3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.53-0.16 (SLES 10),
6 2.6.18-8.1.14.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
21 Description: Update to SLES9 kernel-2.6.5-7.287.3.
25 Description: Update to RHEL4 latest kernel.
29 Description: Remove 2.4 kernel support from lustre 1.8.0
30 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
31 quota_interface.c in quota subsystem.
32 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
34 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
35 lvfs_linux.c upcall_cache.c in lvfs subsystem.
36 Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
37 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
39 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
40 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
42 Remove 2.4 kernel check from lproc_osc.c, osc_request.c in osc
44 Remove 2.4 kernel check from filter.c, lproc_obdfilter.c in
46 Remove 2.4 kernel code from lustre/obdecho/echo_client.c.
47 Remove 2.4 kernel check from lustre/mgs/lproc_mgs.c,
49 Remove 2.4 kernel code from lustre/mgc/mgc_request.c.
50 Remove 2.4 kernel check from lustre/mdt/mdt_lproc.c.
51 Remove 2.4 kernel code from lustre/mdc/mdc_request.c.
52 Remove 2.4 kernel check from lustre/lov/lproc_lov.c.
53 Remove 2.4 kernel check from lustre/lmv/lproc_lmv.c.
54 Remove lock_24kernel from lustre/lvfs/fsfilt_ext3.c.
55 Remove 2.4 kernel code from include/liblustre.h,
56 include/lustre_disk.h, include/lustre_quota.h,
57 include/linux/lprocfs_status.h, include/linux/lvfs.h,
58 include/linux/lustre_lite.h, include/linux/lustre_net.h,
59 include/linux/lustre_user.h, include/linux/lvfs_linux.h,
60 include/linux/obd_class.h, include/linux/lustre_compat25.h.
62 Severity : enhancement
64 Description: Build SLES10 patchless client fails
65 Details : The configure was broken by run ./configure with
66 --with-linux-obj=.... argument for patchless client. When the
67 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
68 can't search header adequately. Use absolute path such as
69 -I($LINUX)/include instead.
73 Description: interrupt oig_wait produce painc on resend.
74 Details : brw_redo_request can be used for resend requests from ptlrpcd and
75 private set, and this produce situation when rq_ptlrpcd_data not
76 copyed to new allocated request and triggered LBUG on assert
77 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
78 ptlrpcd set if request is changed and can be safety changed to use
83 Description: LASSERT_{REQ,REP}SWAB macros are buggy
84 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
85 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
86 these macros and replace them with their difinitions instead.
90 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
91 Details : Update which_patch & target file for SLES10 latest kernel.
94 Description: Remove client patches from SLES 10 kernel.
95 Details : This causes SLES 10 clients to behave as patchless clients
96 even on a Lustre-patched (server) kernel.
98 Severity : enhancement
100 Description: self-adjustable client's lru lists
101 Details : use adaptive algorithm for managing client cached locks lru
102 lists according to current server load, other client's work
103 pattern, memory activities, etc. Both, server and client
104 side namespaces provide number of proc tunables for controlling
107 Severity : enhancement
109 Description: light-weight GSS support
110 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
111 and reduce performance overhead.
113 Severity : enhancement
115 Description: Linux keyring support
116 Details : Support using service of Linux keyring for Lustre GSS internal
117 context refresh/cache mechanism.
121 Description: Fix errors in lfs documentation
122 Details : Fixes man pages
126 Description: don't use GFP_* in generic Lustre code.
127 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
131 Description: obdclass is limited by single OBD_ALLOC(idarray)
132 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
136 Description: warnings with build patchless client with vanila 2.6.19 and up
137 Details : change old ctl_table style and replace ctl_table/ctl_table_header
138 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
142 Frequency : Only for SLES
143 Description: sanity_quota fail test_1
144 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
145 file, which makes TSTID[2] inunique.
149 Frequency : Only for SLES9
150 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
151 Details : Update target/ChangeLog/which_patch .
155 Frequency : Only for test_10 in sanity-quota.sh
156 Description: a bug in quota test code
157 Details : checking a test flag in a wrong place
161 Frequency : Only for test_13 in sanity-quota.sh
162 Description: a bug in quota test script
163 Details : shouldn't check overfull allocated quota
167 Description: jbd statistics
168 Details : Port older jbd statistics patch for sles10
170 Severity : enhancement
172 Description: Add printing inode info into message about error in writepage.
177 Description: reply_lock_interpret crash due to race with it and lock cancel.
178 Details : Do not replay locks that are being cancelled. Do not reference
179 locks by their address during replay, just by their handle.
183 Frequency : When flocks are used.
184 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
185 Details : Flock locks might destroy just granted lock if it could be merged
186 with another existing flock, this is done in completion handler,
187 so teach ldlm_cli_enquque_fini that this is a valid case for
193 Description: Oops in read and write path when failing to allocate lock.
194 Details : Check if lock allocation failed and return error back.
198 Description: lstripe command fails for valid OST index
199 Details : The stripe offset is compared to lov->desc.ld_tgt_count
200 instead of lov->desc.ld_active_tgt_count.
204 Description: sanity.sh failed test 103
205 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
210 Description: df doesn't work properly if diskfs blocksize != 4K
211 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
215 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
216 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
220 Description: obd_health_check_timeout too short
221 Details : set obd_health_check_timeout as 1.5x of obd_timeout
225 Description: llapi_file_create() does not allow some changes
226 Details : add llapi_file_open() that allows specifying the mode and
227 open flags, and also returns an open file handle.
