1 tbd Sun Microsystems, Inc.
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.2-cfs5
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
13 Severity : enhancement
15 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
17 Severity : enhancement
19 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
23 Description: make mgs_setparam() handle fsname containing dash
24 Details : fsname containing a dash does not work with lctl conf_param
26 Severity : enhancement
28 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
30 Severity : enhancement
32 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
36 Frequency : rare, at shutdown
37 Description: access already free / zero obd_namespace.
38 Details : if client_disconnect_export was called without force flag set,
39 and exist connect request in flight, this can produce access to
40 NULL pointer (or already free pointer) when connect_interpret
41 store ocd flags in obd_namespace.
45 Frequency : only at startup
46 Description: not alloc memory with spinlock held.
47 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
50 Severity : enhancement
52 Description: make lustre randomly fail allocating memory
53 Details : Make lustre randomly failed allocating memory for testing purpose.
55 Severity : enhancement
57 Description: lost problems with lov objid file
58 Details : Fixes some scability and access to not inited memory problems
59 in work with lov objdid file.
64 Description: lfs find does not continue on file error
65 Details : Continue other files processing when a file/dir is absent.
69 Description: Inconsistent usage of lustre_pack_reply()
70 Details : Standardize the usage of lustre_pack_reply() such that it
71 always generate a CERROR on failure.
74 Frequency : occasional
76 Description: MDS hang or stay in waiting lock
77 Details : If client receive lock with CBPENDING flag ldlm need send lock
78 cancel as separate rpc, to avoid situation when cancel request
79 can't processed due all i/o threads stay in wait lock.
83 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
87 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
91 Description: Update to SLES9 kernel-2.6.5-7.287.3.
95 Description: Update to RHEL4 latest kernel.
97 Severity : enhancement
99 Description: Build SLES10 patchless client fails
100 Details : The configure was broken by run ./configure with
101 --with-linux-obj=.... argument for patchless client. When the
102 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
103 can't search header adequately. Use absolute path such as
104 -I($LINUX)/include instead.
108 Description: interrupt oig_wait produce painc on resend.
109 Details : brw_redo_request can be used for resend requests from ptlrpcd and
110 private set, and this produce situation when rq_ptlrpcd_data not
111 copyed to new allocated request and triggered LBUG on assert
112 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
113 ptlrpcd set if request is changed and can be safety changed to use
118 Description: LASSERT_{REQ,REP}SWAB macros are buggy
119 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
120 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
121 these macros and replace them with their difinitions instead.
125 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
126 Details : Update which_patch & target file for SLES10 latest kernel.
129 Description: Remove client patches from SLES 10 kernel.
130 Details : This causes SLES 10 clients to behave as patchless clients
131 even on a Lustre-patched (server) kernel.
133 Severity : enhancement
135 Description: self-adjustable client's lru lists
136 Details : use adaptive algorithm for managing client cached locks lru
137 lists according to current server load, other client's work
138 pattern, memory activities, etc. Both, server and client
139 side namespaces provide number of proc tunables for controlling
142 Severity : enhancement
144 Description: light-weight GSS support
145 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
146 and reduce performance overhead.
148 Severity : enhancement
150 Description: Linux keyring support
151 Details : Support using service of Linux keyring for Lustre GSS internal
152 context refresh/cache mechanism.
156 Description: Fix errors in lfs documentation
157 Details : Fixes man pages
161 Description: don't use GFP_* in generic Lustre code.
162 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
166 Description: obdclass is limited by single OBD_ALLOC(idarray)
167 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
171 Description: warnings with build patchless client with vanila 2.6.19 and up
172 Details : change old ctl_table style and replace ctl_table/ctl_table_header
173 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
177 Frequency : Only for SLES
178 Description: sanity_quota fail test_1
179 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
180 file, which makes TSTID[2] inunique.
184 Frequency : Only for SLES9
185 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
186 Details : Update target/ChangeLog/which_patch .
190 Frequency : Only for test_10 in sanity-quota.sh
191 Description: a bug in quota test code
192 Details : checking a test flag in a wrong place
196 Frequency : Only for test_13 in sanity-quota.sh
197 Description: a bug in quota test script
198 Details : shouldn't check overfull allocated quota
202 Description: jbd statistics
203 Details : Port older jbd statistics patch for sles10
205 Severity : enhancement
207 Description: Add printing inode info into message about error in writepage.
212 Description: reply_lock_interpret crash due to race with it and lock cancel.
213 Details : Do not replay locks that are being cancelled. Do not reference
214 locks by their address during replay, just by their handle.
218 Frequency : When flocks are used.
219 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
220 Details : Flock locks might destroy just granted lock if it could be merged
221 with another existing flock, this is done in completion handler,
222 so teach ldlm_cli_enquque_fini that this is a valid case for
228 Description: Oops in read and write path when failing to allocate lock.
229 Details : Check if lock allocation failed and return error back.
233 Description: lstripe command fails for valid OST index
234 Details : The stripe offset is compared to lov->desc.ld_tgt_count
235 instead of lov->desc.ld_active_tgt_count.
239 Description: sanity.sh failed test 103
240 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
245 Description: df doesn't work properly if diskfs blocksize != 4K
246 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
250 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
251 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
255 Description: obd_health_check_timeout too short
256 Details : set obd_health_check_timeout as 1.5x of obd_timeout
260 Description: llapi_file_create() does not allow some changes
261 Details : add llapi_file_open() that allows specifying the mode and
262 open flags, and also returns an open file handle.
266 Description: merge and cleanup kernel patches.
267 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
271 Description: Add journal checksum support.(Kernel part)
272 Details : The journal checksum feature adds two new flags i.e
273 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
274 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
275 indicates that the commit block contains the checksum for
276 the blocks described by the descriptor blocks. Now commit
277 record can be sent to disk without waiting for descriptor
278 blocks to be written to disk. This behavior is controlled
279 using JBD2_FEATURE_ASYNC_COMMIT flag.
283 Description: OSS needs mutliple precreate threads
284 Details : Add ability to start more than one create thread per OSS.
288 Description: Sanity.sh test_65k failed: setstripe should have succeeded
289 Details : execute lfs setstripe on client
293 Description: mds_obd_create error creating tmp object
294 Details : When the user sets quota on root, llog will be affected and can't
295 create files and write files.
298 Frequency : Always on ia64 patchless client, and possibly others.
300 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
301 Details : This allows the patchless client to be loaded on architectures
306 Description: RedHat Update kernel for RHEL5
307 Details : Add the kernel config file for RHEL5.
311 Description: RedHat Update kernel for RHEL5
312 Details : Modify the kernel config file more closer RHEL5.
316 Description: Build failure against Centos5 (RHEL5)
317 Details : Define PAGE_SIZE when it isn't present.
321 Description: test_7 on sanity-quota.sh can't be used on separate machines
322 Details : add facet to handle it
326 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
327 Details : with NFS, the anon dentry's parent was set to itself in
328 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
329 corresponding dentry other than use rec->ur_name.
331 Severity : enhancement
333 Description: lfs setstripe enhancement
334 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
338 Description: enable data checksumming by default
339 Details : enable checksum by default, allow --disable-checksum
340 configure option and "-o nochecksum" mount option. Checksums
341 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
345 Description: lustre support for RHEL5
346 Details : Add support for RHEL5.
350 Description: Client eviction due to failover config
351 Details : after a connection loss, the lustre client should attempt to
352 reconnect to the last active server first before trying the
353 other potential connections.
357 Description: when mds and osts use different quota unit(32bit and 64bit),
358 quota will be released repeatly.
359 Details : void sending multiple quota reqs to mds, which will keep the
360 status between the reqs.
364 Description: osts not allocated evenly to files
365 Details : change the condition to increase offset_idx
368 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
370 Description: Data corruption for OSTs that are formatted larger than 2TB
372 Details : When generating the bio request for lustre file writes the
373 sector number would overflow a temporary variable before being
374 used for the IO. The data reads correctly from Lustre (which
375 will overflow in a similar manner) but other file data or
376 filesystem metadata may be corrupted in some cases.
380 Description: Tune the kernel for good SCSI performance.
381 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
382 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
387 Description: rewrite ext2-derived code in obdclass/uuid.c
388 Details : rewrite inherited code (uuid parsing code from ext2 utils)
389 from scratch preserving functionality.
393 Description: "lfs find -obd UUID" prints directories
394 Details : "lfs find -obd UUID" will return all directory names instead
395 of just file names. It is incorrect because the directories
396 do not reside on the OSTs.
400 Description: use i_size_read and i_size_write in 2.6 port
401 Details : replace inode->i_size access with i_size_read/write()
404 Frequency : when removing large files
406 Description: scheduling issue during removal of large Lustre files
407 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
408 It causes scheduling issues when removing large files (17TB in the
412 Frequency : only with liblustre clients on XT3
414 Description: evictions taking too long
415 Details : allow llrd to evict clients directly on OSTs
418 Frequency : only on ppc
420 Description: /proc/fs/lustre/devices broken on ppc
421 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
422 We should make sure the type of 'index' is loff_t and then cast
423 explicitly as needed below. Do not assign an explicitly cast
427 Frequency : only for rhel5
429 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
430 Details : Modify the target file & which_kernel.
432 Severity : enhancement
434 Description: omit set fsid for export NFS
435 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
436 as FS_REQUIRES_DEV for avoid problems with generate fsid.
440 Frequency : Always, for kernels after 2.6.16
441 Description: Fix warning idr_remove called for id=.. which is not allocated.
442 Details : Last kernels save old s_dev before kill super and not allow
443 to restore from callback - restore it before call kill_anon_super.
447 Description: buffer overruns could theoretically occur
448 Details : llapi_semantic_traverse() modifies the "path" argument by
449 appending values to the end of the origin string, and a
450 overrun may occur. Adding buffer overrun check in liblustreapi.
454 Description: Fix error on 'ls .' at the top of the Lustre mount.
455 Details : Don't revalidate dentry if it is a root dentry.
459 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
460 Details : Modify vm-tunables-rhel4.patch.
464 Description: Kernel config for 2.6.18-vanilla.
465 Details : Modify targets/2.6-vanilla.target.in.
466 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
467 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
468 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
469 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
473 Description: improve handling recoverable errors
474 Details : if request processig with error which can be recoverable on server
475 request should be resend, otherwise page released from cache and
480 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
481 Details : Modify target file & which_patch.
482 A flaw was found in the IA32 system call emulation provided
483 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
484 value could be stored in the %RAX register, which could trigger an
485 out-of-bounds system call table access. An untrusted local user
486 could exploit this flaw to run code in the kernel
487 (ie a root privilege escalation). (CVE-2007-4573).
491 Description: O_DIRECT bypasses client statistics.
492 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
493 read_ahead_stats were updated. copied the stats section from
494 osc_send_oap_rpc() into async_internal().
498 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
502 Description: change order of libsysio includes
503 Details : '#include sysio.h' should always come before '#include xtio.h'
507 Description: handle "serious error: objid * already exists" more gracefully
508 Details : If LAST_ID value on disk is smaller than the objects existing in
509 the O/0/d* directories, it indicates disk corruption and causes an
510 LBUG(). If the object is 0-length, then we should use the existing
511 object. This will help to avoid a full fsck in most cases.
513 Severity : enhancement
515 Description: adapt the lustre_config script to support the upgrade case
516 Details : Add "-u" option for lustre_config script to support upgrading 1.4
517 server targets to 1.6 in parallel.
521 Description: To avoid grant space > avaible space when the disk is almost
522 full. Without this patch you might see the error "grant XXXX >
523 available" or some LBUG about grant, when the disk is almost
525 Details : In filter_check_grant, for non_grant cache write, we should
526 check the left space by if (*left > ungranted + bytes), instead
527 of (*left > ungranted), because only we are sure the left space
528 is enough for another "bytes", then the ungrant space should be
529 increase. In client, we should update cl_avail_grant only there
530 is OBD_MD_FLGRANT in the reply.
534 Description: Update RHEL 4 kernel to fix local root privilege escalation.
535 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
536 described in CVE-2007-4573. This problem could allow untrusted
537 local users to gain root access.
540 Frequency : when using O_DIRECT and quotas
542 Description: Incorrect file ownership on O_DIRECT output files
543 Details : block usage reported by 'lfs quota' does not take into account
544 files that have been written with O_DIRECT.
549 Description: touch file failed when fs is not full
550 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
551 otherwise we can get ENOSP while fs is not full.
555 Description: parallel lock callbacks
556 Details : Instead of sending blocking and completion callbacks as separated
557 requests, adding them to a set and sending in parallel.
560 Frequency : only for Cray XT3
561 Bugzilla : 12829/13455
562 Description: Changing primary group doesn't change the group lustre assigns to
564 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
565 overridden with the primary group provided by the group upcall,
566 whereas the supplied fsgid can be trusted if it is in the list of
567 supplementary groups returned by the group upcall.
569 Severity : enhancement
571 Description: Allow masking D_WARNING, D_ERROR messages from console
572 Details : Console messages can now be disabled via lnet.printk.
574 --------------------------------------------------------------------------------
576 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
578 * Support for kernels:
579 2.6.5-7.283 (SLES 9),
580 2.6.9-55.EL (RHEL 4),
581 2.6.16.46-0.14 (SLES 10),
582 2.6.18.8 vanilla (kernel.org)
583 * Client support for unpatched kernels:
584 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
585 2.6.16 - 2.6.22 vanilla (kernel.org)
586 * Due to recently discovered recovery problems, we do not recommend
587 using patchless RHEL 4 clients with this or any earlier release.
588 * Recommended e2fsprogs version: 1.39.cfs8
589 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
590 * Starting with this release, the ldiskfs backing filesystem required
591 by Lustre is now in its own package, lustre-ldiskfs. This package
592 should be installed. It is versioned separately from Lustre and
593 may be released separately in future.
597 Description: block reactivating mgc import until all deactivates complete
598 Details : Fix race when failing back MDT/MGS to itself (testing)
601 Frequency : at statup only
603 Description: mds_lov_synchronize race leads to various problems
604 Details : simultaneous MDT->OST connections at startup can cause the
605 sync to abort, leaving the OSC in a bad state.
607 Severity : enhancement
609 Description: add optional extra BUILD_VERSION info
610 Details : add a new environment variable (namely LUSTRE_VERS) which allows
611 to override the lustre version.
614 Frequency : 2.6.18 servers only
616 Description: ll_kern_mount() doesn't release the module reference
617 Details : The ldiskfs module reference count never drops down to 0
618 because ll_kern_mount() doesn't release the module reference.
623 Description: server LBUG when using old ost_num_threads parameter
624 Details : Accept the old ost_num_threads parameter but warn that it
625 is deprecated, and fix an off-by-one error that caused an LBUG.
630 Description: Transient SCSI error results in persistent IO issue
631 Details : iobuf->dr_error is not reinitialized to 0 between two
635 Frequency : sometimes when underlying device returns I/O errors
637 Description: OSTs not going read-only during write failures
638 Details : OSTs are not remounted read-only when the journal commit threads
639 get I/O errors because fsfilt_ext3 calls journal_start/stop()
640 instead of the ext3 wrappers.
644 Description: poor connect scaling with increasing client count
645 Details : Don't run filter_grant_sanity_check for more than 100 exports
646 to improve scaling for large numbers of clients.
649 Frequency : SLES10 only
651 Description: sanity-quota.sh quotacheck failed: rc = -22
652 Details : Quotas cannot be enabled on SLES10.
655 Frequency : liblustre clients only
657 Description: getdirentries does not give error when run on compute nodes
658 Details : getdirentries does not fail when the size specified as an argument
659 is too small to contain at least one entry
661 Severity : enhancement
663 Description: Add LNET router traceability for debug purposes
664 Details : If a checksum failure occurs with a router as part of the
665 IO path, the NID of the last router that forwarded the bulk data
666 is printed so it can be identified.
671 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
672 Details : Due to a race condition, liblustre clients were occasionally
675 Severity : enhancement
677 Description: lfs setstripe use optional parameters instead of postional
680 Severity : enhancement
682 Description: Nanosecond timestamp support for ldiskfs
683 Details : The on-disk ldiskfs filesystem has added support for nanosecond
684 resolution timestamps. There is not yet support for this at
685 the Lustre filesystem level.
688 Frequency : during server recovery
690 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
691 Details : request with rq_no_resend flag not awake l_wait_event if they get a
695 Frequency : nfs export on patchless client
697 Description: connectathon hang when test nfs export over patchless client
698 Details : Disconnected dentry cannot be found with lookup, so we do not need
699 to unhash it or make it invalid
702 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
703 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
704 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
709 Description: open req refcounting wrong on reconnect
710 Details : If reconnect happened between getting open reply from server and
711 call to mdc_set_replay_data in ll_file_open, we will schedule
712 replay for unreferenced request that we are about to free.
713 Subsequent close will crash in variety of ways.
714 Check that request is still eligible for replay in
715 mdc_set_replay_data().
