute a file in a lustre FS would lock inode (1203)
3942 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3943 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3944 - obd_uuid2tgt would walk off the end of the list (1255)
3945 - on IA64 the osc would give portals incorrect bulk size (1258)
3946 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3947 - fix lock inversion caused by new llite matching code (1282)
3948 - limit the number of dirty pages on a client to 10MB (1286)
3949 - timed out locks were not being corrected cancelled (1289)
3950 - fix O_DIRECT above 4GB on IA-32 (1292)
3951 * major user-visible changes
3952 - fail out/fail over policy now controlled by the upcall (993)
3954 - add OBD_PING to check server availability and failure (954)
3955 - lustre messages are now sent in sending host order (288, 340, 891)
3956 - add eadatalen to MDS getattr reply (340)
3957 - OST read replies may contain second buffer, with per-page status (593)
3959 2003-03-11 Phil Schwan <phil@clusterfs.com>
3962 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3963 - failed open()s could cause deadlock; fixed (867, 869)
3964 - stop cancelling OST locks when files are closed (481)
3965 - overlapping XID spaces caused network corruption (851, 853)
3966 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3967 - setattr_raw added, to do single-RPC, server-side setattrs
3968 - lmc/lconf syntax change for OST UUIDs
3969 - fix crashy race condition between ptlrpc_free_req and osc_close
3970 - don't use request in mdc_enqueue if we hit a timeout (889)
3971 - don't set the inode i_size for regular files from the MDS (896)
3972 - handle out of order completion AST (842)
3973 - don't LBUG if a lock request times out after receiving AST (913)
3974 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3975 - if a bad lock AST arrives, send an error instead of dropping entirely
3976 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3977 - fix leak in bulk IO when only partially completed (899, 900, 926)
3978 - fix O_DIRECT for ia64 (55)
3979 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3980 - C-z after timeout could hang a process forever; fixed (977)
3982 - client-side I/O cache (678, 924, 929, 941, 970)
3984 - READPAGE and SETATTRs which don't take server-side locks get
3987 2003-02-11 Phil Schwan <phil@clusterfs.com>
3990 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3991 - Fix fsfilt_extN_readpage() to read a full page of directory
3992 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3993 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3994 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3995 - Most intent operations were removed, in favour of a new RPC mode
3996 that does a single RPC to the server and bypasses most of the VFS
3997 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3998 - Aggressively cancel local locks on DLM servers
3999 - mds_reint_unlink sends EA to the client if it's the last nlink.
4000 client uses that EA to unlink OST objects.
4001 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4002 - recursive symlinks were fixed (439)
4003 - fixed NULL deref in DEBUG_REQ
4004 - filter_update_lastobjid no longer calls sync, which annoyed extN
4005 - fixed multi-client small-writes to a single file problem (445)
4006 - fixed mtime updates during file writes (607)
4007 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4008 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4009 - fixed Linux OST size reporting problem (444, 656)
4010 - OST now updates object mtime with writes or setattr (607, 619)
4011 - client verifies file size before zeroing page past EOF (445)
4012 - OST now writes last allocated objid to disk with allocation (108)
4013 - LOV on echo now works (409)
4015 - mds_reint_unlink sends a new buffer, with the EA included. this
4016 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4017 set if a regular file was being unlinked, and it was the last link
4018 - use PtlGet from the target for bulk writes (315)
4019 - OST now updates object mtime with writes or setattr (607, 619)
4020 - LDLM now has a grant-time callback to revalidate locked items, if
4022 - Many MDS operations were reorganized to combat race conditions
4024 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4025 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4027 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4030 - Fully reactivate OST imports after reconnection (512, others)
4031 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4032 - More graceful error handling for truncating on dead OST (515)
4033 - Don't error out unless we're actually accessing dead stripes (474)
4034 - Fix garbage sizes when stripes are missing (410)
4035 - LRU counters were broken, causing constant lock purge (433, 432)
4036 - garbage on read from stripes with failed OSTs (441)
4037 - mark OSCs as active before reconnecting during recovery (438)
4038 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4039 - lfind did not preserve OST order in output (443)
4040 - symlinks cause hung clients, incorrect data (439)
4041 - stop dereferencing request after dropping refcount (457)
4042 - don't LASSERT(spin_is_locked) on non-SMP (455)
4043 - fixes for many rename() bugs
4044 - fstat didn't correctly synchronize attributes (399)
4045 - server must handle lock cancellation during blocking AST prep (487)
4046 - bulk descriptors were free()d too soon (511)
4047 - fix paths in lconf, which would load incorrect modules (451, 507)