231 Description: merge and cleanup kernel patches.
232 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
236 Description: Add journal checksum support.(Kernel part)
237 Details : The journal checksum feature adds two new flags i.e
238 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
239 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
240 indicates that the commit block contains the checksum for
241 the blocks described by the descriptor blocks. Now commit
242 record can be sent to disk without waiting for descriptor
243 blocks to be written to disk. This behavior is controlled
244 using JBD2_FEATURE_ASYNC_COMMIT flag.
248 Description: OSS needs mutliple precreate threads
249 Details : Add ability to start more than one create thread per OSS.
253 Description: Sanity.sh test_65k failed: setstripe should have succeeded
254 Details : execute lfs setstripe on client
258 Description: mds_obd_create error creating tmp object
259 Details : When the user sets quota on root, llog will be affected and can't
260 create files and write files.
263 Frequency : Always on ia64 patchless client, and possibly others.
265 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
266 Details : This allows the patchless client to be loaded on architectures
271 Description: RedHat Update kernel for RHEL5
272 Details : Add the kernel config file for RHEL5.
276 Description: RedHat Update kernel for RHEL5
277 Details : Modify the kernel config file more closer RHEL5.
281 Description: Build failure against Centos5 (RHEL5)
282 Details : Define PAGE_SIZE when it isn't present.
286 Description: test_7 on sanity-quota.sh can't be used on separate machines
287 Details : add facet to handle it
291 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
292 Details : with NFS, the anon dentry's parent was set to itself in
293 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
294 corresponding dentry other than use rec->ur_name.
296 Severity : enhancement
298 Description: lfs setstripe enhancement
299 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
303 Description: testing performance impact of enabling checksumming
304 Details : enable checksum by default, allow --disable-checksum
305 configure option and "-o nochecksum" mount option
309 Description: lustre support for RHEL5
310 Details : Add support for RHEL5.
314 Description: Client eviction due to failover config
315 Details : after a connection loss, the lustre client should attempt to
316 reconnect to the last active server first before trying the
317 other potential connections.
321 Description: when mds and osts use different quota unit(32bit and 64bit),
322 quota will be released repeatly.
323 Details : void sending multiple quota reqs to mds, which will keep the
324 status between the reqs.
328 Description: osts not allocated evenly to files
329 Details : change the condition to increase offset_idx
332 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
334 Description: Data corruption for OSTs that are formatted larger than 2TB
336 Details : When generating the bio request for lustre file writes the
337 sector number would overflow a temporary variable before being
338 used for the IO. The data reads correctly from Lustre (which
339 will overflow in a similar manner) but other file data or
340 filesystem metadata may be corrupted in some cases.
344 Description: Tune the kernel for good SCSI performance.
345 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
346 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
351 Description: rewrite ext2-derived code in obdclass/uuid.c
352 Details : rewrite inherited code (uuid parsing code from ext2 utils)
353 from scratch preserving functionality.
357 Description: "lfs find -obd UUID" prints directories
358 Details : "lfs find -obd UUID" will return all directory names instead
359 of just file names. It is incorrect because the directories
360 do not reside on the OSTs.
364 Description: use i_size_read and i_size_write in 2.6 port
365 Details : replace inode->i_size access with i_size_read/write()
368 Frequency : when removing large files
370 Description: scheduling issue during removal of large Lustre files
371 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
372 It causes scheduling issues when removing large files (17TB in the
376 Frequency : only with liblustre clients on XT3
378 Description: evictions taking too long
379 Details : allow llrd to evict clients directly on OSTs
382 Frequency : only on ppc
384 Description: /proc/fs/lustre/devices broken on ppc
385 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
386 We should make sure the type of 'index' is loff_t and then cast
387 explicitly as needed below. Do not assign an explicitly cast
391 Frequency : only for rhel5
393 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
394 Details : Modify the target file & which_kernel.
396 Severity : enhancement
398 Description: omit set fsid for export NFS
399 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
400 as FS_REQUIRES_DEV for avoid problems with generate fsid.
404 Frequency : Always, for kernels after 2.6.16
405 Description: Fix warning idr_remove called for id=.. which is not
407 Details : Last kernels save old s_dev before kill super and not allow
408 to restore from callback - restore it before call kill_anon_super.
412 Description: buffer overruns could theoretically occur
413 Details : llapi_semantic_traverse() modifies the "path" argument by
414 appending values to the end of the origin string, and a
415 overrun may occur. Adding buffer overrun check in liblustreapi.
419 Description: Fix error on 'ls .' at the top of the Lustre mount.
420 Details : Don't revalidate dentry if it is a root dentry.
424 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
425 Details : Modify vm-tunables-rhel4.patch.
429 Description: Kernel config for 2.6.18-vanilla.
430 Details : Modify targets/2.6-vanilla.target.in.
431 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
432 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
433 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
434 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
438 Description: improve handling recoverable errors
439 Details : if request processig with error which can be recoverable on server
440 request should be resend, otherwise page released from cache and
445 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
446 Details : Modify target file & which_patch.
447 A flaw was found in the IA32 system call emulation provided
448 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
449 value could be stored in the %RAX register, which could trigger an
450 out-of-bounds system call table access. An untrusted local user
451 could exploit this flaw to run code in the kernel
452 (ie a root privilege escalation). (CVE-2007-4573).
456 Description: O_DIRECT bypasses client statistics.
457 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
458 read_ahead_stats were updated. copied the stats section from
459 osc_send_oap_rpc() into async_internal().
463 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
467 Description: change order of libsysio includes
468 Details : '#include sysio.h' should always come before '#include xtio.h'
472 Description: handle "serious error: objid * already exists" more gracefully
473 Details : If LAST_ID value on disk is smaller than the objects existing in
474 the O/0/d* directories, it indicates disk corruption and causes an
475 LBUG(). If the object is 0-length, then we should use the existing
476 object. This will help to avoid a full fsck in most cases.