720 Description: disable writes to filesystem when reading health_check file
721 Details : the default for reading the health_check proc file has changed
722 to NOT do a journal transaction and write to disk, because this
723 can cause reads of the /proc file to hang and block HA state
724 checking on a healthy but otherwise heavily loaded system. It
725 is possible to return to the previous behaviour during configure
726 with --enable-health-write.
728 Severity : enhancement
730 Description: 64-bit inode version
731 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
732 to the inode. This will be required for version-based recovery.
737 Description: MDS fails to start if a duplicate client export is detected
738 Details : in some rare cases it was possible for a client to connect to
739 an MDS multiple times. Upon recovery the MDS would detect this
740 and fail during startup. Handle this more gracefully.
742 Severity : enhancement
744 Description: Add -o localflock option to simulate old noflock
746 Details : This will achieve local-only flock/fcntl locks
752 Description: log_commit_thread vs filter_destroy race leads to crash
753 Details : Take import reference before releasing llog record semaphore
758 Description: Wrong request locking in request set processing
759 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
760 imp_delayed_list, in this place should be used imp_lock.
763 Frequency : when reconnection
765 Description: Grant Leak when osc reconnect to OST
766 Details : When osc reconnect ost, OST(filter) should check whether it
767 should grant more space to client by comparing fed_grant and
768 cl_avail_grant, and return the granted space to client instead
769 of "new granted" space, because client will call osc_init_grant
770 to update the client grant space info.
773 Frequency : when client reconnect to OST
775 Description: Grant Leak when osc do resend and replay bulk write
776 Details : When osc reconnect to OST, OST(filter)should clear grant info of
777 bulk write request, because the grant info while be sync between
778 OSC and OST when reconnect, and we should ignore the grant info
779 these of resend/replay write req.
784 Description: Grant space more than avaiable left space sometimes.
785 Details : When then OST is about to be full, if two bulk writing from
786 different clients came to OST. Accord the avaliable space of the
787 OST, the first req should be permitted, and the second one
788 should be denied by ENOSPC. But if the seconde arrived before
789 the first one is commited. The OST might wrongly permit second
790 writing, which will cause grant space > avaiable space.
793 Frequency : when client is evicted
795 Description: Grant might be wrongly erased when osc is evicted by OST
796 Details : when the import is evicted by server, it will fork another
797 thread ptlrpc_invalidate_import_thread to invalidate the
798 import, where the grant will be set to 0. While the original
799 thread will update the grant it got when connecting. So if
800 the former happened latter, the grant will be wrongly errased
801 because of this race.
806 Description: Checking Stale with correct fid
807 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
808 to check whether it is stale, because sometimes, we want the
809 enqueue happened anyway, and op_data.fid2 will not be initialized.
811 Severity : enhancement
813 Description: update patchless client
814 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
817 Frequency : only with 2.4 kernel
819 Description: random memory corruption
820 Details : size of struct ll_inode_info is to big for union inode.u and this
821 can be cause of random memory corruption.
826 Description: Memory leak in recovery
827 Details : Lov_mds_md was not free in an error handler in mds_create_object.
828 It should also check obd_fail before fsfilt_start, otherwise if
829 fsfilt_start return -EROFS,(failover mds during mds recovery).
830 then the req will return with repmsg->transno = 0 and rc = EROFS.
831 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
832 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
833 be freed no matter whether fsfilt_commit success or not.
836 Frequency : only with huge count clients
838 Description: Prevents from taking the superblock lock in llap_from_page for
840 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
841 for llap_from_page call in ll_removepage prevents from taking the
842 superblock lock for a soon died page.
847 Description: Not check open intent error before release open handle
848 Details : in some rare cases, the open intent error is not checked before
849 release open handle, which may cause
850 ASSERTION(open_req->rq_transno != 0), because it tries to release
851 the failed open handle.
856 Description: Set cat log bitmap only after create log success.
857 Details : in some rare cases, the cat log bitmap is set too early. and it
858 should be set only after create log success.
862 Description: Accessing a block bevice can re-enable I/O when Lustre is
863 tearing down a device.
864 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
869 Description: service threads may hog cpus when there are a lot of requests
870 Details : Insert cond_resched to give other threads a chance to use some CPU
875 Description: the cat log was not initialized in recovery
876 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
877 unlink log on mds will not be initialized until mds post
878 recovery. And also in mds post recovery, the unlink log will
879 initialization will be done asynchronausly, so there will be race
880 between add unlink log and unlink log initialization.
884 Description: brw_stats were being printed incorrectly
885 Details : brw_stats were being printed as log2 but all of them were not
886 recorded as log2. Also remove some code duplication arising from
887 filter_tally_{read,write}.
891 Frequency : rare, only in recovery.
892 Description: ASSERTION(req->rq_type != LI_POISON) failed
893 Details : imp_lock should be held while iterating over imp_sending_list for
894 prevent destroy request after get timeout in ptlrpc_queue_wait.
898 Description: replay-single.sh test 52 fails
899 Details : A lock's skiplist need to be cleanup when it being unlinked
900 from its resource list.
904 Description: Short directio read returns full requested size rather than
906 Details : Direct I/O operations should return actual amount of bytes
907 transferred rather than requested size.
909 Severity : enhancement
911 Description: metadata RPC reduction (e.g. for rm performance)
912 Details : decrease the amount of synchronous RPC between clients and servers
913 by canceling conflicing lock before the operation on the client
914 and packing thier handles into the main operation RPC to server.
916 Severity : enhancement
918 Description: Async OSC create to avoid the blocking unnecessarily.
919 Details : If a OST has no remain object, system will block on the creating
920 when need to create a new object on this OST. Now, ways use
921 pre-created objects when available, instead of blocking on an
922 empty osc while others are not empty. If we must block, we block
923 for the shortest possible period of time.
925 Severity : enhancement
927 Description: refine locking for avoid write wrong info into lov_objid file
928 Details : fix possible races with add new target and write/update data in
931 --------------------------------------------------------------------------------
933 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
938 Frequency : on some architectures
940 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
941 Details : Uninitialized flags sometimes cause configuration commands to
945 Frequency : patchless clients only
947 Description: missing __iget() symbol export
948 Details : The __iget() symbol export is missing. To avoid the need for
949 this on patchless clients the deathrow inode reaper is turned
950 off, and we depend on the VM to clean up old inodes. This
951 dependency was during via the fix for bug 12181.
953 --------------------------------------------------------------------------------
955 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
957 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
958 INTEROPERATE with older versions automatically. In many cases a
959 special upgrade step is needed. Please read the
960 user documentation before upgrading any part of a live system.
961 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
962 version will not interoperate with 1.6 betas before beta5 (1.5.95).
963 * WARNING: Lustre configuration and startup changes are required with
964 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
970 Severity : enhancement
972 Description: MountConf
973 Details : Lustre configuration is now managed via mkfs and mount
974 commands instead of lmc and lconf. New obd types (MGS, MGC)
975 are added for dynamic configuration management. See
976 https://mail.clusterfs.com/wikis/lustre/MountConf for
979 Severity : enhancement
981 Description: dynamic OST addition
982 Details : OSTs can now be added to a live filesystem
984 Severity : enhancement
986 Description: startup order invariance
987 Details : MDTs and OSTs can be started in any order. Clients only
988 require the MDT to complete startup.
990 Severity : enhancement
992 Description: parallel, asynchronous orphan cleanup
993 Details : orphan cleanup is now performed in separate threads for each
994 OST, allowing parallel non-blocking operation.
996 Severity : enhancement
998 Description: optimized stripe assignment
999 Details : stripe assignments are now made based on ost space available,
1000 ost previous usage, and OSS previous usage, in order to try
1001 to optimize storage space and networking resources.
1003 Severity : enhancement
1005 Description: Permanently set tunables
1006 Details : All writable /proc/fs/lustre tunables can now be permanently
1007 set on a per-server basis, at mkfs time or on a live system.
1009 Severity : enhancement
1011 Description: Lustre message v2
1012 Details : Add lustre message format v2.
1014 Severity : enhancement
1016 Description: client OST exclusion list
1017 Details : Clients can be started with a list of OSTs that should be
1018 declared "inactive" for known non-responsive OSTs.
1022 Description: SPEC SFS validation failure on NFS v2 over lustre.
1023 Details : Changes the blocksize for regular files to be 2x RPC size,
1024 and not depend on stripe size.
1026 Severity : enhancement
1028 Description: Multiple MD RPCs in flight.
1029 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1030 To avoid overly-overloading MDS, introduce a limit on number of
1031 MDS RPCs in flight for a single client and add /proc controls
1032 to adjust this limit.
1034 Severity : enhancement
1036 Description: client read/write statistics
1037 Details : Add client read/write call usage stats for performance
1038 analysis of user processes.
1039 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1040 file access. extents_stats shows chunk size distribution.
1041 extents_stats_per_process show chunk size distribution per
1044 Severity : enhancement
1046 Description: mds statistics
1047 Details : Add detailed mds operations statistics in
1048 /proc/fs/lustre/mds/*/stats.
1052 Description: Failure of copying files with lustre special EAs.
1053 Details : Client side always return success for setxattr call for lustre
1054 special xattr (currently only "trusted.lov").
1059 Description: Refcount LNET uuids
1060 Details : The global LNET uuid list grew linearly with every startup;
1061 refcount repeated list entries instead of always adding to
1064 Severity : enhancement
1066 Description: Dynamic service threads
1067 Details : Within a small range, start extra service threads
1068 automatically when the request queue builds up.
1071 Frequency : mixed-endian client/server environments
1073 Description: mixed-endian crashes
1074 Details : The new msg_v2 system had some failures in mixed-endian
1077 Severity : enhancement
1079 Description: Easy OST removal
1080 Details : OSTs can be permanently deactivated with e.g. 'lctl
1081 conf_param lustre-OST0001.osc.active=0'
1083 Severity : enhancement
1085 Description: MGS proc entries
1086 Details : Added basic proc entries for the MGS showing what filesystems
1089 Severity : enhancement
1091 Description: provide MGS failover
1092 Details : Added config lock reacquisition after MGS server failover.
1094 Severity : enhancement
1096 Description: add Linux 2.4 support
1097 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1101 Description: a large application tries to do I/O to the same resource and dies
1102 in the middle of it.
1103 Details : Check the req->rq_arrival time after the call to
1104 ost_brw_lock_get(), but before we do anything about
1105 processing it & sending the BULK transfer request. This
1106 should help move old stale pending locks off the queue as
1107 quickly as obd_timeout.
1110 Frequency : when an incorrect nid is specified during startup
1112 Description: ptlrpc connect to non-existant node causes kernel crash
1113 Details : LNET can't be re-entered from an event callback, which
1114 happened when we expire a message after the export has been
1115 cleaned up. Instead, hand the zombie cleanup off to another
1118 Severity : enhancement
1120 Description: plain/inodebits lock performance improvement
1121 Details : Grouping plain/inodebits in granted list by their request modes
1122 and bits policy, thus improving the performance of search through
1126 Frequency : only if OST filesystem is corrupted
1128 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1129 Details : for a short time RPCs with bulk IO are in the replay list,
1130 but replay of bulk IOs is unimplemented. If the OST filesystem
1131 is corrupted due to disk cache incoherency and then replay is
1132 started it is possible to trip an assertion. Avoid putting
1133 committed RPCs into the replay list at all to avoid this issue.
1136 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1138 Description: System hang on startup
1139 Details : This bug allowed the liblustre (e.g. catamount) client to
1140 return to the app before handling all startup RPCs. This
1141 could leave the node unresponsive to lustre network traffic
1142 and manifested as a server ptllnd timeout.
1144 Severity : enhancement
1146 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1147 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1148 boolean to control whether to print peer debug info when a
1149 client's RPC times out.
1152 Frequency : only for kernels with patches from Lustre below 1.4.3
1154 Description: Remove old rdonly API
1155 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1158 Frequency : only for devices with external journals
1160 Description: Set external device read-only also
1161 Details : During a commanded failover stop, we set the disk device
1162 read-only while the server shuts down. We now also set any
1163 external journal device read-only at the same time.
1166 Frequency : when upgrading from 1.4 while trying to change parameters
1168 Description: The wrong (new) MDC name was used when setting parameters for
1169 upgraded MDT's. Also allows changing of OSC (and MDC)
1170 parameters if --writeconf is specified at tunefs upgrade time.
1173 Frequency : when setting specific ost indicies
1175 Description: QOS code breaks on skipped indicies
1176 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1177 created with --index need not be sequential.
1179 Severity : enhancement
1181 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1182 Details : The uninit_groups feature works in conjunction with the kernel
1183 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1184 up the pass1 processing of e2fsck. This is a read-only feature
1185 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1186 mount filesystems that have had this feature enabled.
1188 Severity : enhancement
1190 Description: Improve multi-block allocation algorithm to avoid fragmentation
1191 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1192 allocation locality and avoid filesystem fragmentation.
1194 ------------------------------------------------------------------------------
1196 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1198 * Support for kernels:
1199 2.4.21-47.0.1.EL (RHEL 3)
1200 2.6.5-7.283 (SLES 9)
1201 2.6.9-42.0.10.EL (RHEL 4)
1202 2.6.12.6 vanilla (kernel.org)
1203 2.6.16.27-0.9 (SLES 10)
1204 * Recommended e2fsprogs version: 1.39.cfs5
1206 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1210 Frequency : occasional, depends on client load and configuration
1211 Bugzilla : 12181, 12203
1212 Description: data loss for recently-modified files
1214 Details : In some cases it is possible that recently written or created
1215 files may not be written to disk in a timely manner (this should
1216 normally be within 30s unless client IO load is very high).
1217 The problem appears as zero-length files or files that are a
1218 multiple of 1MB in size after a client crash or client eviction
1219 that are missing data at the end of the file.
1221 This problem is more likely to be hit on clients where files are
1222 repeatedly created and unlinked in the same directory, clients
1223 have a large amount of RAM, have many CPUs, the filesystem has
1224 many OSTs, the clients are rebooted frequently, and/or the files
1225 are not accessed by other nodes after being written.
1227 The presence of the problem can be detected by looking at
1228 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1229 smaller than the second (nr_unused) then dirty files will not
1230 be flushed automatically to disk. "sync; sleep 10" should be
1231 run several times on the node before unmounting it to update
1232 Lustre (this is also safe to run on nodes without this problem).
1234 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1235 that can cause this same problem, so customers using that kernel
1236 also need to update the kernel in addition to Lustre. In order
1237 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1239 It is normal that files written just before a client crash (less
1240 than 30s) may not yet have been flushed to disk, even for local
1244 Frequency : frequent on thin XT3 nodes
1246 Description: UUID collision on thin XT3 Linux nodes
1247 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1248 randomly, since we relied on an insufficiently-seeded PRNG.
1253 Description: OSS hangs after "All ost request buffers busy"
1254 Details : A deadlock between quota and journal operations caused OSS
1255 hangs after printing "All ost request buffers busy."
1258 Frequency : always on liblustre builds
1260 Description: Cleanup compiler warnings on liblustre
1263 Frequency : always on liblustre builds on XT3
1265 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1269 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1270 > 32000 subdirectories
1271 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1272 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1273 subdirectory count crosses 32000. This will aid e2fsck to
1274 correctly handle more than 32000 subdirectories.
1277 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1279 Description: System hang on startup
1280 Details : This bug allowed the liblustre (e.g. catamount) client to
1281 return to the app before handling all startup RPCs. This
1282 could leave the node unresponsive to lustre network traffic
1283 and manifested as a server ptllnd timeout.
1285 Severity : enhancement
1287 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1288 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1289 boolean to control whether to print peer debug info when a
1290 client's RPC times out.
1295 Description: make O_SYNC working on 2.6 kernels
1296 Details : 2.6 kernels use different method for mark pages for write,
1297 so need add a code to lustre for O_SYNC work.
1302 Description: Failure to close file and release space on NFS
1303 Details : Put inode details into lock acquired in ll_intent_file_open.
1304 Use mdc_intent_lock in ll_intent_open to properly
1305 detect all kind of errors unhandled by mdc_enqueue.
1310 Description: proc file read during shutdown sometimes raced obd removal,
1312 Details : Add lock to prevent obd access after proc file removal.
1315 Frequency : Only for files larger than 4GB on 32-bit clients.
1317 Description: improperly doing page alignment of locks
1318 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1319 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1324 Description: RPCs being resent when they shouldn't be
1325 Details : Some RPCs that should not be resent are being resent. This
1326 can cause inconsistencies in the RPC state machine. Do not
1327 resend such requests.
1330 Frequency : rare, only with NFS export
1332 Description: Crash on NFS re-export node
1333 Details : under very unusual load conditions an assertion is hit in
1334 ll_intent_file_open()
1337 Frequency : only if OST filesystem is corrupted
1339 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1340 Details : for a short time RPCs with bulk IO are in the replay list,
1341 but replay of bulk IOs is unimplemented. If the OST filesystem
1342 is corrupted due to disk cache incoherency and then replay is
1343 started it is possible to trip an assertion. Avoid putting
1344 committed RPCs into the replay list at all to avoid this issue.