4048 - fix confusing lconf 'host not found' error message (386)
4049 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4050 - fix race condition in mdc_blocking_ast() for inode access (526)
4051 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4052 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4053 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4054 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4055 - fix lstripe in several ways (kernel side) (527)
4056 - fix request leak in ldlm_cli_enqueue (262)
4057 - incorrect OSC was marked inactive after OST failure
4058 - call mds_fs_cleanup before unmounting filesystem (524)
4059 - fix races between taking ns_lock and ldlm_lock_change_resource
4060 - fix races updating LOV export open file list
4061 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4062 - fix recovery NULL deref in ldlm_cli_cancel_unused
4063 - fix some DLM races by using new hash table for lock handles (419)
4064 - permit the client to specify desired inodes, at replay
4065 - duplicate requests when we queue them for replay reintegration
4066 - fix last_rcvd offset calculation
4067 - sync after each recovered transaction, so we always make progress
4068 - never, not always, ERESTART requests without transnos
4069 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4071 - skip replay if the MDS says that the client is already connected
4072 - don't check for a recovery-enabled export to match lctl's UUID
4073 - don't INC_USE_COUNT for phantom exports
4074 - don't crash when cleaning up phantom exports (567)
4075 - don't double-finish or set replay data for errored mdc_open requests
4076 - abort requests when they time out, so we don't get old replies
4077 - send/receive replies for AST messages again
4078 - if the client says that it doesn't have the lock, cancel it on the
4080 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4081 mark it as destroyed, it will all work out when decref is called
4082 - fix module use counts (22, 581)
4084 - ASTs now expect a reply (server cancels lock on error reply)
4086 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4089 - fix many simultaneous client startup (392)
4090 - fix dentry->d_it clobbering
4091 - credentials weren't being shipped for readdir/getattr operations
4092 - remove invalid assertions triggered during some concurrent MD
4094 - proper Lustre versions added (336, 389)
4095 - fix memory leak for create error case (398)
4096 - fix LOV locking bug that would get cli/srv out of sync
4097 - fix echo client over LOV (409)
4098 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4099 - fix double-O_EXCL intent crash (424)
4100 - avoid sending multiple lock CANCELs (352)
4102 - MDS can do multi-client recovery (modulo bugs in new code)
4104 - many updates, edits, cleanups
4106 2002-11-18 Phil Schwan <phil@clusterfs.com>
4109 - fix null d_it dereference (346)
4110 - fix full OST/dbench hang (333)
4111 - fix permission problem with file removal (286)
4112 - fix removal of OSCs from LOV when they fail
4113 - fix NULL deref during bulk timeout (214)
4114 - fix problems related to multiple filesystems on one MDS (241)
4115 - fixed serious subtle metadata locking bugs
4116 - free locks on clients when inodes are removed due to memory
4118 - fix inode pointer in lock data (285)
4119 - partial support for multiple MDS on a single host (241)
4120 - data locks weren't cancelled at clear_inode time (290, 311)
4121 - intent locks could lead to unbounded lock growth (205)
4122 - added a maximum lock count, an LRU list, and a flusher
4123 - fix multiple rename (365)
4124 - properly abstracted the echo client
4125 - OSC locked 1 byte too many; fixed
4126 - rewrote brw callback code:
4127 - fixed recovery bugs related to LOVs (306)
4128 - fixed too-many-pages-in-one-write crash (191)
4129 - fixed (again) crash in sync_io_timeout (214)
4130 - probably fixed callback-related race (385)
4132 - Add capability to MDS protocol
4133 - LDLM cancellations and callbacks on different portals
4135 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4138 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4139 - defer open object destruction to close time (601981, 138)
4140 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4141 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4142 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4143 - don't call obd_setattr in ll_file_release for destroyed objects
4144 * protocol change to lustre_msg: move |version| and add |flags|
4145 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4146 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4147 * added replay of create, unlink, link and rename operations during
4148 MDS failover; recovery should be much more robust now
4149 * remove failed OSCs from LOVs (only lov_create uses this so far)
4150 * the lustre-HOWTO was brought (more) up to date (582544)
4152 2002-10-23 Phil Schwan <phil@clusterfs.com>
4155 - in-use dentries weren't being reused properly (617851)
4156 - prevent multiple LDLM setup (599178)
4157 - fix LOV size calculations for truncate (617853)
4158 - fix client handling of MDS intent errors (POSIX)
4159 - fix permission bug in lovstripe.c test (624321)
4160 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4161 - truncate past end of file could corrupt data
4162 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4163 - a race in recovery could return ETIMEDOUT to apps (623947)
4164 - building outside the source directory was fixed