478 Severity : enhancement
480 Description: adapt the lustre_config script to support the upgrade case
481 Details : Add "-u" option for lustre_config script to support upgrading 1.4
482 server targets to 1.6 in parallel.
486 Description: To avoid grant space > avaible space when the disk is almost
487 full. Without this patch you might see the error "grant XXXX >
488 available" or some LBUG about grant, when the disk is almost
490 Details : In filter_check_grant, for non_grant cache write, we should
491 check the left space by if (*left > ungranted + bytes), instead
492 of (*left > ungranted), because only we are sure the left space
493 is enough for another "bytes", then the ungrant space should be
494 increase. In client, we should update cl_avail_grant only there
495 is OBD_MD_FLGRANT in the reply.
499 Description: Update RHEL 4 kernel to fix local root privilege escalation.
500 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
501 described in CVE-2007-4573. This problem could allow untrusted
502 local users to gain root access.
506 Frequency : when using O_DIRECT and quotas
508 Description: Incorrect file ownership on O_DIRECT output files
509 Details : block usage reported by 'lfs quota' does not take into account
510 files that have been written with O_DIRECT.
512 --------------------------------------------------------------------------------
514 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
516 * Support for kernels:
517 2.6.5-7.283 (SLES 9),
518 2.6.9-55.EL (RHEL 4),
519 2.6.16.46-0.14 (SLES 10),
520 2.6.18.8 vanilla (kernel.org)
521 * Client support for unpatched kernels:
522 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
523 2.6.16 - 2.6.22 vanilla (kernel.org)
524 * Due to recently discovered recovery problems, we do not recommend
525 using patchless RHEL 4 clients with this or any earlier release.
526 * Recommended e2fsprogs version: 1.39.cfs8
527 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
528 * Starting with this release, the ldiskfs backing filesystem required
529 by Lustre is now in its own package, lustre-ldiskfs. This package
530 should be installed. It is versioned separately from Lustre and
531 may be released separately in future.
535 Description: block reactivating mgc import until all deactivates complete
536 Details : Fix race when failing back MDT/MGS to itself (testing)
539 Frequency : at statup only
541 Description: mds_lov_synchronize race leads to various problems
542 Details : simultaneous MDT->OST connections at startup can cause the
543 sync to abort, leaving the OSC in a bad state.
545 Severity : enhancement
547 Description: add optional extra BUILD_VERSION info
548 Details : add a new environment variable (namely LUSTRE_VERS) which allows
549 to override the lustre version.
552 Frequency : 2.6.18 servers only
554 Description: ll_kern_mount() doesn't release the module reference
555 Details : The ldiskfs module reference count never drops down to 0
556 because ll_kern_mount() doesn't release the module reference.
561 Description: server LBUG when using old ost_num_threads parameter
562 Details : Accept the old ost_num_threads parameter but warn that it
563 is deprecated, and fix an off-by-one error that caused an LBUG.
568 Description: Transient SCSI error results in persistent IO issue
569 Details : iobuf->dr_error is not reinitialized to 0 between two
573 Frequency : sometimes when underlying device returns I/O errors
575 Description: OSTs not going read-only during write failures
576 Details : OSTs are not remounted read-only when the journal commit threads
577 get I/O errors because fsfilt_ext3 calls journal_start/stop()
578 instead of the ext3 wrappers.
582 Description: poor connect scaling with increasing client count
583 Details : Don't run filter_grant_sanity_check for more than 100 exports
584 to improve scaling for large numbers of clients.
587 Frequency : SLES10 only
589 Description: sanity-quota.sh quotacheck failed: rc = -22
590 Details : Quotas cannot be enabled on SLES10.
593 Frequency : liblustre clients only
595 Description: getdirentries does not give error when run on compute nodes
596 Details : getdirentries does not fail when the size specified as an argument
597 is too small to contain at least one entry
599 Severity : enhancement
601 Description: Add LNET router traceability for debug purposes
602 Details : If a checksum failure occurs with a router as part of the
603 IO path, the NID of the last router that forwarded the bulk data
604 is printed so it can be identified.
609 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
610 Details : Due to a race condition, liblustre clients were occasionally
613 Severity : enhancement
615 Description: lfs setstripe use optional parameters instead of postional
618 Severity : enhancement
620 Description: Nanosecond timestamp support for ldiskfs
621 Details : The on-disk ldiskfs filesystem has added support for nanosecond
622 resolution timestamps. There is not yet support for this at
623 the Lustre filesystem level.
626 Frequency : during server recovery
628 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
629 Details : request with rq_no_resend flag not awake l_wait_event if they get a
633 Frequency : nfs export on patchless client
635 Description: connectathon hang when test nfs export over patchless client
636 Details : Disconnected dentry cannot be found with lookup, so we do not need
637 to unhash it or make it invalid
640 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
641 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
642 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
647 Description: open req refcounting wrong on reconnect
648 Details : If reconnect happened between getting open reply from server and
649 call to mdc_set_replay_data in ll_file_open, we will schedule
650 replay for unreferenced request that we are about to free.
651 Subsequent close will crash in variety of ways.
652 Check that request is still eligible for replay in
653 mdc_set_replay_data().
658 Description: disable writes to filesystem when reading health_check file
659 Details : the default for reading the health_check proc file has changed
660 to NOT do a journal transaction and write to disk, because this
661 can cause reads of the /proc file to hang and block HA state
662 checking on a healthy but otherwise heavily loaded system. It
663 is possible to return to the previous behaviour during configure
664 with --enable-health-write.