1349 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1350 Details : Large single O_DIRECT read and write calls can fail to allocate
1351 a sufficiently large buffer to process the request. In case of
1352 allocation failure the allocation is retried with a smaller
1353 buffer and broken into smaller requests.
1355 Severity : enhancement
1357 Description: Add -o localflock option to simulate old noflock behaviour.
1358 Details : This will achieve local-only flock/fcntl locks coherentness.
1363 Description: versioning check is incomplete
1364 Details : Checking the version difference of client vs. server, report
1365 error if the gap is too big.
1370 Description: add support PG_writeback bit
1371 Details : add support for PG_writeback bit for Lustre, for more carefull
1372 work with page cache in 2.6 kernel. This also fix some deadlocks
1373 and remove hack for work O_SYNC with 2.6 kernel.
1375 Severity : enhancement
1377 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1378 Details : The uninit_groups feature works in conjunction with the kernel
1379 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1380 up the pass1 processing of e2fsck. This is a read-only feature
1381 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1382 mount filesystems that have had this feature enabled.
1384 Severity : enhancement
1386 Description: Improve multi-block allocation algorithm to avoid fragmentation
1387 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1388 allocation locality and avoid filesystem fragmentation.
1390 ------------------------------------------------------------------------------
1392 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1394 * Support for kernels:
1395 2.6.9-42.0.3.EL (RHEL 4)
1396 2.6.5-7.276 (SLES 9)
1397 2.4.21-47.0.1.EL (RHEL 3)
1398 2.6.12.6 vanilla (kernel.org)
1399 2.6.16.21-0.8 (SLES10)
1400 * Recommended e2fsprogs version: 1.39.cfs2-0
1402 * The backwards-compatible /proc/sys/portals symlink has been removed
1403 in this release. Before upgrading, please ensure that you change
1404 any configuration scripts or /etc/sysctl.conf files that access
1405 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1406 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1407 in advance of the upgrade on any system running Lustre 1.4.6 or
1408 newer, since /proc/sys/lnet was added in that version.
1409 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1413 Frequency : only when quota is used
1415 Description: avoid scanning export list for quota master
1416 Details : Change the algorithms to avoid scanning export list in order
1417 to improve the efficiency.
1420 Frequency : MDS failover only, very rarely
1422 Description: "went back in time" messages on mds failover
1423 Details : The greatest transno may be lost when the current operation
1424 finishes with an error (transno==0) and the client's last_rcvd
1425 record is over-written. Save the greatest transno in the
1426 mds_last_transno for this case.
1429 Frequency : always for specific kernels and striping counts
1431 Description: client may get "Matching packet too big" without ACL support
1432 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1433 when trying to access files in certain configurations. The
1434 clients should in fact be denied when mounting because they do
1435 not understand ACLs.
1438 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1440 Description: many clients connecting with IO in progress causes connect timeouts
1441 Details : Avoid synchronous journal commits to avoid delays caused by many
1442 clients connecting/disconnecting when bulk IO is in progress.
1443 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1444 OST_IO_PORTAL to avoid delays behind potentially many pending
1448 Frequency : occasionally with multiple writers to a single file
1450 Description: shared writes to file may result in wrong size reported by stat()
1451 Details : Allow growing of kms when extent lock is cancelled
1454 Frequency : always with random mmap IO to multi-striped file
1456 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1457 Details : Only if the hole is at the end of OST object so that kms is too
1458 small. Fix is to increase kms accordingly in ll_nopage.
1461 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1463 Description: writes to a missing object would leak memory on the OST
1464 Details : If there is an inconsistency between the MDS and OST filesystems,
1465 such that the MDS references an object that doesn't exist, writes
1466 to that object will leak memory due to incorrect cleanup in the
1467 error handling path, eventually running out of memory on the OST.
1472 Description: Creating too long symlink causes lustre errors
1473 Details : Check symlink and name lengths before sending requests to MDS.
1476 Frequency : only if flock is enabled (not on by default)
1478 Description: posix locks not released on fd closure on 2.6.9+
1479 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1480 kernels, this caused such locks not to be released on fd close and
1481 then assertions on fs unmount about still used locks.
1484 Frequency : MDS failover only, very rarely
1486 Description: clients may get ASSERTION(granted_lock != NULL)
1487 Details : When request was taking a long time, and a client was resending
1488 a getattr by name lock request. The were multiple lock requests
1489 with the same client lock handle and
1490 mds_getattr_name->fixup_handle_for_resent_request found one of the
1491 lock handles but later failed with ASSERTION(granted_lock != NULL).
1496 Description: handle->h_buffer_credits > 0, assertion failure
1497 Details : h_buffer_credits is zero after truncate, causing assertion
1498 failure. This patch extends the transaction or creates a new
1502 Frequency : NFS re-export or patchless client
1503 Bugzilla : 11179, 10796
1504 Description: Crash on NFS re-export node (__d_move)
1505 Details : We do not want to hash the dentry if we don't have a lock.
1506 But if this dentry is later used in d_move, we'd hit uninitialised
1507 list head d_hash, so we just do this to init d_hash field but
1508 leave dentry unhashed.
1511 Frequency : NFS re-export or patchless client
1513 Description: NFS exports has problem with symbolic link
1514 Details : lustre client didn't properly install dentry when re-exported
1515 to NFS or running patchless client.
1518 Frequency : NFS re-export or patchless client
1520 Description: Various nfs/patchless fixes.
1521 Details : fixes reuse disconected alias for lookup process - this fixes
1522 warning "find_exported_dentry: npd != pd",
1523 fix permission error with open files at nfs.
1524 fix apply umask when do revalidate.
1527 Frequency : occasional
1529 Description: Crash on NFS re-export node
1530 Details : calling clear_page() on the wrong pointer triggered oops in
1531 generic_mapping_read().
1534 Frequency : rarely, using O_DIRECT IO
1536 Description: unaligned directio crashes client with LASSERT
1537 Details : check for unaligned buffers before trying any requests.
1540 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1542 Description: stale data returned from RAID cache
1543 Details : If only a small amount of IO is done to the RAID device before
1544 reading it again it is possible to get stale data from the RAID
1545 cache instead of reading it from disk.
1548 Frequency : always for sles10 kernel
1550 Description: sles10 support
1551 Details : ll_follow_link: compile fixes and using of nd_set_link
1552 under newer kernels.
1555 Frequency : depends on arch, kernel and compiler version, always on sles10
1558 Description: recursive or deep enough symlinks cause stack overflow
1559 Details : getting rid of large stack-allocated variable in
1563 Frequency : depends on hardware
1565 Description: lustre write performance loss in the SLES10 kernel
1566 Details : the performance loss is caused by using of write barriers in the
1567 ext3 code. The SLES10 kernel turns barrier support on by
1568 default. The fix is to undo that change for ldiskfs.
1570 ------------------------------------------------------------------------------
1572 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1574 * Support for kernels:
1575 2.6.9-42.0.3EL (RHEL 4)
1576 2.6.5-7.276 (SLES 9)
1577 2.4.21-47.0.1.EL (RHEL 3)
1578 2.6.12.6 vanilla (kernel.org)
1582 Frequency : quota enabled and large files being deleted
1584 Description: releasing more than 4GB of quota at once hangs OST
1585 Details : If a user deletes more than 4GB of files on a single OST it
1586 will cause the OST to spin in an infinite loop. Release
1587 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1592 Description: statfs data retrieved from /proc may be stale or zero
1593 Details : When reading per-device statfs data from /proc, in the
1594 {kbytes,files}_{total,free,avail} files, it may appear
1595 as zero or be out of date.
1598 Frequency : systems with MD RAID1 external journal devices
1600 Description: lconf's call to blkid is confused by RAID1 journal devices
1601 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1602 of returning all block devices that match the journal UUID.
1605 Frequency : always, for aggregate stripe size over 4GB
1607 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1608 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1609 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1612 Frequency : always if "lfs find" used on a local file/directory
1614 Description: "lfs find" segfaults if used on a local file/directory
1615 Details : The case where a directory component was not specified wasn't
1616 handled correctly. Handle this properly.
1619 Frequency : always on ppc64
1621 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1622 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1623 perform endianness conversion.
1626 Frequency : rarely (truncate to non-zero file size after write under load)
1627 Bugzilla : 10730, 10687
1628 Description: Files padded with zeros to next 4K multiple
1629 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1630 it is possible that files that are truncated to a non-zero size
1631 immediately after being written are filled with zero bytes beyond
1632 the truncated size. No file data is lost.
1634 Severity : enhancement
1636 Description: Allow recovery/failover for liblustre clients.
1637 Details : liblustre clients were unaware of failover configurations until
1640 Severity : enhancement
1642 Description: user file locks should fail when not mounting with flock option
1643 Details : Set up an error-returning stub in ll_file_operations.lock field
1644 to prevent incorrect behaviour when client is mounted without
1645 flock option. Also, set up properly f_op->flock field for
1649 Frequency : always on ia64
1651 Description: "lfs df" loops on printing out MDS statfs information
1652 Details : The obd_ioctl_data was not initialized and in some systems
1653 this caused a failure during the ioctl that did not return
1654 an error. Initialize the struct and return an error on failure.
1657 Frequency : SLES 9 only
1659 Description: Error of copying files with lustre special EAs as root
1660 Details : Client side always return success for setxattr call for lustre
1661 special xattr (currently only "trusted.lov").
1664 Frequency : rarely on clusters with both ia64+i386 clients
1666 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1667 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1668 clients are doing concurrent writes to the same file it is
1669 possible that the ia64 clients may LASSERT because the OST
1670 extent locks are not PAGE_SIZE aligned. Ensure that grown
1671 locks are always aligned on the request boundary.
1674 Frequency : specific use, occasional
1676 Description: Overwriting in use executable truncates on-disk binary image
1677 Details : If one node attempts to overwrite an executable in use by
1678 another node, we now correctly return ETXTBSY instead of
1679 truncating the file.
1681 Severity : enhancement
1683 Description: Async OSC create to avoid the blocking unnecessarily.
1684 Details : If a OST has no remain object, system will block on the creating
1685 when need to create a new object on this OST. Now, ways use
1686 pre-created objects when available, instead of blocking on an
1687 empty osc while others are not empty. If we must block, we block
1688 for the shortest possible period of time.
1693 Description: chmod on Lustre root is propagated to other clients
1694 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1695 invalid by the follow_mount time.
1698 Frequency : liblustre clients only
1700 Description: Race in 'instant cancel' lock handling could lead to such locks
1701 never to be granted in case of SMP MDS
1702 Details : Do not destroy not yet granted but cbpending locks in
1706 Frequency : replay/resend of open
1708 Description: non null lock assetion failure in mds_intent_policy
1709 Details : Trying to replay/resend lockless open requests resulted in
1710 mds_open() returning 0 with no lock. Now it sets a flag if
1711 a lock is going to be returned.
1713 Severity : enhancement
1715 Description: Checksum enhancements
1716 Details : New checksum enhancements allow for resending RPCs that failed
1719 Severity : enhancement
1721 Description: Tunables on number of dirty pages in cacche
1722 Details : Allow to set limit on number of dirty pages cached.
1727 Description: client crash on unmount - lock still has references
1728 Details : In some error handling cases it was possible to leak a lock
1729 reference on a client while accessing a file. This was not
1730 harmful to the client during operation, but would cause the
1731 client to crash when the filesystem is unmounted.
1734 Frequency : specific case, rare
1736 Description: ETXTBSY on mds though file not in use by client
1737 Details : ETXTBSY is no longer incorrectly returned when attempting to
1738 chmod or chown a directory that the user previously tried to
1739 execute or a currently-executing binary.
1742 Frequency : extremely rare except on liblustre-based clients
1744 Description: Lustre space not freed when files are deleted
1745 Details : Clean up open-unlinked files after client eviction. Previously
1746 the unlink was skipped and the files remained as orphans.
1751 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1752 Details : In some cases it was possible to send a blocking callback to
1753 a client doing a glimpse, even though that client didn't get
1754 a lock granted. When the glimpse lock is cancelled on the OST
1755 the freed lock is left on the waiting list and corrupted the list.
1758 Frequency : all core dumps
1760 Description: Broke core dumps to lustre
1761 Details : Negative dentry may be unhashed if parent does not have UPDATE
1762 lock, but some callers, e.g. do_coredump, expect dentry to be
1763 hashed after successful create, hash it in ll_create_it.
1765 ------------------------------------------------------------------------------
1767 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1769 * Support for kernels:
1770 2.6.9-42.0.2.EL (RHEL 4)
1771 2.6.5-7.276 (SLES 9)
1772 2.4.21-40.EL (RHEL 3)
1773 2.6.12.6 vanilla (kernel.org)
1777 Frequency : always on RHEL 3
1779 Description: Number of open files grows over time
1780 Details : The number of open files grows over time, whether or not
1781 Lustre is started. This was due to a filp leak introduced
1782 by one of our kernel patches.
1784 ------------------------------------------------------------------------------
1786 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1788 * Support for kernels:
1789 2.6.9-42.EL (RHEL 4)
1790 2.6.5-7.276 (SLES 9)
1791 2.4.21-40.EL (RHEL 3)
1792 2.6.12.6 vanilla (kernel.org)
1797 Bugzilla : 5719, 9635, 9792, 9684
1798 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1799 Details : If a server is under heavy load and cannot reply to new
1800 connection requests before the client resends the (re)connect,
1801 the connection handling code can behave badly if two service
1802 threads are concurrently handing separate (re)connections from
1803 the same client. Add better locking to the connection handling
1804 code, and ensure that only a single connection will be processed
1805 for a given client UUID, even if the lock is dropped.
1807 Severity : enhancement
1809 Description: add TCP zero-copy support to kernel
1810 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1811 sends if the hardware supports scatter-gather and checksumming.
1812 This allows socklnd to do client-write and server-read more
1813 efficiently and reduce CPU utilization from skbuf copying.
1816 Frequency : only if NFS exporting from client
1818 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1819 Details : If setting attributes on a file created under NFS that had
1820 never been opened it would be possible to oops the client
1821 if the file had no objects.
1824 Frequency : always for liblustre
1826 Description: liblustre client does MDS+OSTs setattr RPC for each write
1827 Details : When doing a write from a liblustre client, the client
1828 incorrectly issued an RPC to the MDS and each OST the file was
1829 striped over in order to update the timestamps. When writing
1830 with small chunks and many clients this could overwhelm the MDS
1831 with RPCs. In all cases it would slow down the write because
1832 these RPCs are unnecessary.
1834 Severity : enhancement
1836 Description: allow number of MDS service threads to be changed at module load
1837 Details : It is now possible to change the number of MDS service threads
1838 running. Adding "options mds mds_num_threads={N}" to the MDS's
1839 /etc/modprobe.conf will set the number of threads for the next
1840 time Lustre is restarted (assuming the "mds" module is also
1841 reloaded at that time). The default number of threads will
1842 stay the same, 32 for most systems.
1847 Description: OST crash if filesystem is unformatted or corrupt
1848 Details : If an OST is started on a device that has never been formatted
1849 or if the filesystem is corrupt and cannot even mount then the
1850 error handling cleanup routines would dereference a NULL pointer.
1855 Description: NULL pointer deref in llap_from_page.
1856 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1857 incorrect) mapping if the page was truncated/reclaimed while it was
1858 searched for. Check for this condition and skip such pages when
1859 doing readahead. Introduce extra check to llap_from_page() to
1860 verify page->mapping->host is non-NULL (so page is not anonymous).
1863 Frequency : Sometimes when using sys_sendfile
1865 Description: "page not covered by a lock" warnings from ll_readpage
1866 Details : sendfile called ll_readpage without right page locks present.
1867 Now we introduced ll_file_sendfile that does necessary locking
1868 around call to generic_file_sendfile() much like we do in
1872 Frequency : with certain MDS communication failures at client mount time
1874 Description: NULL pointer deref after failed client mount
1875 Details : a client connection request may delayed by the network layer
1876 and not be sent until after the PTLRPC layer has timed out the
1877 request. If the client fails the mount immediately it will try
1878 to clean up before the network times out the request. Add a
1879 reference from the request import to the obd device and delay
1880 the cleanup until the network drops the request.
1883 Frequency : occasionally during client (re)connect
1885 Description: assertion failure during client (re)connect
1886 Details : processing a client connection request may be delayed by the
1887 client or server longer than the client connect timeout. This
1888 causes the client to resend the connection request. If the
1889 original connection request is replied in this interval, the
1890 client may trip an assertion failure in ptlrpc_connect_interpret()
1891 which thought it would be the only running connect process.