4165 * the lustre-HOWTO was brought (more) up to date (582544)
4166 * major progress was made on recovery functionality
4168 2002-10-10 Phil Schwan <phil@clusterfs.com>
4171 - recovery deadlock fix
4172 - rm -rf causes LBUG fix (617817)
4173 - file open by multiple tasks fix (618962)
4174 - directory permissions bugs (602707 and 620007)
4175 - journal_stop fixed with locking (611313)
4176 - O_APPEND failures resolved (618273, perhaps 614459)
4177 - lconf PATH fix (619770)
4178 - IA64 build fix (621450)
4179 - RPC buffer sizes scale with amount of memory
4181 2002-10-01 Phil Schwan <phil@clusterfs.com>
4184 - locks would be cancelled without throwing away data pages,
4185 resulting in inconsistent data (605627)
4186 - inode attributes were not always being refreshed (605627, 612449)
4187 - lconf now continues to cleanup after lctl reports an error
4188 - MDS now enforces user permissions (602707)
4189 - lprocfs cleanup fixed, but not yet enabled (614157)
4190 - fixed infinite server hang, should a client not respond to an AST
4191 - avoid going into recovery if user calls readlink() with a buffer
4192 that's too small (613941)
4193 - AST RPCs no longer require replies (614867) -- this may be changed
4194 - don't crash server if client sends an IOV that's too big (611336)
4195 - fixed lock conversion deadlock (611892)
4196 - fixed the following of symlinks (614622)
4197 * recovery: the server can remove locks from a client that dies, other
4198 clients can make progress
4199 * more extN patch fixes
4200 * compile-time configurable ptlrpc buffer allocations
4202 - collaborative read cache document
4203 - Lustre Lite Performance CDR document-in-progress
4205 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4208 - fix typo in patch-2.4.18
4210 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4213 - clear ptlrpc request each time in handle_incoming_request()
4214 - unlink of files now destroys the object on the OST
4216 2002-09-19 Peter Braam <braam@clusterfs.com>
4218 * add hard link support
4219 * change obdfile creation method
4220 * kernel patch changed
4222 2002-09-19 Peter Braam <braam@clusterfs.com>
4225 - stack overflow bug in extN fixed
4227 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4229 * documentation updates
4230 - add man pages for config tools
4231 - update tests/README to describe testing with new config tools
4232 - finish metadata API descriptions
4233 * bug fixes and cleanups
4234 - statfs workaround for 16TB limit
4235 - LOV stripe allocation improved, can stripe on subset of OSTs
4236 - LOV file size/IO offset was wrong for files > 4GB in size
4237 - object EA data was being dropped, caused files to be unreadable
4238 - memory overflow with non-LOV OST caused memory corruption
4239 - fixed regression tests to work with new config tools, obdfilter
4240 - fixed bug when directory size became larger than 1 block
4241 - fixed bug (for single client case) when PWD was deleted
4242 - invalidate local directory pages when doing intent-based ops
4243 - avoid LDLM oops when lock callback contained bad data
4245 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4247 * documentation updates
4248 * bug fixes and cleanups
4249 - configuration tools
4252 - 64-bit compile warnings
4253 - 64-bit internal statfs data
4255 * test_brw on persistent OST devices
4257 * lprocfs (disabled)
4259 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4261 * documentation updates
4262 * bug fixes and cleanups
4263 * configuration tools
4265 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4268 * many small fixes to 0_5_4
4269 * io/network handling
4270 * thinkos in MDS operations
4272 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4275 * crucial basic fixes to 0.5.3
4276 * IOR, Iozone work over Elan
4279 2002-08-07 Phil Schwan <phil@clusterfs.com>
4280 * version 0_5_3, our first alpha
4281 * we use the new Portals iovs
4282 * documentation updates
4283 * bug fixes and cleanups
4284 * small changes in the DLM wire protocol
4286 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4287 * version 0_5_1 with some initial stability,
4288 * locking on MD and file I/O.
4289 * documentation updates
4290 * several bug fixes since 0.5.0
4291 * small changes in wire protocol
4293 2002-07-18 Phil Schwan <phil@clusterfs.com>
4295 * delivered as Lustre Light Alpha
4296 * fixed a crash after handling invalid MDS requests
4297 * fixed directory pages for architectures with non-4k pages sizes
4299 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4300 * release version v0_4_4
4301 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4303 * Fixes a number of interruption problems with OST operations.
4304 * Update documentation for portals header changes
4305 * Move all wire protocol structs/defines to lustre_idl.h
4306 * Fixes symlink length bug.
4307 * Add tcpdump to repository.
4309 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4310 * release version v0_4_3
4311 * Fixes statfs for inodes on extN.
4312 * Fixes bug in runtests which would delete /etc/hosts.
4313 * Use 64-bit object IDs wherever possible (not into VFS though)
4314 Remove ost_get_info, which is unused by lustre, and out of date.
4316 2002-07-03 Peter Braam <braam@clusterfs.com>
4317 * release version v0_4_2 Fixes a lookup error (type not passed)
4318 * move forward to head of Portals
4319 * move forward to latest Lustre kernel
4321 2002-06-25 Peter Braam <braam@clusterfs.com>
4322 * release version v0_4_1. Hopefully stable on single node use.