666 Severity : enhancement
668 Description: 64-bit inode version
669 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
670 to the inode. This will be required for version-based recovery.
675 Description: MDS fails to start if a duplicate client export is detected
676 Details : in some rare cases it was possible for a client to connect to
677 an MDS multiple times. Upon recovery the MDS would detect this
678 and fail during startup. Handle this more gracefully.
680 Severity : enhancement
682 Description: Add -o localflock option to simulate old noflock
684 Details : This will achieve local-only flock/fcntl locks
690 Description: log_commit_thread vs filter_destroy race leads to crash
691 Details : Take import reference before releasing llog record semaphore
696 Description: Wrong request locking in request set processing
697 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
698 imp_delayed_list, in this place should be used imp_lock.
701 Frequency : when reconnection
703 Description: Grant Leak when osc reconnect to OST
704 Details : When osc reconnect ost, OST(filter) should check whether it
705 should grant more space to client by comparing fed_grant and
706 cl_avail_grant, and return the granted space to client instead
707 of "new granted" space, because client will call osc_init_grant
708 to update the client grant space info.
711 Frequency : when client reconnect to OST
713 Description: Grant Leak when osc do resend and replay bulk write
714 Details : When osc reconnect to OST, OST(filter)should clear grant info of
715 bulk write request, because the grant info while be sync between
716 OSC and OST when reconnect, and we should ignore the grant info
717 these of resend/replay write req.
722 Description: Grant space more than avaiable left space sometimes.
723 Details : When then OST is about to be full, if two bulk writing from
724 different clients came to OST. Accord the avaliable space of the
725 OST, the first req should be permitted, and the second one
726 should be denied by ENOSPC. But if the seconde arrived before
727 the first one is commited. The OST might wrongly permit second
728 writing, which will cause grant space > avaiable space.
731 Frequency : when client is evicted
733 Description: Grant might be wrongly erased when osc is evicted by OST
734 Details : when the import is evicted by server, it will fork another
735 thread ptlrpc_invalidate_import_thread to invalidate the
736 import, where the grant will be set to 0. While the original
737 thread will update the grant it got when connecting. So if
738 the former happened latter, the grant will be wrongly errased
739 because of this race.
744 Description: Checking Stale with correct fid
745 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
746 to check whether it is stale, because sometimes, we want the
747 enqueue happened anyway, and op_data.fid2 will not be initialized.
749 Severity : enhancement
751 Description: update patchless client
752 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
755 Frequency : only with 2.4 kernel
757 Description: random memory corruption
758 Details : size of struct ll_inode_info is to big for union inode.u and this
759 can be cause of random memory corruption.
764 Description: Memory leak in recovery
765 Details : Lov_mds_md was not free in an error handler in mds_create_object.
766 It should also check obd_fail before fsfilt_start, otherwise if
767 fsfilt_start return -EROFS,(failover mds during mds recovery).
768 then the req will return with repmsg->transno = 0 and rc = EROFS.
769 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
770 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
771 be freed no matter whether fsfilt_commit success or not.
774 Frequency : only with huge count clients
776 Description: Prevents from taking the superblock lock in llap_from_page for
778 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
779 for llap_from_page call in ll_removepage prevents from taking the
780 superblock lock for a soon died page.
785 Description: Not check open intent error before release open handle
786 Details : in some rare cases, the open intent error is not checked before
787 release open handle, which may cause
788 ASSERTION(open_req->rq_transno != 0), because it tries to release
789 the failed open handle.
794 Description: Set cat log bitmap only after create log success.
795 Details : in some rare cases, the cat log bitmap is set too early. and it
796 should be set only after create log success.
800 Description: Accessing a block bevice can re-enable I/O when Lustre is
801 tearing down a device.
802 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
807 Description: service threads may hog cpus when there are a lot of requests
809 Details : Insert cond_resched to give other threads a chance to use some of
815 Description: the cat log was not initialized in recovery
816 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
817 unlink log on mds will not be initialized until mds post
818 recovery. And also in mds post recovery, the unlink log will
819 initialization will be done asynchronausly, so there will be race
820 between add unlink log and unlink log initialization.
824 Description: brw_stats were being printed incorrectly
825 Details : brw_stats were being printed as log2 but all of them were not
826 recorded as log2. Also remove some code duplication arising from
827 filter_tally_{read,write}.
831 Frequency : rare, only in recovery.
832 Description: ASSERTION(req->rq_type != LI_POISON) failed
833 Details : imp_lock should be held while iterating over imp_sending_list for
834 prevent destroy request after get timeout in ptlrpc_queue_wait.
838 Description: replay-single.sh test 52 fails
839 Details : A lock's skiplist need to be cleanup when it being unlinked
840 from its resource list.
844 Description: Short directio read returns full requested size rather than
846 Details : Direct I/O operations should return actual amount of bytes
847 transferred rather than requested size.
849 Severity : enhancement
851 Description: metadata RPC reduction (e.g. for rm performance)
852 Details : decrease the amount of synchronous RPC between clients and servers
853 by canceling conflicing lock before the operation on the client side
854 and packing thier handles into the main operation RPC to server.
856 Severity : enhancement
858 Description: Async OSC create to avoid the blocking unnecessarily.
859 Details : If a OST has no remain object, system will block on the creating
860 when need to create a new object on this OST. Now, ways use
861 pre-created objects when available, instead of blocking on an
862 empty osc while others are not empty. If we must block, we block
863 for the shortest possible period of time.