1894 Frequency : only with obd_echo servers and clients that are rebooted
1896 Description: kernel BUG accessing uninitialized data structure
1897 Details : When running an obd_echo server it did not start the ping_evictor
1898 thread, and when a client was evicted an uninitialized data
1899 structure was accessed. Start the ping_evictor in the RPC
1900 service startup instead of the OBD startup.
1902 Severity : enhancement
1903 Bugzilla : 10193 (patchless)
1904 Description: Remove dependency on various unexported kernel interfaces.
1905 Details : No longer need reparent_to_init, exit_mm, exit_files,
1906 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1909 Frequency : rare (only users of deprecated and unsupported LDAP config)
1911 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1912 Details : LDAP apparently contains 'lustreName' attributes instead of
1913 'name'. A simple remapping of the name is sufficient.
1916 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1918 Description: waiting_locks_callback trips kernel BUG if client is evicted
1919 Details : Running with the dump_on_timeout debug flag turned on makes
1920 it possible that the waiting_locks_callback() can try to dump
1921 the Lustre kernel debug logs from an interrupt handler. Defer
1922 this log dumping to the expired_lock_main() thread.
1924 Severity : enhancement
1926 Description: Support NFS exporting on 2.6 kernels.
1927 Details : Implement non-rawops metadata methods for NFS server to use without
1928 changing NFS server code.
1931 Frequency : very rare (synthetic metadata workload only)
1933 Description: two racing renames might cause an MDS thread to deadlock
1934 Details : Running the "racer" program may cause one MDS thread to rename
1935 a file from being the source of a rename to being the target of
1936 a rename at exactly the same time that another thread is doing
1937 so, and the second thread has already enqueued these locks after
1938 doing a lookup of the target and is trying to relock them in
1939 order. Ensure that we don't try to re-lock the same resource.
1942 Frequency : only very large systems with liblustre clients
1944 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1945 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1946 all OSTs in parallel. This allows the eviction of stale liblustre
1947 clients to proceed much faster than if they were done in series,
1948 and also offers similar improvements for other set_info RPCs.
1953 Description: excessive CPU usage during initial read phase on client
1954 Details : During the initial read phase on a client, it would agressively
1955 retry readahead on the file, consuming too much CPU and impacting
1956 performance (since 1.4.5.8). Improve the readahead algorithm
1957 to avoid this, and also improve some other common cases (read
1958 of small files in particular, where "small" is files smaller than
1959 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1964 Description: MDS crash when receiving packet with unknown intent.
1965 Details : Do not LBUG in unknown intent case, just return -EFAULT
1967 Severity : enhancement
1968 Bugzilla : 9293, 9385
1969 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1970 Details : Do not serialize getattr (non-intent version) and statfs.
1973 Frequency : occasional, when OST network is overloaded/intermittent
1975 Description: client evicted by OST after bulk IO timeout
1976 Details : If a client sends a bulk IO request (read or write) the OST
1977 may evict the client if it is unresposive to its data GET/PUT
1978 request. This is incorrect if the network is overloaded (takes
1979 too long to transfer the RPC data) or dropped the OST GET/PUT
1980 request. There is no need to evict the client at all, since
1981 the pinger and/or lock callbacks will handle this, and the
1982 client can restart the bulk request.
1985 Frequency : Always when mmapping file with no objects
1987 Description: client crashes when mmapping file with no objects
1988 Details : Check that we actually have objects in a file before doing any
1989 operations on objects in ll_vm_open, ll_vm_close and
1995 Description: Request leak when working with deleted CWD
1996 Details : Introduce advanced request refcount tracking for requests
1997 referenced from lustre intent.
1999 Severity : Enhancement
2001 Description: Cache open file handles on client.
2002 Details : MDS now will return special lock along with openhandle, if
2003 requested and client is allowed to hold openhandle, even if unused,
2004 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2005 closing files for every read/write openration.
2007 Severity : Enhancement
2009 Description: Cache open negative dentries on client when possible.
2010 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2011 negative dentries on lock revocation.
2016 Description: Remounting a client read-only wasn't possible with a zconf mount
2017 Details : It wasn't possible to remount a client read-only with llmount.
2019 Severity : enhancement
2020 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2021 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2022 MPICH is included to add Lustre-specific ADIO interfaces.
2023 This is based closely on the UFS ADIO layer and only differs
2024 in file creation, in order to allow the OST striping to be set.
2025 This is user-contributed code and not supported by CFS.
2030 Description: extended inode attributes (immutable, append-only) work improperly
2031 when 2.4 and 2.6 kernels are used on client/server or vice versa
2032 Details : Introduce kernel-independent values for these flags.
2034 Severity : enhancement
2037 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2038 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2039 granularity. Now, Fractional MB granularity is be supported,
2040 this is very useful for low memory system.
2042 Severity : enhancement
2044 Description: Getattr by fid
2045 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2046 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2047 have locks and avoids getattr rpc requests for it on every stat.
2050 Frequency : Always, for filesystems larger than 2TB
2052 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2053 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2054 filesystems larger than 2TB to be mounted with mballoc enabled.
2057 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2059 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2060 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2061 filesystems larger than 2TB is mounted then the kernel will
2062 silently corrupt the start of the filesystem. CONFIG_LBD is
2063 enabled for all CFS-supported kernels, but the possibility of
2064 this happening with a modified kernel config exists.
2066 Severity : enhancement
2068 Description: add client O_DIRECT support for 2.6 kernels
2069 Details : It is now possible to do O_DIRECT reads and writes to files
2070 in the Lustre client mountpoint on 2.6 kernel clients.
2072 Severity : enhancement
2074 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2075 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2076 parallel, not waiting for response from every OST before sending
2077 a rpc to the next OST.
2082 Description: setattr vs write race when updating file timestamps
2083 Details : Client processes that update a file timestamp into the past
2084 right after writing to the file (e.g. tar) it is possible that
2085 the updated file modification time can be reset to the current
2086 time due to a race between processing the setattr and write RPC.
2088 Severity : enhancement
2090 Description: Bring 'lfs find' closer in line with regular Linux find.
2091 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2092 -print0 options and obtains all the needed info through the lustre
2095 Severity : enhancement
2097 Description: support up to 1024 configured devices on one node
2098 Details : change obd_dev array from statically allocated to dynamically
2099 allocated structs as they are first used to reduce memory usage
2104 Description: Flush dirty partially truncated pages during truncate
2105 Details : Immediatelly flush partially truncated pages in filter_setattr,
2106 this way we completely avoid having any pages in page cache on OST
2107 and can retire ugly workarounds during writes to flush such pages.
2112 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2113 Details : protect inode from truncation within vfs_unlink() context
2114 just take a reference before calling vfs_unlink() and release it
2115 when parent's i_sem is free.
2120 Description: last_id value checked outside lock on OST caused LASSERT failure
2121 Details : If there were multiple MDS->OST object precreate requests in
2122 flight, it was possible that the OST's last object id was checked
2123 outside a lock and incorrectly tripped an assertion. Move checks
2124 inside locks, and discard old precreate requests.
2127 Frequency : always, if extents are used on OSTs
2129 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2130 Details : OSTs using the extents format would not zero the high 16 bits of
2131 the index physical block number. This is not a problem for any
2132 OST filesystems smaller than 16TB, and no kernels support ext3
2133 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2134 new/modified files) and can be fixed for existing filesystems
2135 with e2fsprogs-1.39-cfs1.
2140 Description: import connection selection may be incorrect if timer wraps
2141 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2142 import connections to be ignored if the 32-bit jiffies counter
2143 wraps. Use a 64-bit jiffies counter.
2146 Frequency : very large clusters immediately after boot
2148 Description: LNET request buffers exhausted under heavy short-term load
2149 Details : If a large number of client requests are generated on a service
2150 that has previously never seen so many requests it is possible
2151 that the request buffer growth cannot keep up with the spike in
2152 demand. Instead of dropping incoming requests, they are held in
2153 the LND until the RPC service can accept more requests.
2156 Frequency : Sometimes during replay
2158 Description: Assertion failure in ll_local_open after replay.
2159 Details : If replay happened on an open request reply before we were able
2160 to set replay handler, reply will become not swabbed tripping the
2161 assertion in ll_local_open. Now we set the handler right after
2162 recognising of open request
2165 Frequency : very rare
2167 Description: kernel reports "badness in vsnprintf"
2168 Details : Reading from the "recovery_status" /proc file in small chunks
2169 may cause a negative length in lprocfs_obd_rd_recovery_status()
2170 call to vsnprintf() (which is otherwise harmless). Exit early
2171 if there is no more space in the output buffer.
2173 Severity : enhancement
2175 Description: clear OBD RPC statistics by writing to them
2176 Details : It is now possible to clear the OBD RPC statistics by writing
2177 to the "stats" file.
2182 Description: Client mtime is not the same on different clients after utimes
2183 Details : In some cases, the client was using the utimes() syscall on
2184 a file cached on another node. The clients now validate the
2185 ctime from the MDS + OSTs to determine which one is right.
2190 Description: Inability to activate failout mode
2191 Details : lconf script incorrectly assumed that in pythong string's numeric
2192 value is used in comparisons.
2195 Frequency : always with multiple stripes per file
2197 Description: Inefficient object allocation for mutli-stripe files
2198 Details : When selecting which OSTs to stripe files over, for files with
2199 a stripe count that divides evenly into the number of OSTs,
2200 the MDS is always picking the same starting OST for each file.
2201 Return the OST selection heuristic to the original design.
2206 Description: mount failures may take full timeout to return an error
2207 Details : Under some heavy load conditions it is possible that a
2208 failed mount can wait for the full obd_timeout interval,
2209 possibly several minutes, before reporting an error.
2210 Instead return an error as soon as the status is known.
2212 Frequency : quota enabled and large files being deleted
2214 Description: releasing more than 4GB of quota at once hangs OST
2215 Details : If a user deletes more than 4GB of files on a single OST it
2216 will cause the OST to spin in an infinite loop. Release
2217 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2222 Description: statfs data retrieved from /proc may be stale or zero
2223 Details : When reading per-device statfs data from /proc, in the
2224 {kbytes,files}_{total,free,avail} files, it may appear
2225 as zero or be out of date.
2228 Frequency : systems with MD RAID1 external journal devices
2230 Description: lconf's call to blkid is confused by RAID1 journal devices
2231 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2232 of returning all block devices that match the journal UUID.
2235 Frequency : always, for aggregate stripe size over 4GB
2237 Description: assertion fails when trying to use 4GB stripe size
2238 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2239 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2242 Frequency : always on ppc64
2244 Description: the first write on an ext3 filesystem with mballoc got stuck
2245 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2246 perform endianness conversion.
2248 ------------------------------------------------------------------------------
2250 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2252 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2253 INTEROPERATE with older versions automatically. Please read the
2254 user documentation before upgrading any part of a live system.
2255 * WARNING: Lustre networking configuration changes are required with
2256 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2259 * Support for kernels:
2260 2.6.9-22.0.2.EL (RHEL 4)
2261 2.6.5-7.244 (SLES 9)
2262 2.6.12.6 vanilla (kernel.org)
2265 Severity : enhancement
2266 Bugzilla : 7981/8208
2267 Description: Introduced Lustre Networking (LNET)
2268 Details : LNET is new networking infrastructure for Lustre, it includes
2269 a reorganized network configuration mode (see the user
2270 documentation for full details) as well as support for routing
2271 between different network fabrics. Lustre Networking Devices
2272 (LNDs) for the supported network fabrics have also been
2273 created for this new infrastructure.
2275 Severity : enhancement
2276 Description: Introduced Access control lists
2277 Details : clients can set ACLs on files and directories in order to have
2278 more fine-grained permissions than the standard Unix UGO+RWX.
2279 The MDS must be started with the "-o acl" mount option.
2281 Severity : enhancement
2282 Description: Introduced filesystem quotas
2283 Details : Administrators may now establish per-user quotas on the
2286 Severity : enhancement
2288 Description: Configuration change for the XT3
2289 The PTLLND is now used to run Lustre over Portals on the XT3
2290 The configure option(s) --with-cray-portals are no longer used.
2291 Rather --with-portals=<path-to-portals-includes> is used to
2292 enable building on the XT3. In addition to enable XT3 specific
2293 features the option --enable-cray-xt3 must be used.
2298 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2299 Details : A race between allocating a new llcd and re-getting the llcd_lock
2300 allowed another thread to grab newly-allocated llcd.
2302 Severity : enhancement
2304 Description: 2.6 OST async journal commit and locking fix to improve performance
2305 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2306 2.6 kernels are now async as they already were in 2.4 kernels so
2307 that they can commit concurrently with the network bulk transfer.
2308 For block-allocated files the filter allocation semaphore is held
2309 to avoid filesystem fragmentation during allocation. BKL lock
2310 removed for 2.6 xattr operations where it is no longer needed.
2315 Description: lconf incorrectly determined whether two IP networks could talk
2316 Details : In some more complicated routing and multiple-network
2317 configurations, lconf will avoid trying to make a network
2318 connection to a disjoint part of the IP space. It was doing the
2319 math incorrectly for one set of cases.
2324 Description: Fix for potential infinite loop processing records in an llog.
2325 Details : If an llog record is corrupted/zeroed, it is possible to loop
2326 forever in llog_process(). Validate the llog record length
2327 and skip the remainder of the block on error.
2330 Frequency : occasional (liblustre only)
2332 Description: liblustre could not open files whose last component is a symlink
2333 Details : sysio_path_walk() would incorrectly pass the open intent to
2334 intermediate path components.
2337 Frequency : rare (liblustre only with non-standard tuning)
2338 Bugzilla : 7201 (7350)
2339 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2340 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2341 after it has been released, only do this for referenced locks
2343 Severity : enhancement
2345 Description: specifying an (invalid) directory default stripe_size of -1
2346 would reset the directory default striping
2347 Details : stripe_size -1 was used internally to signal directory stripe
2348 removal, now use "all default" to signal dir stripe removal
2349 as a directory striping of "all default" is not useful
2352 Frequency : common for large clusters running liblustre clients
2354 Description: doing an ls when liblustre clients are running is slow
2355 Details : sending a glimpse AST to a liblustre client waits for every AST
2356 to time out, as liblustre clients will not respond. Since they
2357 cannot cache data we refresh the OST lock LVB from disk instead.
2359 Severity : enhancement
2361 Description: doing an ls at the same time as file IO can be slow
2362 Details : enqueue and other "small" requests can be blocked behind many
2363 large IO requests. Create a new OST IO portal for non-IO
2364 requests so they can be processed faster.
2367 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2369 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2370 created file with mode 000
2371 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2372 allow an ftruncate() as a non-root user to a file with mode 000.
2373 Lustre now respects this flag to disable mode checks when
2374 truncating a file owned by the user
2377 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2380 Description: Revoking locks from clients that went dead or catatonic might take
2382 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2383 cancellation of such locks instant on servers without waiting for
2384 any reply from clients. Clients drops these locks when cancel
2385 notification from server is received without replying.
2388 Frequency : liblustre-only, when liblustre client dies or becomes busy
2390 Description: Doing ls on Linux clients can take a long time with active
2392 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2393 granting such locks to it in the first place if possible. Locks
2394 are taken by proxy on the OST during the read or write and
2395 dropped immediately afterward. Add connect flags handling, do
2396 not grant locks to liblustre clients for glimpse ASTs.
2398 Severity : enhancement
2400 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2401 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2402 sequential reads and behaves badly with applications doing random
2403 reads. Improve it by reading ahead at least read region, and
2404 avoiding excessive large RPC for small reads.
2406 Severity : enhancement
2408 Description: Creating more than 1000 files for a single job may cause a load
2409 imbalance on the OSTs if there are also a large number of OSTs.
2410 Details : qos_prep_create() uses an OST index reseed value that is an
2411 even multiple of the number of available OSTs so that if the
2412 reseed happens in the middle of the object allocation it will
2413 still utilize the OSTs as uniformly as possible.
2418 Description: OST or MDS may oops in ping_evictor_main()
2419 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2420 but doesn't restart at beginning of obd_exports_timed list
2423 Severity : enhancement
2425 Description: improve by-nid export eviction on the MDS and OST
2426 Details : allow multiple exports with the same NID to be evicted at one
2427 time without re-searching the exports list.
2430 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2432 Description: MDS may oops in groups_free()
2433 Details : in rare race conditions a newly allocated group_info struct is
2434 freed again, and this can be NULL. The 2.4 compatibility code
2435 for groups_free() checked for a NULL pointer, but 2.6 did not.
2438 Frequency : common for liblustre clients doing little filesystem IO
2439 Bugzilla : 9352, 7313
2440 Description: server may evict liblustre clients accessing contended locks
2441 Details : if a client is granted a lock or receives a completion AST
2442 with a blocking AST already set it would not reply to the AST
2443 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2446 Frequency : lfs setstripe, only systems with more than 160 OSTs
2448 Description: unable to set striping with a starting offset beyond OST 160
2449 Details : llapi_create_file() incorrectly limited the starting stripe
2450 index to the maximum single-file stripe count.
2453 Frequency : LDAP users only
2455 Description: lconf did not handle in-kernel recovery with LDAP properly
2456 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2458 Severity : enhancement
2460 Description: bind OST threads to NUMA nodes to improve performance
2461 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2462 node and do their allocations there to localize memory access
2464 Severity : enhancement
2466 Description: llmount can determine client NID directly from Myrinet (GM)
2467 Details : the client NID code from gmnalnid was moved directly into
2468 llmount, removing the need to use this or specifying the
2469 client NID explicitly when mounting GM clients with zeroconf
2472 Frequency : if client is started with down MDS
2474 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2475 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2476 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2477 ptlrpc_import_delayed_req() because we didn't check if the
2478 request was interrupted, and we also didn't break out of the
2479 event loop if there was no timeout
2484 Description: data loss during non-page-aligned writes to a single file from
2485 both multiple nodes and multiple threads on one node at same time
2486 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2487 inconsistency led to false short-reads, that were cached and later
2488 used by ->prepare_write() to fill in partially written page,
2489 leading to data loss.
2492 Frequency : always, if lconf --abort_recovery used
2494 Description: lconf --abort_recovery fails with 'Operation not supported'
2495 Details : lconf was attempting to abort recovery on the MDT device and not
2498 Severity : enhancement
2500 Description: remove cleanup logs
2501 Details : replace lconf-generated cleanup logs with lustre internal
2502 cleanup routines. Eliminates the need for client-cleanup and
2505 Severity : enhancement
2507 Description: add support for EAs (user and system) on lustre filesystems
2508 Details : it is now possible to store extended attributes in the Lustre
2509 client filesystem, and with the user_xattr mount option it
2510 is possible to allow users to store EAs on their files also
2512 Severity : enhancement
2514 Description: Add possibility (config option) to show minimal available OST free
2516 Details : When compiled with --enable-mindf configure option, statfs(2)
2517 (and so, df) will return least minimal free space available from
2518 all OSTs as amount of free space on FS, instead of summary of
2519 free spaces of all OSTs.
2521 Severity : enhancement
2523 Description: do not expand extent locks acquired on OST-side
2524 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2525 by server: they are not cached anyway.
2528 Frequency : when mmap is used/binaries executed from Lustre
2530 Description: Unmmap pages before throwing them away from read cache.
2531 Details : llap_shrink cache now attempts to unmap pages before discarding
2532 them (if unmapping failed - do not discard). SLES9 kernel has
2533 extra checks that trigger if this unmapping is not done first.
2538 Description: lconf didn't resolve symlinks before checking to see whether a
2539 given mountpoint was already in use
2542 Frequency : when migrating failover services
2543 Bugzilla : 6395, 9514
2544 Description: When migrating a subset of services from a node (e.g. failback
2545 from a failover service node) the remaining services would
2546 time out and evict clients.
2547 Details : lconf --force (implied by --failover) sets the global obd_timeout
2548 to 5 seconds in order to quickly disconnect, but this caused
2549 other RPCs to time out too quickly. Do not change the global
2550 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2552 Severity : enhancement
2553 Frequency : if MDS is started with down OST
2554 Bugzilla : 9439,5706
2555 Description: Allow startup/shutdown of an MDS without depending on the
2556 availability of the OSTs.
2557 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2558 Add appropriate locking and lov-osc refcounts for safe
2559 cleaning. Add osc abort_inflight calls in case the
2560 synchronize never started.
2563 Frequency : occasional (Cray XT3 only)
2565 Description: root not authorized to access files in CRAY_PORTALS environment
2566 Details : The client process capabilities were not honoured on the MDS in
2567 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2568 been accessed by an authorized user then root was able to access
2569 the file on the local client also. The root user capabilities
2570 are now allowed on the MDS, as this environment has secure UID.
2573 Frequency : occasional
2575 Description: ldiskfs "too long searching" message happens too often
2576 Details : A debugging message (otherwise harmless) prints too often on
2577 the OST console. This has been reduced to only happen when
2578 there are fragmentation problems on the filesystem.
2583 Description: Division by zero in statfs when all OSCs are inactive
2584 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2585 lov_statfs divides by value returned by lov_get_stripecnt().
2589 Bugzilla : 9489, 3273
2590 Description: First write from each client to each OST was only 4kB in size,
2591 to initialize client writeback cache, which caused sub-optimal
2592 RPCs and poor layout on disk for the first writen file.
2593 Details : Clients now request an initial cache grant at (re)connect time
2594 and so that they can start streaming writes to the cache right
2595 away and always do full-sized RPCs if there is enough data.
2596 If the OST is rebooted the client also re-establishes its grant
2597 so that client cached writes will be honoured under the grant.
2602 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2603 Details : Now I/O RPCs go to different portal number and (presumably) fast
2604 lock requests (and glimses) and other RPCs get their own service
2605 threads pool that should be able to service those RPCs
2608 Severity : enhancement
2610 Description: Ability to exchange lustre version between client and servers and
2611 issue warnings at client side if client is too old. Also for
2612 liblustre clients there is ability to refuse connection of too old
2614 Details : New 'version' field is added to connect data structure that is
2615 filled with version info. That info is later checked by server and
2619 Frequency : rare, liblustre only.
2620 Bugzilla : 9296, 9581
2621 Description: Two simultaneous writes from liblustre at offset within same page
2622 might proceed at the same time overwriting eachother with stale
2624 Details : I/O lock withing llu_file_prwv was released too early, before data
2625 actually was hitting the wire. Extended lock-holding time until
2626 server acknowledges receiving data.
2629 Frequency : extremely rare. Never observed in practice.
2631 Description: avoid generating lustre_handle cookie of 0.
2632 Details : class_handle_hash() generates handle cookies by incrementing
2633 global counter, and can hit 0 occasionaly (this is unlikely, but
2634 not impossible, because initial value of cookie counter is
2635 selected randonly). Value of 0 is used as a sentinel meaning
2636 "unassigned handle" --- avoid it. Also coalesce two critical
2637 sections in this function into one.
2639 Severity : enhancement
2641 Description: allow liblustre clients to delegate truncate locking to OST
2642 Details : To avoid overhead of locking, liblustre client instructs OST to
2643 take extent lock in ost_punch() on client's behalf. New connection
2644 flag is added to handle backward compatibility.
2646 Severity : enhancement
2647 Bugzilla : 4928, 7341, 9758
2648 Description: allow number of OST service threads to be specified
2649 Details : a module parameter allows the number of OST service threads
2650 to be specified via "options ost ost_num_threads={N}" in the
2651 OSS's /etc/modules.conf or /etc/modprobe.conf.
2655 Bugzilla : 6146, 9635, 9895
2656 Description: servers crash with bad pointer in target_handle_connect()
2657 Details : In rare cases when a client is reconnecting it was possible that
2658 the connection request was the last reference for that export.
2659 We would temporarily drop the export reference and get a new
2660 one, but this may have been the last reference and the export
2661 was just destroyed. Get new reference before dropping old one.
2663 Severity : enhancement
2664 Frequency : if client is started with failover MDS
2666 Description: Allow multiple MDS hostnames in the mount command
2667 Details : Try to read the configuration from all specified MDS
2668 hostnames during a client mount in case the "primary"
2671 Severity : enhancement
2673 Description: Stop sending data to evicted clients as soon as possible.
2674 Details : Check if the client we are about to send or are sending data to
2675 was evicted already. (Check is done every second of waiting,
2676 for which l_wait_event interface was extended to allow checking
2677 of exit condition at specified intervals).
2680 Frequency : rare, normally only when NFS exporting is done from client
2682 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2684 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2687 Frequency : always, if srand() is called before liblustre initialization
2689 Description: Liblustre uses system PRNG disturbing its usage by user application
2690 Details : Introduce internal to lustre fast and high-quality PRNG for
2691 lustre usage and make liblustre and some other places in generic
2692 lustre code to use it.
2694 Severity : enhancement
2695 Bugzilla : 9477, 9557, 9870
2696 Description: Verify that the MDS configuration logs are updated when xml is
2697 Details : Check if the .xml configuration logs are newer than the config
2698 logs stored on the MDS and report an error if this is the case.
2699 Request --write-conf, or allow starting with --old_conf.
2701 Severity : enhancement
2703 Description: Handle symlinks in the path when checking if Lustre is mounted.
2704 Details : Resolve intermediate symlinks when checking if a client has
2705 mounted a filesystem to avoid duplicate client mounts.
2710 Description: lconf can hit an error exception but still return success.
2711 Details : The lconf command catches the Command error exception at the top
2712 level script context and will exit with the associated exit
2713 status, but doesn't ensure that this exit status is non-zero.
2718 Description: failure of ptlrpc thread startup can cause oops
2719 Details : Starting a ptlrpc service thread can fail if there are a large
2720 number of threads or the server memory is very fragmented.
2721 Handle this without oopsing.
2724 Frequency : always, only if liblustre and non-default acceptor port was used
2726 Description: liblustre cannot connect to servers with non-default acceptor port
2727 Details : tcpnal_set_default_params() was not called and was therefore
2728 ignoring the environment varaible TCPNAL_PORT, as well as other
2729 TCPNAL_ environment variables
2734 Description: two objects could be created on the same OST for a single file
2735 Details : If an OST is down, in some cases it was possible to create two
2736 objects on a single OST for a single file. No problems other
2737 than potential performance impact and spurious error messages.
2741 Bugzilla : 5681, 9562
2742 Description: Client may oops in ll_unhash_aliases
2743 Details : Client dcache may become inconsistent in race condition.
2744 In some cases "getcwd" can fail if the current directory is
2750 Description: Inode refcounting problems in NFS export code
2751 Details : link_raw functions used to call d_instantiate without obtaining
2752 extra inode reference first.
2756 Bugzilla : 9942, 9903
2757 Description: Referencing freed requests leading to crash, memleaks with NFS.
2758 Details : We used to require that call to ll_revalidate_it was always
2759 followed by ll_lookup_it. Also with revalidate_special() it is
2760 possible to call ll_revalidate_it() twice for the same dentry
2761 even if first occurence returned success. This fix changes semantic
2762 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2763 reference on a request referred from the intent.
2764 ll_intent_release() then releases such a request.
2767 Frequency : rare, normally benchmark loads only
2769 Description: unlinked inodes were kept in memory on the client
2770 Details : If a client is repeatedly creating and unlinking files it
2771 can accumulate a lot of stale inodes in the inode slab cache.
2772 If there is no other client load running this can cause the
2773 client node to run out of memory. Instead flush old inodes
2774 from client cache that have the same inode number as a new inode.
2777 Frequency : SLES9 2.6.5 kernel and long filenames only
2778 Bugzilla : 9969, 10379
2779 Description: utime reports stale NFS file handle
2780 Details : SLES9 uses out-of-dentry names in some cases, which confused
2781 the lustre dentry revalidation. Change it to always use the
2785 Frequency : rare, unless heavy write-truncate concurrency is continuous
2786 Bugzilla : 4180, 6984, 7171, 9963, 9331
2787 Description: OST becomes very slow and/or deadlocked during object unlink
2788 Details : filter_destroy() was holding onto the parent directory lock
2789 while truncating+unlinking objects. For very large objects this
2790 may block other threads for a long time and slow overall OST
2791 responsiveness. It may also be possible to get a lock ordering
2792 deadlock in this case, or run out of journal credits because of
2793 the combined truncate+unlink. Solution is to do object truncate
2794 first in one transaction without parent lock, and then do the
2795 final unlink in a new transaction with the parent lock. This
2796 reduces the lock hold time dramatically.
2799 Frequency : rare, 2.4 kernels only
2801 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2802 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2803 before unmounting their filesystem in order to avoid deadlock.
2804 The kernel_locked() function in 2.4 kernels only checks whether
2805 the kernel lock is held, not whether it is this process that is
2806 holding it as 2.6 kernels do.
2811 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2812 Details : The client ptlrpc code may be trying to reconnect to a down
2813 server before a previous connection attempt has timed out.
2814 Increase the reconnect interval to be longer than the connection
2815 timeout interval to avoid sending duplicate connections to
2819 Frequency : echo_client brw_test command
2821 Description: fix echo_client to work with OST preallocated code
2822 Details : OST preallocation code (5137) didn't take echo_client IO path
2823 into account: echo_client calls filter methods outside of any
2824 OST thread and, hence, there is no per-thread preallocated
2825 pages and buffers to use. Solution: hijack pga pages for IO. As
2826 a byproduct, this avoids unnecessary data copying.
2830 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2831 Description: Client can oops in mdc_commit_close() after open replay
2832 Details : It was possible for the MDS to return an open request with no
2833 transaction number in mds_finish_transno() if the client was
2834 evicted, but without actually returning an error. Clients
2835 would later try to replay that open and may trip an assertion
2836 Simplify the client close codepath, and always return an error
2837 from the MDS in case the open is not successful.
2840 Frequency : rare, 2.6 OSTs only
2842 Description: OST may deadlock under high load on fragmented files
2843 Details : If there was a heavy load and highly-fragmented OST filesystems
2844 it was possible to have all the OST threads deadlock waiting on
2845 allocation of biovecs, because the biovecs were not released
2846 until the entire RPC IO was completed. Instead, release biovecs
2847 as soon as they are complete to ensure forward IO progress.
2849 Severity : enhancement
2851 Description: Support for specifying external journal device at mount
2852 Details : If an OST or MDS device is formatted with an external journal
2853 device, this device major/minor is stored in the ext3 superblock
2854 and may not be valid for failover. Allow detecting and
2855 specifying the external journal at mount time.
2860 Description: Mounting an MDS with pending unlinked files may cause oops
2861 Details : target_finish_recovery() calls mds_postrecov() which returned
2862 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2863 considers this an error and immediately begins cleaning up the
2864 lov, just after starting the mds_lov process
2866 Severity : enhancement
2868 Description: Implement 'lfs df' to report actual free space on per-OST basis
2869 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2870 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2871 usage of objects; '-h' to report in human readable format.
2873 ------------------------------------------------------------------------------
2875 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2882 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2883 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2884 at startup during orphan unlinking. Instead, skip buddy update
2885 before setup, as it will be regenerated after recovery is complete.
2890 Description: If an OST is inactive, its locks might reference stale inodes.
2891 Details : lov_change_cbdata() must iterate over all namespaces, even if
2892 they are inactive to clear inode references from the lock.
2894 Severity : enhancement
2895 Frequency : occasional, if non-standard max_dirty_mb used
2897 Description: Client will block write RPCs if not enough grant
2898 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2899 then the client will block writes while waiting for another RPC
2900 to complete instead of consuming its dirty limit. With change
2901 we get improved performance when max_dirty_mb is small.
2903 Severity : enhancement
2904 Bugzilla : 3389, 6253
2905 Description: Add support for supplementary groups on the MDS.
2906 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2907 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2908 lookups for user supplementary groups into a cache.
2912 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2913 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2914 for liblustre O_EXCL is enforced by MDS.
2918 Description: readdir never returns NULL in liblustre.
2919 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2920 dentry in struct dirent.
2924 Frequency : liblustre only, depends on application IO pattern
2925 Description: liblustre clients evicted if not contacting servers
2926 Details : Don't put liblustre clients into the ping_evictor list, so
2927 they will not be evicted by the pinger ever.
2929 Severity : enhancement
2931 Description: Add ability to evict clients by NID from MDS.
2932 Details : By echoing "nid:$NID" string into
2933 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2934 $NID would be instantly evicted from this MDS and from all active
2935 OSTs connected to it.
2939 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2940 Details : lookup_it_finish() used to query file size from OSTs that was not
2945 Description: service threads change working directory to that of init
2946 Details : Starting lustre service threads may pin the working directory
2947 of the parent thread, making that filesystem busy. Threads
2948 now change to the working directory of init to avoid this.
2952 Frequency : during shutdown only
2953 Description: shutdown with a failed MDS or OST can cause unmount to hang
2954 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2959 Frequency : 2.6 only
2960 Description: chmod/chown may include an extra supplementary group
2961 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2962 supplementary group and if none is specified this is used.
2965 Bugzilla : 5479 (6816)
2967 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2968 Details : If lookup is in progress on a file that is unlinked we might try
2969 to revalidate the inode and fail in revalidate after lookup is
2970 complete and ll_file_open() enqueues the open again but
2971 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2974 Frequency : always, if lconf --abort_recovery used
2976 Description: lconf --abort_recovery fails with 'Operation not supported'
2977 Details : lconf was attempting to abort recovery on the MDT device and not
2980 ------------------------------------------------------------------------------
2982 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2987 Frequency : rare (only unsupported configurations with a node running as an
2989 Bugzilla : 6514, 5137
2990 Description: Mounting a Lustre file system on a node running as an OST could
2992 Details : OSTs now preallocates memory needed to write out data at
2993 startup, instead of when needed, to avoid having to
2994 allocate memory in possibly low memory situations.