865 --------------------------------------------------------------------------------
867 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
872 Frequency : on some architectures
874 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
875 Details : Uninitialized flags sometimes cause configuration commands to
879 Frequency : patchless clients only
881 Description: missing __iget() symbol export
882 Details : The __iget() symbol export is missing. To avoid the need for
883 this on patchless clients the deathrow inode reaper is turned
884 off, and we depend on the VM to clean up old inodes. This
885 dependency was during via the fix for bug 12181.
887 --------------------------------------------------------------------------------
889 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
891 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
892 INTEROPERATE with older versions automatically. In many cases a
893 special upgrade step is needed. Please read the
894 user documentation before upgrading any part of a live system.
895 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
896 version will not interoperate with 1.6 betas before beta5 (1.5.95).
897 * WARNING: Lustre configuration and startup changes are required with
898 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
904 Severity : enhancement
906 Description: MountConf
907 Details : Lustre configuration is now managed via mkfs and mount
908 commands instead of lmc and lconf. New obd types (MGS, MGC)
909 are added for dynamic configuration management. See
910 https://mail.clusterfs.com/wikis/lustre/MountConf for
913 Severity : enhancement
915 Description: dynamic OST addition
916 Details : OSTs can now be added to a live filesystem
918 Severity : enhancement
920 Description: startup order invariance
921 Details : MDTs and OSTs can be started in any order. Clients only
922 require the MDT to complete startup.
924 Severity : enhancement
926 Description: parallel, asynchronous orphan cleanup
927 Details : orphan cleanup is now performed in separate threads for each
928 OST, allowing parallel non-blocking operation.
930 Severity : enhancement
932 Description: optimized stripe assignment
933 Details : stripe assignments are now made based on ost space available,
934 ost previous usage, and OSS previous usage, in order to try
935 to optimize storage space and networking resources.
937 Severity : enhancement
939 Description: Permanently set tunables
940 Details : All writable /proc/fs/lustre tunables can now be permanently
941 set on a per-server basis, at mkfs time or on a live
944 Severity : enhancement
946 Description: Lustre message v2
947 Details : Add lustre message format v2.
949 Severity : enhancement
951 Description: client OST exclusion list
952 Details : Clients can be started with a list of OSTs that should be
953 declared "inactive" for known non-responsive OSTs.
957 Description: SPEC SFS validation failure on NFS v2 over lustre.
958 Details : Changes the blocksize for regular files to be 2x RPC size,
959 and not depend on stripe size.
961 Severity : enhancement
963 Description: Multiple MD RPCs in flight.
964 Details : Further unserialise some read-only MDS RPCs - learn about intents.
965 To avoid overly-overloading MDS, introduce a limit on number of
966 MDS RPCs in flight for a single client and add /proc controls
967 to adjust this limit.
969 Severity : enhancement
971 Description: client read/write statistics
972 Details : Add client read/write call usage stats for performance
973 analysis of user processes.
974 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
975 file access. extents_stats shows chunk size distribution.
976 extents_stats_per_process show chunk size distribution per
979 Severity : enhancement
981 Description: mds statistics
982 Details : Add detailed mds operations statistics in
983 /proc/fs/lustre/mds/*/stats.
987 Description: Failure of copying files with lustre special EAs.
988 Details : Client side always return success for setxattr call for lustre
989 special xattr (currently only "trusted.lov").
994 Description: Refcount LNET uuids
995 Details : The global LNET uuid list grew linearly with every startup;
996 refcount repeated list entries instead of always adding to
999 Severity : enhancement
1001 Description: Dynamic service threads
1002 Details : Within a small range, start extra service threads
1003 automatically when the request queue builds up.
1006 Frequency : mixed-endian client/server environments
1008 Description: mixed-endian crashes
1009 Details : The new msg_v2 system had some failures in mixed-endian
1012 Severity : enhancement
1014 Description: Easy OST removal
1015 Details : OSTs can be permanently deactivated with e.g. 'lctl
1016 conf_param lustre-OST0001.osc.active=0'
1018 Severity : enhancement
1020 Description: MGS proc entries
1021 Details : Added basic proc entries for the MGS showing what filesystems
1024 Severity : enhancement
1026 Description: provide MGS failover
1027 Details : Added config lock reacquisition after MGS server failover.
1029 Severity : enhancement
1031 Description: add Linux 2.4 support
1032 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1036 Description: a large application tries to do I/O to the same resource and dies
1037 in the middle of it.
1038 Details : Check the req->rq_arrival time after the call to
1039 ost_brw_lock_get(), but before we do anything about
1040 processing it & sending the BULK transfer request. This
1041 should help move old stale pending locks off the queue as
1042 quickly as obd_timeout.
1045 Frequency : when an incorrect nid is specified during startup
1047 Description: ptlrpc connect to non-existant node causes kernel crash
1048 Details : LNET can't be re-entered from an event callback, which
1049 happened when we expire a message after the export has been
1050 cleaned up. Instead, hand the zombie cleanup off to another
1053 Severity : enhancement
1055 Description: plain/inodebits lock performance improvement
1056 Details : Grouping plain/inodebits in granted list by their request modes
1057 and bits policy, thus improving the performance of search through
1061 Frequency : only if OST filesystem is corrupted
1063 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1064 Details : for a short time RPCs with bulk IO are in the replay list,
1065 but replay of bulk IOs is unimplemented. If the OST filesystem
1066 is corrupted due to disk cache incoherency and then replay is
1067 started it is possible to trip an assertion. Avoid putting
1068 committed RPCs into the replay list at all to avoid this issue.