2995 Specifically, if the file system is mounted on on OST,
2996 memory pressure could force it to try to write out data,
2997 which it needed to allocate memory to do. Due to the low
2998 memory, it would be unable to do so and the node would
2999 become unresponsive.
3001 Severity : enhancement
3003 Description: Addition of lconf --service command line option
3004 Details : lconf now accepts a '--service <arg>' option, which is
3005 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3007 Severity : enhancement
3009 Description: Failover mode is now the default for OSTs.
3010 Details : By default, OSTs will now run in failover mode. To return to
3011 the old behaviour, add '--failout' to the lmc line for OSTs.
3013 Severity : enhancement
3015 Description: Health checks are now provided for MDS and OSTs
3016 Details : Additional detailed health check information on MSD and OSTs
3017 is now provided through the procfs health_check value.
3020 Frequency : occasional, depends on IO load
3022 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3023 after numerous create/delete cycles
3024 Details : The ext3 inode allocation policy would not allocate new inodes
3025 very well on the OSTs because there are no new directories
3026 being created. Instead we look for groups with free space if
3027 the parent directories are nearly full.
3032 Description: Network or server problems during mount may cause partially
3033 mounted clients instead of returning an error.
3034 Details : The config llog parsing code may overwrite the error return
3035 code during mount error handling, returning success instead
3041 Description: MDS can fail to allocate large reply buffers
3042 Details : After long uptimes the MDS can fail to allocate large reply
3043 buffers (e.g. zconf client mount config records) due to memory
3044 fragmentation or consumption by the buffer cache. Preallocate
3045 some large reply buffers so that these replies can be sent even
3046 under memory pressure.
3050 Frequency : rare (liblustre)
3051 Description: fsx running with liblustre complained that using truncate() to
3052 extend the file doesn't work. This patch corrects that issue.
3053 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3054 ATTR_SIZE and lsm use in llu_setattr_raw.
3058 Frequency : rare, only 2.6 kernels
3059 Description: Unusual file access patterns on the MDS may result in inode
3060 data being lost in very rare circumstances.
3061 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3062 code in the 2.6 kernel caused the inode and/or EA data not to
3063 be read from disk, causing single-file corruption.
3067 Frequency : rare, only 2.6 filesystems using extents
3068 Description: Heavy concurrent write and delete load may cause data corruption.
3069 Details : It was possible under high-load situations to have an extent
3070 metadata block in the block device cache from a just-unlinked
3071 file overwrite a newly-allocated data block. We now unmap any
3072 metadata buffers that alias just-allocated data blocks.
3076 Frequency : filesystems with default stripe_count larger than 77
3077 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3078 Details : lconf specifies an inode size of 4096 bytes when the default
3079 stripe_count is larger than 77. This conflicts with the default
3080 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3081 this case to avoid pinning too much memory for large EAs.
3083 ------------------------------------------------------------------------------
3085 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3090 Frequency : rare (extremely heavy IO load with hundreds of clients)
3092 Description: Client is evicted, gets IO error writing to file
3093 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3094 caused clients to be evicted for AST timeouts. The fixes in
3095 bug 5192 mean we no longer need to have such short AST timeouts
3096 so ldlm_timeout has been increased.
3099 Frequency : occasional during --force or --failover shutdown under load
3100 Bugzilla : 5949, 4834
3101 Description: Server oops/LBUG if stopped with --force or --failover under load
3102 Details : a collection of import/export refcount and cleanup ordering
3103 issues fixed for safer force cleanup
3106 Frequency : only filesystems larger than 120 OSTs
3107 Bugzilla : 5990, 6223
3108 Description: lfs getstripe would oops on a very large filesystem
3109 Details : lov_getconfig used kfree on vmalloc'd memory
3112 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3113 Bugzilla : 6242, 6243
3114 Description: reading from files that had been truncated to a non-zero size
3115 but never opened returned no data
3116 Details : ll_file_read() reads zeros from no-object files to EOF
3121 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3122 Details : There's little harm in aborting MDS/OSS recovery and letting it
3123 try again, so I removed the LASSERT and return an error instead.
3125 Severity : enhancement
3127 Description: New debugging infrastructure for tracking down data corruption
3128 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3129 (b) cover more of the client-side code path, and (c) try to narrow
3130 down where problems occurred
3134 Bugzilla : 3819, 4364, 4397, 6313
3135 Description: Racing close and eviction MDS could cause assertion in mds_close
3136 Details : It was possible to get multiple mfd references during close and
3137 client eviction, leading to one thread referencing a freed mfd.
3139 Severity: : enhancement
3140 Bugzilla : 3262, 6359
3141 Description: Attempts to reconnect to servers are now more aggressive.
3142 Details : This builds on the enhanced upcall-less recovery that was added
3143 in 1.4.2. When trying to reconnect to servers, clients will
3144 now try each server in the failover group every 10 seconds. By
3145 default, clients would previously try one server every 25 seconds.
3150 Description: After recovery, certain operations trigger a failed
3151 assertion on a client.
3152 Details : Failing over an mds, using lconf -d --failover, while a
3153 client was doing a readdir() call would cause the client to
3154 LBUG after recovery completed and the readdir() was resent.
3156 Severity : enhancement
3158 Description: Default groups are now added by lconf
3159 Details : You can now run lconf --group <servicename> without having to
3160 manually add groups with lmc.
3163 Frequency : occasional
3165 Description: Nodes with an elan id of 0 trigger a failed assertion
3168 Frequency : always when accessing e.g. tty/console device nodes
3170 Description: tty and some other devices nodes cannot be used on lustre
3171 Details : file's private_data field is used by device data and lustre
3172 values in there got lost. New field was added to struct file to
3173 store fs-specific private data.
3176 Frequency : when exporting Lustre via NFS
3178 Description: NFSD failed occasionally when looking up a path component
3179 Details : NFSD is looking up ".." which was broken in ext3 directories
3180 that had grown large enough to become hashed.
3183 Frequency : Clusters with multiple interfaces not on the same subnet
3185 Description: Nodes will repeatedly try to reconnect to an interface which it
3186 cannot reach and report an error to the log.
3187 Details : Extra peer list entries will be created by lconf with some peers
3188 unreachable. lconf now validates the peer before adding it.
3191 Frequency : Only if a default stripe is set on the filesystem root.
3193 Description: Setting a default stripe on the filesystem root prevented the
3194 filesystem from being remounted.
3195 Details : The client was sending extra request flags in the root getattr
3196 request and did not allocate a reply buffer for the dir EA.
3199 Frequency : occasional, higher if lots of files are accessed by one client
3200 Bugzilla : 6159, 6097
3201 Description: Client trips assertion regarding lsm mismatch/magic
3202 Details : While revalidating inodes the VFS looks up inodes with ifind()
3203 and in rare cases can find an inode that is being freed.
3204 The ll_test_inode() code will free the lsm during ifind()
3205 when it finds an existing inode and then the VFS later attaches
3206 this free lsm to a new inode.
3210 Bugzilla : 6422, 7030
3211 Description: MDS deadlock between mkdir and client eviction
3212 Details : Creating a new file via mkdir or mknod (starting a transaction
3213 and getting the ns lock) can deadlock with client eviction
3214 (gets ns lock and trying to finish a synchronous transaction).
3217 Frequency : occasional
3218 Description: While starting a server, the fsfilt_ext3 module could not be
3220 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3221 kernels. Previously, lconf would still use the ext3 name
3222 when trying to load modules. Now, it will correctly use
3223 ext3 on 2.4 and ldiskfs on 2.6.
3225 Severity : enhancement
3226 Description: The default stripe count has been changed to 1
3227 Details : The interpretation of the default stripe count (0, to lfs
3228 or lmc) has been changed to mean striping across a single
3229 OST, rather than all available. For general usage we have
3230 found a stripe count of 1 or 2 works best.
3232 Severity : enhancement
3233 Description: Add support for compiling against Cray portals.
3234 Details : Conditional compiling for some areas that are different
3238 Frequency : occasional
3239 Bugzilla : 6409, 6834
3240 Description: Creating files with an explicit stripe count may lead to
3241 a failed assertion on the MDS
3242 Details : If some OSTs are full or unavailable, creating files may
3243 trigger a failed assertion on the MDS. Now, Lustre will
3244 try to use other servers or return an error to the
3248 Frequency : occasional
3250 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3251 caused lock timeout during readahead (since 1.4.2).
3252 Details : Processes doing readahead might match a lock that hasn't been
3253 granted yet if there are overlapping and conflicting lock
3254 requests. The readahead process waits on ungranted lock
3255 (original lock is CBPENDING), while OST waits for that process
3256 to cancel CBPENDING read lock and eventually evicts client.
3258 Severity : enhancement
3260 Description: Initial enabling of flock support for clients
3261 Details : Implements fcntl advisory locking and file status functions.
3262 This feature is provided as an optional mount flag (default
3263 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3264 locking are implemented yet, and those that are are not guaranteed
3265 to be completely correct in production environments.
3266 mount -t lustre -o [flock|noflock] ...
3269 Frequency : occasional
3271 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3272 assertion in the ext3 JBD (journaling) layer.
3273 Details : The b_committed_data struct is protected by the big kernel lock
3274 in 2.4 kernels, serializing journal_commit_transaction() and
3275 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3276 access to this struct. In 2.6 kernels there is finer grained
3277 locking to improve SMP performance of the JBD layer.
3281 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3286 Description: merge and cleanup kernel patches.
3289 -----------------------------------------------------------------------------
3291 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3293 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3294 versions of Lustre. Please update all servers and clients to
3295 version 1.4.2 or later at the same time. You must also run
3296 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3297 to update the configuration logs.
3299 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3300 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3301 - swab llog records of type '0' so we get proper header size/idx (5861)
3302 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3303 - fix rename of one directory over another leaking an inode (5953)
3304 - avoid SetPageDirty on 2.6 (5981)
3305 - don't re-add just-being-destroyed locks to the waiting list (5653)
3306 - when creating new directories, inherit the parent's custom
3307 striping settings if present parent (3048)
3308 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3309 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3310 - don't hold client locks on temporary worklist from l_lru (5666)
3311 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3312 - automatically evict dead clients (5921)
3313 - Update file size properly in create+truncate+fstat case (6196)
3314 - Do not unhash mountpoint dentries, do not allow removal of
3316 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3317 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3318 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3319 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3320 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3321 - Correctly handle reads of files with no objects (6243)
3322 - lctl recover will also mark a device active if deactivate used (5933)
3324 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3325 - allow --write-conf on an MDS with different nettype than client (5619)
3326 - don't write config llogs to MDS for mounts not from that MDS (5617)
3327 - lconf should create multiple TCP connections from a client (5201)
3328 - init scripts are now turned off by default; run chkconfig --on
3329 lustre and chkconfig --on lustrefs to use them
3330 - upcalls are no longer needed for clients to recover to failover
3332 - add --abort-recovery option to lconf to abort recovery on device
3334 - add support for an arbitrary number of OSTs (3026)
3335 - Quota support protocol changes.
3336 - forward compatibility changes to wire structs (6007)
3337 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3338 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3339 - improved "lustre status" script
3340 - initialize blocksize for non-regular files (6062)
3341 - added --disable-server and --disable-client configure options (5782)
3342 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3343 - Vanilla 2.4.29 support
3344 - increase maximum number of obd devices to 520 (6242)
3345 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3346 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3347 SLES 9 2.6.5 kernels
3348 - limit stripes per file to 160 (the maximum EA size) (6093)
3350 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3353 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3354 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3355 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3356 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3357 - small ext3 extents cleanups and fixes (5733)
3358 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3359 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3360 - avoid needless client->OST connect, fix handle mismatch (5317)
3361 - fix DLM error path that led to out-of-sync client, long delays (5779)
3362 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3363 - fix several locking issues related to i_size (5492,5624,5654,5672)
3364 - don't move pending lock onto export if it is already evicted (5683)
3365 - fix kernel oops when creating .foo in unlinked directory (5548)
3366 - fix deadlock in obdfilter statistics vs. object create (5811)
3367 - use time_{before,after} to avoid timer jiffies wrap (5882)
3368 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3369 - Do not leak request if server was not able to process it (5154)
3370 - If mds_open unable to find parent dir, make that negative lookup(5154)
3371 - don't create new directories with extent-mapping (5909, 5936)
3373 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3374 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3375 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3376 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3377 - support for RHEL4 kernel on i686 (5773)
3378 - provide error messages when incompatible logs are encountered (5898)
3380 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3381 * version 1.4.0.10 (1.4.1 release candidate 1)
3383 - don't keep a lock reference when lock is not granted (4238)
3384 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3385 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3386 - drop import inflight refcount on signal_completed_replay error (5255)
3387 - unlock page after async write error during send (3677)
3388 - handle missing objects in filter_preprw_read properly (5265)
3389 - no transno return for symlink open, don't save no-trasno open (3440)
3390 - don't try to complete elan receive that already failed (4012)
3391 - free RPC server reply state on error (5406)
3392 - clean up thread from ptlrpc_start_thread() on error (5160)
3393 - readahead could read extra page into cache that wasn't ejected (5388)
3394 - prevent races in class_attach/setup/cleanup/detach (5260)
3395 - don't dereference de->d_inode after l_dput of de (5458)
3396 - use "int" for stripe value returned from lock_to_stripe (5544)
3397 - mballoc allocation and error-checking fixes in 2.6 (5504)
3398 - block device patches to fix I/O request sizes in 2.6 (5482)
3399 - look up hostnames for IB nals (5602)
3400 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3401 - don't start multiple acceptors for the same port (5277)
3402 - fix incorrect LASSERT in mds_getattr_name (5635)
3403 - export a proc file for general "ping" checking (5628)
3404 - fix "lfs check" to not block when the MDS is down (5628)
3406 - service request history (4965)
3407 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3408 - create an "evict_client" /proc entry on OSTs, like the MDS has
3409 - fix mount usage message, return errors per mount(8) (5168)
3410 - change grep [] to grep "[]" in tests so they work in more UMLs
3411 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3412 - remove ancient LOV_MAGIC_V0 EA support (5047)
3413 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3414 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3415 - allow readahead allocations to fail when low on memory (5383)
3416 - mmap locking landed again, after considerable improvement (2828)
3417 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3419 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3422 - send OST transaction number in read/write reply to free req (4966)
3423 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3424 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3425 - the watchdog thread now runs as interruptible (5246)
3426 - flock/lockf fixes (but it's still disabled, pending 5135)
3427 - don't use EXT3 constants in llite code (5094)
3428 - memory shortage at startup could cause assertion (5176)
3430 - reorganization of lov code
3431 - single portals codebase
3433 - add extents/mballoc support (5025)
3434 - direct I/O reads in the obdfilter (4048)
3435 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3437 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3440 - send OST transaction number in read/write reply to free req (4966)
3441 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3442 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3443 - the watchdog thread now runs as interruptible (5246)
3444 - handle missing objects in filter_preprw_read properly (5265)
3445 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3446 - drop import inflight refcount on signal_completed_replay error (5255)
3447 - unlock page after async write error during send (3677)
3448 - return original error code on reconstructed replies (3761)
3449 - no transno return for symlink open, don't save no-trasno open (3440)
3451 - add pid to ldlm debugging output (4922)
3452 - bump the watchdog timeouts -- we can't handle 30sec yet
3453 - extra debugging for orphan dentry/inode bug (5259)
3455 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3458 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3459 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3460 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3461 - fix NULL dereference in /proc/sys/portals/routes (4827)
3462 - allow failed mdc_close() operations to be interrupted (4561)
3463 - stop precreate on OST before MDS would time out on it (4778)
3464 - don't send partial-page writes before EOF from client (4410)
3465 - discard client grant for sub-page writes on large-page clients (4520)
3466 - don't free dentries not owned by NFS code, check generation (4806)
3467 - fix lsm leak if mds_create_objects() fails (4801)
3468 - limit debug_daemon file size, always print CERROR messages (4789)
3469 - use transno after validating reply (3892)
3470 - process timed out requests if import state changes (3754)
3471 - update mtime on OST during writes, return in glimpse (4829)
3472 - add mkfsoptions to LDAP (4679)
3473 - use ->max_readahead method instead of zapping global ra (5039)
3474 - don't interrupt __l_wait_event() during strace
3476 - add software watchdogs to catch hung threads quickly (4941)
3477 - make lustrefs init script start after nfs is mounted
3478 - fix CWARN/ERROR duplication (4930)
3479 - return async write errors to application if possible (2248)
3480 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3481 - print NAL number in %x format (4645)
3482 - update barely-supported suse-2.4.21-171 series (4842)
3483 - support for sles 9 %post scripts
3484 - support for building 2.6 kernel-source packages
3485 - support for sles km_* packages
3487 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3490 - ignore -ENOENT errors in osc_destroy (3639)
3491 - notify osc create thread that OSC is being cleaned up (4600)
3492 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3493 - reconstruct ost_handle() like mds_handle() (4657)
3494 - create a new thread to do import eviction to avoid deadlock (3969)
3495 - let lconf resolve symlinked-to devices (4629)
3496 - don't unlink "objects" from directory with default EA (4554)
3497 - hold socknal file ref over connect in case target is down (4394)
3498 - allow more than 32000 subdirectories in a single directory (3244)
3499 - fix blocks count for O_DIRECT writes (3751)
3500 - OST returns ENOSPC from object create when no space left (4539)
3501 - don't send truncate RPC if file size isn't changing (4410)
3502 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3503 - bind to privileged port in socknal and tcpnal (3689)
3505 - rate limit CERROR/CWARN console message to avoid overload (4519)
3506 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3507 - basic mmap support (3918)
3508 - kernel patch series update from b1_4 (4711)
3510 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3513 - avoid crash during MDS cleanup with OST shut down (2775)
3514 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3515 - don't use bad inodes on the MDS (3744)
3516 - dynamic object preallocation to improve recovery speed (4236)
3517 - don't hold spinlock over lock dumping or change debug flags (4401)
3518 - don't zero obd_dev when it is force cleaned (3651)
3519 - print grants to console if they go negative (4431)
3520 - "lctl deactivate" will stop automatic recovery attempts (3406)
3521 - look for existing locks in ldlm_handle_enqueue() (3764)
3522 - don't resolve lock handle twice in recovery avoiding race (4401)
3523 - revalidate should check working dir is a directory (4134)
3525 - don't always mark "slow" obdfilter messages as errors (4418)
3527 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3530 - don't close LustreDB during write_conf until it is done (3860)
3531 - fix typo in lconf for_each_profile (3821)
3532 - allow dumping logs from multiple threads at one time (3820)
3533 - don't allow multiple threads in OSC recovery (3812)
3534 - fix debug_size parameters (3864)
3535 - fix mds_postrecov to initialize import for llog ctxt (3121)
3536 - replace config semaphore with spinlock (3306)
3537 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3538 - don't allow enqueue to complete on a destroyed export (3822)
3539 - down write_lock before checking llog header bitmap (3825)
3540 - recover from lock replay timeout (3764)
3541 - up llog sem before sending rpc (3652)
3542 - reduce ns lock hold times when setting kms (3267)
3543 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3544 - fix NULL deref and obd_dev leak on setup error (3312)
3545 - replace some LBUG about llog ops with error handling (3841)
3546 - don't match INVALID dentries from d_lookup and spin (3784)
3547 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3548 - fix invalid assertion in ptlrpc_set_wait (3880)
3550 - add libwrap support for the TCP acceptor (3996)
3551 - add /proc/sys/portals/routes for non-root route listing (3994)
3552 - allow setting MDS UUID in .xml (2580)
3553 - print the stack of a process that LBUGs (4228)
3555 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3558 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3559 - make sure to unset replay flag from failed open requests (3440)
3560 - if default stripe count is 0, use OST count for inode size (3636)
3561 - update parent mtime/ctime on client for create/unlink (2611)
3562 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3563 - free recovery state on server during a forced cleanup (3571)
3564 - unregister_reply for resent reqs (3063)
3565 - loop back devices mounting and status check on 2.6 (3563)
3566 - fix resource-creation race that can provoke i_size == 0 (3513)
3567 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3568 - more debugging for page-accounting assertion (3746)
3569 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3570 - avoid deadlock after precreation failure (3758)
3571 - fix race and lock order deadlock in orphan handling (3450, 3750)
3572 - add validity checks when grabbing inodes from l_ast_data (3599)
3574 - add /proc/.../recovery_status to obdfilter (3428)
3575 - lightweight CDEBUG infrastructure, debug daemon (3668)
3576 - change default OSC RPC parameters to be better on small clusters
3577 - turn off OST read cache for files smaller than 32MB
3578 - install man pages and include them in rpms (3100)
3579 - add new init script for (un)mounting lustre filesystems (2593)
3580 - run chkconfig in %post for init scripts (3701)
3581 - drop scimac NAL (unmaintained)
3583 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3586 - clean kiobufs before and after use (3485)
3587 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3588 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3589 - change init script to fail more clearly if not run as root (1528)
3590 - allow clients to reconnect during replay (1742)
3591 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3592 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3593 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3594 - fix cond_resched (3554)
3595 - only evict unfinished clients after recovery (3515)
3596 - allow bulk resend, prevent data loss (3570)
3597 - dynamic ptlrpc request buffer allocation (2102)
3598 - don't allow unlinking open directory if it isn't empty (2904)
3599 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3600 - remove extraneous obd dereference causing LASSERT failure (3334)
3601 - don't use get_cycles() when creating temp. files on the mds (3156)
3602 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3603 - handle EEXIST for set-stripe, set proper directory name (3336)
3605 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3606 - fix ksocknal_fmb_callback() error messages (2918)
3608 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3611 - don't copy lvb into (possibly NULL) reply on error (2983)
3612 - don't deref dentry after dput, don't free lvb on error (2922)
3613 - use the kms to determine writeback rpc length (2947)
3614 - increment oti_logcookies when osc is inactive (2948)
3615 - update client's i_blocks count via lvb messages (2543)
3616 - handle intent open/close of special files properly (1557)
3617 - mount MDS with errors=remount-ro, like obdfilter (2009)
3618 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3619 - don't use cancelling-locks' kms values (2947)
3620 - use highest lock extent for kms, not last one (2925)
3621 - don't dereference ERR_PTR() dentry in error handling path (3107)
3622 - fix thread race in portals_debug_dumplog() (3122)
3623 - create lprocfs device entries at setup instead of at attach (1519)
3624 - common AST error handler, don't evict client on completion race (3145)
3625 - zero nameidata in detach_mnt in 2.6 (3118)
3626 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3627 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3628 - initialize RPC timeout timer earlier for 2.6 (3219)
3629 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3630 - print nal/nid for unknown nid (3258)
3631 - additional checks for oscc recovery before doing precreate (3284)
3632 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3633 - don't crash in mdc_close for bad permissions on open (3285)
3634 - zero i_rdev for non-device files (3147)
3635 - clear page->private before handing to FS, better assertion (3119)
3636 - tune the read pipeline (3236)
3637 - fix incorrect decref of invalidated dentry (2350)
3638 - provide read-ahead stats and refine rpc in flight stats (3328)
3639 - don't hold journal transaction open across create RPC (3313)
3640 - update atime on MDS at close time (3265)
3641 - close LDAP connection when recovering to avoid server load (3315)
3642 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3643 - don't leak open file on MDS after open resend (3325)
3644 - serialize filter_precreate and filter_destroy_precreated (3329)
3645 - loop device shouldn't call sync_dev() for nul device (3092)
3646 - clear page cache after eviction (2766)
3647 - resynchronize MDS->OST in background (2824)
3648 - refuse to mount the same filesystem twice on same mountpoint (3394)
3649 - allow llmount to create routes for mounting behind routers (3320)
3650 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3651 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3652 - fix rare problem with rename on htree directories (3417)
3654 - allow default OST striping configuration per directory (1414)
3655 - fix compilation for qswnal for 2.6 kernels (3125)
3656 - increase maximum number of MDS request buffers for large systems
3657 - change liblustreapi to be useful for external progs like lfsck (3098)
3658 - increase local configuration timeout for slow disks (3353)
3659 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3661 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3664 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3665 - don't overwrite extent policy data in reply if lock was blocked (2901)
3666 - drop filter export grants atomically with removal from device (2663)
3667 - del obd_self_export from work_list in class_disconnect_exports (2908)
3668 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3669 - swab reply message in mdc_close, other PPC fixes (2464)
3670 - fix destroying of named logs (2325)
3671 - overwrite old logs when running lconf --write_conf (2264)
3672 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3673 - fix race in target_handle_connect (2898)
3674 - mds_reint_create() should take same inode create lock (2926)
3675 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3676 - don't close files for self_export to avoid uninitialized obd (2936)
3677 - allow MDS with the same name as client node (2939)
3678 - hold dentry reference for closed log files for unlink (2325)
3679 - reserve space for all logs during transactions (2059)
3680 - don't evict page beyond end of stripe extent (2925)
3681 - don't oops on a deleted current working directory (2399)
3682 - handle hard links to targets without a parent properly (2517)
3683 - don't dereference NULL lock when racing during eviction (2867)
3684 - don't grow lock extents when lots of conflicting locks (2919)
3686 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3689 - account for cache space usage on clients to avoid data loss (974)
3690 - lfsck support in lustre kernel code (2349)
3691 - reduce journal credits needed for BRW writes (2370)
3692 - orphan handling to avoid losing space on client/server crashes
3693 - ptlrpcd can be blocked, stopping ALL progress (2477)
3694 - use lock value blocks to assist in proper KMS, faster stat (1021)
3695 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3696 - recovery for initial connections (2355)
3697 - fixes for mds_cleanup_orphans (1934)
3698 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3699 - block all file creations until orphan recovery completes (1901)
3700 - client remove rq_connection from request struct (2423)
3701 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3702 - recovery timer race (2670)
3703 - mdc_close recovey bug (2532)
3704 - ptlrpc cleanup bug (2710)
3705 - mds timeout on local locks (2588)
3706 - namespace lock held during RPCs (2431)
3707 - handle interrupted sync write properly (2503)
3708 - don't try to handle a message that hasn't been replied to (2699)
3709 - client assert failure during cleanup after abort recovery (2701)
3710 - leak mdc device after failed mount (2712)
3711 - ptlrpc_check_set allows timedout requests to complete (2714)
3712 - wait for inflight reqs when ptlrpcd finishes (2710)
3713 - make sure unregistered services are removed from the srv_list
3714 - reset bulk XID's when resending them (caught by 1138 test)
3715 - unregister_bulk after timeout
3716 - fix lconf error (2694)
3717 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3718 - readahead locks pages, leaves pending causing memory pressure (2673)
3719 - increase OST request buffers to 4096 on large machines (2729)
3720 - fix up permission of existing directories in simple_mkdir (2661)
3721 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3722 - don't assign transno to errored transactions (2742)
3723 - don't delete objects on OST if given a bogus objid from MDS (2751)
3724 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3725 - if rq_no_resend, then timeout request after recovery (2432)
3726 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3727 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3728 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3729 - MDS recovery shouldn't delete objects that it has given out (2730)
3730 - if enqueue arrives after completion, don't clobber LVB (2819)
3731 - don't unlock pages twice when trigger_group_io returns error (2814)
3732 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3733 - don't write pages to disk if there was an error (1450)
3734 - don't ping imports that have recovery disabled (2676)
3735 - take buffered bytes into account when balancing socknal conn (2817)
3736 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3737 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3738 - remove little-endian swabbing of llog records (1987)
3739 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3740 - retry reposting request buffers if they fail (1191)
3741 - grow extent at grant time to avoid granting a revoked lock (2809)
3742 - lock revoke doesn't evict page if covered by a second lock (2765)
3743 - disable VM readahead to avoid reading outside lock extents (2805)
3745 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3746 - updated kernel patches for hp-2.4.20 kernel (2681)
3748 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3751 - fix truncated write corruption (2366)
3752 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3754 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3755 - ensure len(uuid) < 37 in lmc (1171)
3756 - fix ia64 OOPS in llog_test (2255)
3757 - zero end of page at obdfilter for partial page writes (2648)
3758 - don't leave stale dentries around after renames (bug 2428)
3759 - fix timeouts when evicting a client with a single lock held (2642)
3760 - set deadline for the initial HELLO message to drain (2634)
3761 - print out dotted-quad IP addresses in the socknal (2302)
3763 - additional debugging for MDS client eviction problem (2443)
3764 - fix mkfsoptions support for osts (2603, 2604)
3766 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3769 - add series for the vanilla 2.6.0 kernel
3770 - add series for the vanilla 2.4.24 kernel
3771 - add series for a cray x86/64 UL kernel drop
3772 - fix xattr patches for the vanilla 2.4.19 series
3774 - generate true UUIDs in lmc (1171)
3775 - have portals stack dumping break in UML (2466)
3776 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3777 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3778 - wake callback waiting threads on client eviction (2460)
3779 - Add --ptldebug and --subsystem to lmc (1719)
3780 - update assertion to allow safe interrupt allocation
3781 - set rq_no_resend for cancel requests (2432)
3782 - recalculate ptlrpcd timeout after resend (2494)
3783 - call vfs_rmdir when removing pending directories (2368)
3784 - fix renaming a file to itself (2429)
3785 - lmc creates a default one-stripe lov (2454)
3786 - expand procfs space to handle large clusters (2326)
3787 - increase UML stack to avoid overflow
3788 - update lconf's list of debug and subsystem masks
3789 - fix lfs find --obd (2510)
3790 - /proc tunable for disabling filter read caching (2591)
3791 - stop rpm packages from altering slapd.conf (2301)
3792 - disable nagle in the socknal under 0conf (2578)
3793 - choose mds inode size based on stripe count (2572)
3794 - fix kernel-source rpm problems (2516)
3796 - add --disable-doc to avoid pdf generation (2421)
3797 - update documentation, tests, type-os, comments
3798 - avoid format warnings on ia64
3799 - remove the TOE NAL
3800 - tiny code cleanups by removing unused fields
3802 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3805 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3806 - remove the most blatant lies from BUILDING file (2371)
3807 - change default debug level to reasonable production setting
3808 - reduce client side cache size to reduce cache flush time
3809 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3810 - make TCP zerocopy and pinger support enabled by default (2476)
3811 - sync writes completed after process exits caused crashes (2319)
3812 - maintain correct mount count on the MDS (2356)
3813 - backout 1557, because 2316 wasn't really fixed
3814 - better file I/O statistics gathering in /proc
3815 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3816 - another kernel patch to fix zero-copy TCP function export
3817 - don't take duplicate lock when processing re-sent getattr (2420)
3818 - lctl uses obd_self_export instead of creating new conn (2353)
3819 - MDS/OST recovery case which requires object creation asserted (2425)
3820 - move lfs from /usr/sbin to /usr/bin in packages
3821 - fix race between mds_client_add and mds_client_free (2417)
3822 - use kmalloc instead of slabs in portals (2430)
3823 - don't create duplicate records when a failover MDS is present (2442)
3824 - remove unnecessary mount age check (2332)
3825 - don't remove directory inodes from locks prematurely (2451)
3826 - don't break if MDS service name is the same as hostname (2103)
3827 - fix races in client write RPC generation when cache full (2482)
3829 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3832 - remove now-unused request->rq_obd (278)
3833 - if an allocation fails, print out how much memory we've used (1933)
3834 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3835 - add the "configurable stack size" patch to most series files (1256)
3836 - ability to write large log records, for 100+ OST configs (2306)
3837 - fix NULL deref when filter_prep fails (2314)
3838 - fix operator precedence error in filter_sync
3839 - dynamic allocation of socknal TX descriptors (2315)
3840 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3841 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3842 - ensure that CWARN messages in Portals always get to the syslog
3843 - __init/__exit are not for prototype decls (ldlm_init/exit)
3844 - x86-64 compile warning fixes
3845 - fix gateway LMC keyword conflict (2318)
3846 - fix MDS lock inversions in getattr/reint paths (1844)
3847 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3848 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3849 - DLM race condition prevented some lock evictions (2328)
3850 - ENOMEM detection and retry on socknal sends (2230)
3851 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3852 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3853 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3854 - better error messages when a client is rejected during recovery (1505)
3855 - avoid cancelling locks which were never granted, after failure (2330)
3856 - fix i_sem/journal inversion in mds_client_add (2333)
3857 - fix truncate/getattr lock cycle deadlock (2334)
3858 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3859 - fix two rare exit paths which could leak an l_lock() ref (2321)
3860 - fencepost error in MDS/OST orphan recovery (2226)
3861 - make log record alignment 8 bytes (1988)
3862 - lstripe now fails when requested offset > ost_count (2237)
3863 - ensure that all kernel series have a complete list.h (1607)
3864 - fix crashes in special-file operations (2316)
3865 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3867 - allow configurable automake binary, for testing new versions
3868 - small update to the lfs documentation
3870 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3872 * fix negative export reference count in fsfilt_sync (2312)
3874 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3875 * release candidate 0.9.1
3877 - orphans are moved into the PENDING directory for possible recovery