1071 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1073 Description: System hang on startup
1074 Details : This bug allowed the liblustre (e.g. catamount) client to
1075 return to the app before handling all startup RPCs. This
1076 could leave the node unresponsive to lustre network traffic
1077 and manifested as a server ptllnd timeout.
1079 Severity : enhancement
1081 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1082 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1083 boolean to control whether to print peer debug info when a
1084 client's RPC times out.
1087 Frequency : only for kernels with patches from Lustre below 1.4.3
1089 Description: Remove old rdonly API
1090 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1093 Frequency : only for devices with external journals
1095 Description: Set external device read-only also
1096 Details : During a commanded failover stop, we set the disk device
1097 read-only while the server shuts down. We now also set any
1098 external journal device read-only at the same time.
1101 Frequency : when upgrading from 1.4 while trying to change parameters
1103 Description: The wrong (new) MDC name was used when setting parameters for
1104 upgraded MDT's. Also allows changing of OSC (and MDC)
1105 parameters if --writeconf is specified at tunefs upgrade time.
1108 Frequency : when setting specific ost indicies
1110 Description: QOS code breaks on skipped indicies
1111 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1112 created with --index need not be sequential.
1114 Severity : enhancement
1116 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1117 Details : The uninit_groups feature works in conjunction with the kernel
1118 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1119 up the pass1 processing of e2fsck. This is a read-only feature
1120 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1121 mount filesystems that have had this feature enabled.
1123 Severity : enhancement
1125 Description: Improve multi-block allocation algorithm to avoid fragmentation
1126 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1127 allocation locality and avoid filesystem fragmentation.
1129 ------------------------------------------------------------------------------
1131 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1133 * Support for kernels:
1134 2.4.21-47.0.1.EL (RHEL 3)
1135 2.6.5-7.283 (SLES 9)
1136 2.6.9-42.0.10.EL (RHEL 4)
1137 2.6.12.6 vanilla (kernel.org)
1138 2.6.16.27-0.9 (SLES 10)
1139 * Recommended e2fsprogs version: 1.39.cfs5
1141 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1145 Frequency : occasional, depends on client load and configuration
1146 Bugzilla : 12181, 12203
1147 Description: data loss for recently-modified files
1149 Details : In some cases it is possible that recently written or created
1150 files may not be written to disk in a timely manner (this should
1151 normally be within 30s unless client IO load is very high).
1152 The problem appears as zero-length files or files that are a
1153 multiple of 1MB in size after a client crash or client eviction
1154 that are missing data at the end of the file.
1156 This problem is more likely to be hit on clients where files are
1157 repeatedly created and unlinked in the same directory, clients
1158 have a large amount of RAM, have many CPUs, the filesystem has
1159 many OSTs, the clients are rebooted frequently, and/or the files
1160 are not accessed by other nodes after being written.
1162 The presence of the problem can be detected by looking at
1163 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1164 smaller than the second (nr_unused) then dirty files will not
1165 be flushed automatically to disk. "sync; sleep 10" should be
1166 run several times on the node before unmounting it to update
1167 Lustre (this is also safe to run on nodes without this problem).
1169 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1170 that can cause this same problem, so customers using that kernel
1171 also need to update the kernel in addition to Lustre. In order
1172 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1174 It is normal that files written just before a client crash (less
1175 than 30s) may not yet have been flushed to disk, even for local
1179 Frequency : frequent on thin XT3 nodes
1181 Description: UUID collision on thin XT3 Linux nodes
1182 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1183 randomly, since we relied on an insufficiently-seeded PRNG.
1188 Description: OSS hangs after "All ost request buffers busy"
1189 Details : A deadlock between quota and journal operations caused OSS
1190 hangs after printing "All ost request buffers busy."
1193 Frequency : always on liblustre builds
1195 Description: Cleanup compiler warnings on liblustre
1198 Frequency : always on liblustre builds on XT3
1200 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1204 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1205 > 32000 subdirectories
1206 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1207 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1208 subdirectory count crosses 32000. This will aid e2fsck to
1209 correctly handle more than 32000 subdirectories.
1212 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1214 Description: System hang on startup
1215 Details : This bug allowed the liblustre (e.g. catamount) client to
1216 return to the app before handling all startup RPCs. This
1217 could leave the node unresponsive to lustre network traffic
1218 and manifested as a server ptllnd timeout.
1220 Severity : enhancement
1222 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1223 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1224 boolean to control whether to print peer debug info when a
1225 client's RPC times out.
1230 Description: make O_SYNC working on 2.6 kernels
1231 Details : 2.6 kernels use different method for mark pages for write,
1232 so need add a code to lustre for O_SYNC work.
1237 Description: Failure to close file and release space on NFS
1238 Details : Put inode details into lock acquired in ll_intent_file_open.
1239 Use mdc_intent_lock in ll_intent_open to properly
1240 detect all kind of errors unhandled by mdc_enqueue.
1245 Description: proc file read during shutdown sometimes raced obd removal,
1247 Details : Add lock to prevent obd access after proc file removal.
1250 Frequency : Only for files larger than 4GB on 32-bit clients.
1252 Description: improperly doing page alignment of locks
1253 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1254 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1259 Description: RPCs being resent when they shouldn't be
1260 Details : Some RPCs that should not be resent are being resent. This
1261 can cause inconsistencies in the RPC state machine. Do not
1262 resend such requests.
1265 Frequency : rare, only with NFS export
1267 Description: Crash on NFS re-export node
1268 Details : under very unusual load conditions an assertion is hit in
1269 ll_intent_file_open()
1272 Frequency : only if OST filesystem is corrupted
1274 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1275 Details : for a short time RPCs with bulk IO are in the replay list,
1276 but replay of bulk IOs is unimplemented. If the OST filesystem
1277 is corrupted due to disk cache incoherency and then replay is
1278 started it is possible to trip an assertion. Avoid putting
1279 committed RPCs into the replay list at all to avoid this issue.