3878 - replayed opens now open by fid for orphan/rename safety (1042)
3879 - last close of an orphan inode generates a transno (683)
3880 - chdir() and mount() now pin the directory entry (1020)
3881 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3882 - discard very old requests without processing them (1502)
3883 - remove some common, well-understood CERRORs (1505)
3884 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3885 - clear "grant" flags in OST replies until OST grant code lands (1644)
3886 - fix read performance by not clobbering i_blksize on client (1598)
3887 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3888 - make LRU size a /proc tunable, clears locks when reduced (707)
3889 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3890 - server locks take references on exports now (1558)
3891 - build fixes for 2.4.20-rh trees (1663)
3892 - return an error from lov_create if all OSCs are inactive (1751)
3893 - fix import levels when a reconnect happens without a timeout (1597)
3894 - exit early from mds_open if we get a lookup error (1749)
3895 - partial page read at EOF wouldn't wait for disk before sending (1642)
3896 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3897 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3898 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3899 - write/truncate lock inversion (1639)
3900 - Don't auto-load obdclass, portals modules during cleanup (1495)
3901 - fix timestamps from jumping to "now" (1763)
3902 - extra journal assertions (1648)
3903 - add an extra multiunlink test (1771)
3904 - fix read_record/write_record API (1776)
3905 - fix leak of offset_extent, possible incorrect i_size later (1772)
3906 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3907 - Debugging for the kqswnal_get_idle_tx problems (1820)
3908 - Allow recovery to be attempted multiple times (1536)
3909 - Write out MDS last_rcvd file after it is first created (1600)
3910 - Fix tx_descriptor leak in failed transmit situations (1827)
3911 - ext3 journaling fixes for assertion failure after IO error (1871)
3912 - class_export_put() on freed export after completion AST error (1896)
3913 - Fix revalidate looping in VFS (1322)
3914 - Don't access a freed export during MDS_REINT timeout (1521)
3915 - Add open-unlink recovery support on the MDS (1673,1764)
3916 - Return an error if no MDS data was read from last_rcvd (1946)
3917 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3918 - Don't setup a disk device that doesn't match exported UUID (317)
3919 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3920 - avoid committing NULL handle in force close
3921 - local.sh is now a one-stripe LOV configuration
3922 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3923 - fix ext3 htree duplicate directory entry corruption (1516)
3924 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3925 - update to new LOV EA format (2097)
3926 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3928 2003-06-15 Phil Schwan <phil@clusterfs.com>
3931 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3932 - per-import/export recovery handling (958, 931, 959)
3933 - multiple last-rcvd slots, for serving multiple FSes (949)
3934 - connections are again shared between multiple imp/exports (963, 964)
3935 - "umount -f" would hang if any requests needed to be sent (393, 978)
3936 - avoid pinning large req buffer by copying for queued messages (989)
3937 - add "uuid" to "lctl device" command to help upcalls (991)
3938 - "open" RPCs with transnos would confuse recovery counters (1037)
3939 - do proper endian conversion of all wire messages (288, 340, 891)
3940 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3941 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3942 - fix ptlrpc_connection leak in target_handle_connect (1174)
3943 - fix import refcounting bug in OST and MDS cleanup (1134)
3944 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3945 - fix very unlikely obd_types race condition (501)
3946 - remove osc_open hack for echo_client (1187)
3947 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3948 - a failure in read_inode2 leads to deadlock (1139)
3949 - cancel ack-locks as soon as transaction is committed (1072)
3950 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3951 - make sure to commitrw after any preprw to avoid deadlock (1162)
3952 - failing to execute a file in a lustre FS would lock inode (1203)
3953 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3954 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3955 - obd_uuid2tgt would walk off the end of the list (1255)
3956 - on IA64 the osc would give portals incorrect bulk size (1258)
3957 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3958 - fix lock inversion caused by new llite matching code (1282)
3959 - limit the number of dirty pages on a client to 10MB (1286)
3960 - timed out locks were not being corrected cancelled (1289)
3961 - fix O_DIRECT above 4GB on IA-32 (1292)
3962 * major user-visible changes
3963 - fail out/fail over policy now controlled by the upcall (993)
3965 - add OBD_PING to check server availability and failure (954)
3966 - lustre messages are now sent in sending host order (288, 340, 891)
3967 - add eadatalen to MDS getattr reply (340)
3968 - OST read replies may contain second buffer, with per-page status (593)
3970 2003-03-11 Phil Schwan <phil@clusterfs.com>
3973 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3974 - failed open()s could cause deadlock; fixed (867, 869)
3975 - stop cancelling OST locks when files are closed (481)
3976 - overlapping XID spaces caused network corruption (851, 853)
3977 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3978 - setattr_raw added, to do single-RPC, server-side setattrs
3979 - lmc/lconf syntax change for OST UUIDs
3980 - fix crashy race condition between ptlrpc_free_req and osc_close
3981 - don't use request in mdc_enqueue if we hit a timeout (889)
3982 - don't set the inode i_size for regular files from the MDS (896)
3983 - handle out of order completion AST (842)
3984 - don't LBUG if a lock request times out after receiving AST (913)
3985 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3986 - if a bad lock AST arrives, send an error instead of dropping entirely
3987 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3988 - fix leak in bulk IO when only partially completed (899, 900, 926)
3989 - fix O_DIRECT for ia64 (55)
3990 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3991 - C-z after timeout could hang a process forever; fixed (977)
3993 - client-side I/O cache (678, 924, 929, 941, 970)
3995 - READPAGE and SETATTRs which don't take server-side locks get
3998 2003-02-11 Phil Schwan <phil@clusterfs.com>
4001 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4002 - Fix fsfilt_extN_readpage() to read a full page of directory
4003 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4004 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4005 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4006 - Most intent operations were removed, in favour of a new RPC mode
4007 that does a single RPC to the server and bypasses most of the VFS
4008 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4009 - Aggressively cancel local locks on DLM servers
4010 - mds_reint_unlink sends EA to the client if it's the last nlink.
4011 client uses that EA to unlink OST objects.
4012 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4013 - recursive symlinks were fixed (439)
4014 - fixed NULL deref in DEBUG_REQ
4015 - filter_update_lastobjid no longer calls sync, which annoyed extN
4016 - fixed multi-client small-writes to a single file problem (445)
4017 - fixed mtime updates during file writes (607)
4018 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4019 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4020 - fixed Linux OST size reporting problem (444, 656)
4021 - OST now updates object mtime with writes or setattr (607, 619)
4022 - client verifies file size before zeroing page past EOF (445)
4023 - OST now writes last allocated objid to disk with allocation (108)
4024 - LOV on echo now works (409)
4026 - mds_reint_unlink sends a new buffer, with the EA included. this
4027 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4028 set if a regular file was being unlinked, and it was the last link
4029 - use PtlGet from the target for bulk writes (315)
4030 - OST now updates object mtime with writes or setattr (607, 619)
4031 - LDLM now has a grant-time callback to revalidate locked items, if
4033 - Many MDS operations were reorganized to combat race conditions
4035 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4036 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4038 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4041 - Fully reactivate OST imports after reconnection (512, others)
4042 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4043 - More graceful error handling for truncating on dead OST (515)
4044 - Don't error out unless we're actually accessing dead stripes (474)
4045 - Fix garbage sizes when stripes are missing (410)
4046 - LRU counters were broken, causing constant lock purge (433, 432)
4047 - garbage on read from stripes with failed OSTs (441)
4048 - mark OSCs as active before reconnecting during recovery (438)
4049 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4050 - lfind did not preserve OST order in output (443)
4051 - symlinks cause hung clients, incorrect data (439)
4052 - stop dereferencing request after dropping refcount (457)
4053 - don't LASSERT(spin_is_locked) on non-SMP (455)
4054 - fixes for many rename() bugs
4055 - fstat didn't correctly synchronize attributes (399)
4056 - server must handle lock cancellation during blocking AST prep (487)
4057 - bulk descriptors were free()d too soon (511)
4058 - fix paths in lconf, which would load incorrect modules (451, 507)
4059 - fix confusing lconf 'host not found' error message (386)
4060 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4061 - fix race condition in mdc_blocking_ast() for inode access (526)
4062 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4063 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4064 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4065 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4066 - fix lstripe in several ways (kernel side) (527)
4067 - fix request leak in ldlm_cli_enqueue (262)
4068 - incorrect OSC was marked inactive after OST failure
4069 - call mds_fs_cleanup before unmounting filesystem (524)
4070 - fix races between taking ns_lock and ldlm_lock_change_resource
4071 - fix races updating LOV export open file list
4072 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4073 - fix recovery NULL deref in ldlm_cli_cancel_unused
4074 - fix some DLM races by using new hash table for lock handles (419)
4075 - permit the client to specify desired inodes, at replay
4076 - duplicate requests when we queue them for replay reintegration
4077 - fix last_rcvd offset calculation
4078 - sync after each recovered transaction, so we always make progress
4079 - never, not always, ERESTART requests without transnos
4080 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4082 - skip replay if the MDS says that the client is already connected
4083 - don't check for a recovery-enabled export to match lctl's UUID
4084 - don't INC_USE_COUNT for phantom exports
4085 - don't crash when cleaning up phantom exports (567)
4086 - don't double-finish or set replay data for errored mdc_open requests
4087 - abort requests when they time out, so we don't get old replies
4088 - send/receive replies for AST messages again
4089 - if the client says that it doesn't have the lock, cancel it on the
4091 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4092 mark it as destroyed, it will all work out when decref is called
4093 - fix module use counts (22, 581)
4095 - ASTs now expect a reply (server cancels lock on error reply)
4097 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4100 - fix many simultaneous client startup (392)
4101 - fix dentry->d_it clobbering
4102 - credentials weren't being shipped for readdir/getattr operations
4103 - remove invalid assertions triggered during some concurrent MD
4105 - proper Lustre versions added (336, 389)
4106 - fix memory leak for create error case (398)
4107 - fix LOV locking bug that would get cli/srv out of sync
4108 - fix echo client over LOV (409)
4109 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4110 - fix double-O_EXCL intent crash (424)
4111 - avoid sending multiple lock CANCELs (352)
4113 - MDS can do multi-client recovery (modulo bugs in new code)
4115 - many updates, edits, cleanups
4117 2002-11-18 Phil Schwan <phil@clusterfs.com>
4120 - fix null d_it dereference (346)
4121 - fix full OST/dbench hang (333)
4122 - fix permission problem with file removal (286)
4123 - fix removal of OSCs from LOV when they fail
4124 - fix NULL deref during bulk timeout (214)
4125 - fix problems related to multiple filesystems on one MDS (241)
4126 - fixed serious subtle metadata locking bugs
4127 - free locks on clients when inodes are removed due to memory
4129 - fix inode pointer in lock data (285)
4130 - partial support for multiple MDS on a single host (241)
4131 - data locks weren't cancelled at clear_inode time (290, 311)
4132 - intent locks could lead to unbounded lock growth (205)
4133 - added a maximum lock count, an LRU list, and a flusher
4134 - fix multiple rename (365)
4135 - properly abstracted the echo client
4136 - OSC locked 1 byte too many; fixed
4137 - rewrote brw callback code:
4138 - fixed recovery bugs related to LOVs (306)
4139 - fixed too-many-pages-in-one-write crash (191)
4140 - fixed (again) crash in sync_io_timeout (214)
4141 - probably fixed callback-related race (385)
4143 - Add capability to MDS protocol
4144 - LDLM cancellations and callbacks on different portals
4146 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4149 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4150 - defer open object destruction to close time (601981, 138)
4151 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4152 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4153 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4154 - don't call obd_setattr in ll_file_release for destroyed objects
4155 * protocol change to lustre_msg: move |version| and add |flags|
4156 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4157 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4158 * added replay of create, unlink, link and rename operations during
4159 MDS failover; recovery should be much more robust now
4160 * remove failed OSCs from LOVs (only lov_create uses this so far)
4161 * the lustre-HOWTO was brought (more) up to date (582544)
4163 2002-10-23 Phil Schwan <phil@clusterfs.com>
4166 - in-use dentries weren't being reused properly (617851)
4167 - prevent multiple LDLM setup (599178)
4168 - fix LOV size calculations for truncate (617853)
4169 - fix client handling of MDS intent errors (POSIX)
4170 - fix permission bug in lovstripe.c test (624321)
4171 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4172 - truncate past end of file could corrupt data
4173 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4174 - a race in recovery could return ETIMEDOUT to apps (623947)
4175 - building outside the source directory was fixed
4176 * the lustre-HOWTO was brought (more) up to date (582544)
4177 * major progress was made on recovery functionality
4179 2002-10-10 Phil Schwan <phil@clusterfs.com>
4182 - recovery deadlock fix
4183 - rm -rf causes LBUG fix (617817)
4184 - file open by multiple tasks fix (618962)
4185 - directory permissions bugs (602707 and 620007)
4186 - journal_stop fixed with locking (611313)
4187 - O_APPEND failures resolved (618273, perhaps 614459)
4188 - lconf PATH fix (619770)
4189 - IA64 build fix (621450)
4190 - RPC buffer sizes scale with amount of memory
4192 2002-10-01 Phil Schwan <phil@clusterfs.com>
4195 - locks would be cancelled without throwing away data pages,
4196 resulting in inconsistent data (605627)
4197 - inode attributes were not always being refreshed (605627, 612449)
4198 - lconf now continues to cleanup after lctl reports an error
4199 - MDS now enforces user permissions (602707)
4200 - lprocfs cleanup fixed, but not yet enabled (614157)
4201 - fixed infinite server hang, should a client not respond to an AST
4202 - avoid going into recovery if user calls readlink() with a buffer
4203 that's too small (613941)
4204 - AST RPCs no longer require replies (614867) -- this may be changed
4205 - don't crash server if client sends an IOV that's too big (611336)
4206 - fixed lock conversion deadlock (611892)
4207 - fixed the following of symlinks (614622)
4208 * recovery: the server can remove locks from a client that dies, other
4209 clients can make progress
4210 * more extN patch fixes
4211 * compile-time configurable ptlrpc buffer allocations
4213 - collaborative read cache document
4214 - Lustre Lite Performance CDR document-in-progress
4216 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4219 - fix typo in patch-2.4.18
4221 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4224 - clear ptlrpc request each time in handle_incoming_request()
4225 - unlink of files now destroys the object on the OST
4227 2002-09-19 Peter Braam <braam@clusterfs.com>
4229 * add hard link support
4230 * change obdfile creation method
4231 * kernel patch changed
4233 2002-09-19 Peter Braam <braam@clusterfs.com>
4236 - stack overflow bug in extN fixed
4238 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4240 * documentation updates
4241 - add man pages for config tools
4242 - update tests/README to describe testing with new config tools
4243 - finish metadata API descriptions
4244 * bug fixes and cleanups
4245 - statfs workaround for 16TB limit
4246 - LOV stripe allocation improved, can stripe on subset of OSTs
4247 - LOV file size/IO offset was wrong for files > 4GB in size
4248 - object EA data was being dropped, caused files to be unreadable
4249 - memory overflow with non-LOV OST caused memory corruption
4250 - fixed regression tests to work with new config tools, obdfilter
4251 - fixed bug when directory size became larger than 1 block
4252 - fixed bug (for single client case) when PWD was deleted
4253 - invalidate local directory pages when doing intent-based ops
4254 - avoid LDLM oops when lock callback contained bad data
4256 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4258 * documentation updates
4259 * bug fixes and cleanups
4260 - configuration tools
4263 - 64-bit compile warnings
4264 - 64-bit internal statfs data
4266 * test_brw on persistent OST devices
4268 * lprocfs (disabled)
4270 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4272 * documentation updates
4273 * bug fixes and cleanups
4274 * configuration tools
4276 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4279 * many small fixes to 0_5_4
4280 * io/network handling
4281 * thinkos in MDS operations
4283 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4286 * crucial basic fixes to 0.5.3
4287 * IOR, Iozone work over Elan
4290 2002-08-07 Phil Schwan <phil@clusterfs.com>
4291 * version 0_5_3, our first alpha
4292 * we use the new Portals iovs
4293 * documentation updates
4294 * bug fixes and cleanups
4295 * small changes in the DLM wire protocol
4297 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4298 * version 0_5_1 with some initial stability,
4299 * locking on MD and file I/O.
4300 * documentation updates
4301 * several bug fixes since 0.5.0
4302 * small changes in wire protocol
4304 2002-07-18 Phil Schwan <phil@clusterfs.com>
4306 * delivered as Lustre Light Alpha
4307 * fixed a crash after handling invalid MDS requests
4308 * fixed directory pages for architectures with non-4k pages sizes
4310 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4311 * release version v0_4_4
4312 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4314 * Fixes a number of interruption problems with OST operations.
4315 * Update documentation for portals header changes
4316 * Move all wire protocol structs/defines to lustre_idl.h
4317 * Fixes symlink length bug.
4318 * Add tcpdump to repository.
4320 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4321 * release version v0_4_3
4322 * Fixes statfs for inodes on extN.
4323 * Fixes bug in runtests which would delete /etc/hosts.
4324 * Use 64-bit object IDs wherever possible (not into VFS though)
4325 Remove ost_get_info, which is unused by lustre, and out of date.
4327 2002-07-03 Peter Braam <braam@clusterfs.com>
4328 * release version v0_4_2 Fixes a lookup error (type not passed)
4329 * move forward to head of Portals
4330 * move forward to latest Lustre kernel
4332 2002-06-25 Peter Braam <braam@clusterfs.com>
4333 * release version v0_4_1. Hopefully stable on single node use.