1284 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1285 Details : Large single O_DIRECT read and write calls can fail to allocate
1286 a sufficiently large buffer to process the request. In case of
1287 allocation failure the allocation is retried with a smaller
1288 buffer and broken into smaller requests.
1290 Severity : enhancement
1292 Description: Add -o localflock option to simulate old noflock behaviour.
1293 Details : This will achieve local-only flock/fcntl locks coherentness.
1298 Description: versioning check is incomplete
1299 Details : Checking the version difference of client vs. server, report
1300 error if the gap is too big.
1305 Description: add support PG_writeback bit
1306 Details : add support for PG_writeback bit for Lustre, for more carefull
1307 work with page cache in 2.6 kernel. This also fix some deadlocks
1308 and remove hack for work O_SYNC with 2.6 kernel.
1310 Severity : enhancement
1312 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1313 Details : The uninit_groups feature works in conjunction with the kernel
1314 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1315 up the pass1 processing of e2fsck. This is a read-only feature
1316 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1317 mount filesystems that have had this feature enabled.
1319 Severity : enhancement
1321 Description: Improve multi-block allocation algorithm to avoid fragmentation
1322 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1323 allocation locality and avoid filesystem fragmentation.
1325 ------------------------------------------------------------------------------
1327 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1329 * Support for kernels:
1330 2.6.9-42.0.3.EL (RHEL 4)
1331 2.6.5-7.276 (SLES 9)
1332 2.4.21-47.0.1.EL (RHEL 3)
1333 2.6.12.6 vanilla (kernel.org)
1334 2.6.16.21-0.8 (SLES10)
1335 * Recommended e2fsprogs version: 1.39.cfs2-0
1337 * The backwards-compatible /proc/sys/portals symlink has been removed
1338 in this release. Before upgrading, please ensure that you change
1339 any configuration scripts or /etc/sysctl.conf files that access
1340 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1341 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1342 in advance of the upgrade on any system running Lustre 1.4.6 or
1343 newer, since /proc/sys/lnet was added in that version.
1344 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1348 Frequency : only when quota is used
1350 Description: avoid scanning export list for quota master
1351 Details : Change the algorithms to avoid scanning export list in order
1352 to improve the efficiency.
1355 Frequency : MDS failover only, very rarely
1357 Description: "went back in time" messages on mds failover
1358 Details : The greatest transno may be lost when the current operation
1359 finishes with an error (transno==0) and the client's last_rcvd
1360 record is over-written. Save the greatest transno in the
1361 mds_last_transno for this case.
1364 Frequency : always for specific kernels and striping counts
1366 Description: client may get "Matching packet too big" without ACL support
1367 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1368 when trying to access files in certain configurations. The
1369 clients should in fact be denied when mounting because they do
1370 not understand ACLs.
1373 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1375 Description: many clients connecting with IO in progress causes connect timeouts
1376 Details : Avoid synchronous journal commits to avoid delays caused by many
1377 clients connecting/disconnecting when bulk IO is in progress.
1378 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1379 OST_IO_PORTAL to avoid delays behind potentially many pending
1383 Frequency : occasionally with multiple writers to a single file
1385 Description: shared writes to file may result in wrong size reported by stat()
1386 Details : Allow growing of kms when extent lock is cancelled
1389 Frequency : always with random mmap IO to multi-striped file
1391 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1392 Details : Only if the hole is at the end of OST object so that kms is too
1393 small. Fix is to increase kms accordingly in ll_nopage.
1396 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1398 Description: writes to a missing object would leak memory on the OST
1399 Details : If there is an inconsistency between the MDS and OST filesystems,
1400 such that the MDS references an object that doesn't exist, writes
1401 to that object will leak memory due to incorrect cleanup in the
1402 error handling path, eventually running out of memory on the OST.
1407 Description: Creating too long symlink causes lustre errors
1408 Details : Check symlink and name lengths before sending requests to MDS.
1411 Frequency : only if flock is enabled (not on by default)
1413 Description: posix locks not released on fd closure on 2.6.9+
1414 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1415 kernels, this caused such locks not to be released on fd close and
1416 then assertions on fs unmount about still used locks.
1419 Frequency : MDS failover only, very rarely
1421 Description: clients may get ASSERTION(granted_lock != NULL)
1422 Details : When request was taking a long time, and a client was resending
1423 a getattr by name lock request. The were multiple lock requests
1424 with the same client lock handle and
1425 mds_getattr_name->fixup_handle_for_resent_request found one of the
1426 lock handles but later failed with ASSERTION(granted_lock != NULL).
1431 Description: handle->h_buffer_credits > 0, assertion failure
1432 Details : h_buffer_credits is zero after truncate, causing assertion
1433 failure. This patch extends the transaction or creates a new
1437 Frequency : NFS re-export or patchless client
1438 Bugzilla : 11179, 10796
1439 Description: Crash on NFS re-export node (__d_move)
1440 Details : We do not want to hash the dentry if we don't have a lock.
1441 But if this dentry is later used in d_move, we'd hit uninitialised
1442 list head d_hash, so we just do this to init d_hash field but
1443 leave dentry unhashed.
1446 Frequency : NFS re-export or patchless client
1448 Description: NFS exports has problem with symbolic link
1449 Details : lustre client didn't properly install dentry when re-exported
1450 to NFS or running patchless client.
1453 Frequency : NFS re-export or patchless client
1455 Description: Various nfs/patchless fixes.
1456 Details : fixes reuse disconected alias for lookup process - this fixes
1457 warning "find_exported_dentry: npd != pd",
1458 fix permission error with open files at nfs.
1459 fix apply umask when do revalidate.
1462 Frequency : occasional
1464 Description: Crash on NFS re-export node
1465 Details : calling clear_page() on the wrong pointer triggered oops in
1466 generic_mapping_read().
1469 Frequency : rarely, using O_DIRECT IO
1471 Description: unaligned directio crashes client with LASSERT
1472 Details : check for unaligned buffers before trying any requests.
1475 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1477 Description: stale data returned from RAID cache
1478 Details : If only a small amount of IO is done to the RAID device before
1479 reading it again it is possible to get stale data from the RAID
1480 cache instead of reading it from disk.
1483 Frequency : always for sles10 kernel
1485 Description: sles10 support
1486 Details : ll_follow_link: compile fixes and using of nd_set_link
1487 under newer kernels.
1490 Frequency : depends on arch, kernel and compiler version, always on sles10
1493 Description: recursive or deep enough symlinks cause stack overflow
1494 Details : getting rid of large stack-allocated variable in
1498 Frequency : depends on hardware
1500 Description: lustre write performance loss in the SLES10 kernel
1501 Details : the performance loss is caused by using of write barriers in the
1502 ext3 code. The SLES10 kernel turns barrier support on by
1503 default. The fix is to undo that change for ldiskfs.
1505 ------------------------------------------------------------------------------
1507 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1509 * Support for kernels:
1510 2.6.9-42.0.3EL (RHEL 4)
1511 2.6.5-7.276 (SLES 9)
1512 2.4.21-47.0.1.EL (RHEL 3)
1513 2.6.12.6 vanilla (kernel.org)
1517 Frequency : quota enabled and large files being deleted
1519 Description: releasing more than 4GB of quota at once hangs OST
1520 Details : If a user deletes more than 4GB of files on a single OST it
1521 will cause the OST to spin in an infinite loop. Release
1522 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1527 Description: statfs data retrieved from /proc may be stale or zero
1528 Details : When reading per-device statfs data from /proc, in the
1529 {kbytes,files}_{total,free,avail} files, it may appear
1530 as zero or be out of date.
1533 Frequency : systems with MD RAID1 external journal devices
1535 Description: lconf's call to blkid is confused by RAID1 journal devices
1536 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1537 of returning all block devices that match the journal UUID.
1540 Frequency : always, for aggregate stripe size over 4GB
1542 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1543 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1544 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1547 Frequency : always if "lfs find" used on a local file/directory
1549 Description: "lfs find" segfaults if used on a local file/directory
1550 Details : The case where a directory component was not specified wasn't
1551 handled correctly. Handle this properly.
1554 Frequency : always on ppc64
1556 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1557 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1558 perform endianness conversion.
1561 Frequency : rarely (truncate to non-zero file size after write under load)
1562 Bugzilla : 10730, 10687
1563 Description: Files padded with zeros to next 4K multiple
1564 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1565 it is possible that files that are truncated to a non-zero size
1566 immediately after being written are filled with zero bytes beyond
1567 the truncated size. No file data is lost.
1569 Severity : enhancement
1571 Description: Allow recovery/failover for liblustre clients.
1572 Details : liblustre clients were unaware of failover configurations until
1575 Severity : enhancement
1577 Description: user file locks should fail when not mounting with flock option
1578 Details : Set up an error-returning stub in ll_file_operations.lock field
1579 to prevent incorrect behaviour when client is mounted without
1580 flock option. Also, set up properly f_op->flock field for
1584 Frequency : always on ia64
1586 Description: "lfs df" loops on printing out MDS statfs information
1587 Details : The obd_ioctl_data was not initialized and in some systems
1588 this caused a failure during the ioctl that did not return
1589 an error. Initialize the struct and return an error on failure.
1592 Frequency : SLES 9 only
1594 Description: Error of copying files with lustre special EAs as root
1595 Details : Client side always return success for setxattr call for lustre
1596 special xattr (currently only "trusted.lov").
1599 Frequency : rarely on clusters with both ia64+i386 clients
1601 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1602 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1603 clients are doing concurrent writes to the same file it is
1604 possible that the ia64 clients may LASSERT because the OST
1605 extent locks are not PAGE_SIZE aligned. Ensure that grown
1606 locks are always aligned on the request boundary.
1609 Frequency : specific use, occasional
1611 Description: Overwriting in use executable truncates on-disk binary image
1612 Details : If one node attempts to overwrite an executable in use by
1613 another node, we now correctly return ETXTBSY instead of
1614 truncating the file.
1616 Severity : enhancement
1618 Description: Async OSC create to avoid the blocking unnecessarily.
1619 Details : If a OST has no remain object, system will block on the creating
1620 when need to create a new object on this OST. Now, ways use
1621 pre-created objects when available, instead of blocking on an
1622 empty osc while others are not empty. If we must block, we block
1623 for the shortest possible period of time.
1628 Description: chmod on Lustre root is propagated to other clients
1629 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1630 invalid by the follow_mount time.
1633 Frequency : liblustre clients only
1635 Description: Race in 'instant cancel' lock handling could lead to such locks
1636 never to be granted in case of SMP MDS
1637 Details : Do not destroy not yet granted but cbpending locks in
1641 Frequency : replay/resend of open
1643 Description: non null lock assetion failure in mds_intent_policy
1644 Details : Trying to replay/resend lockless open requests resulted in
1645 mds_open() returning 0 with no lock. Now it sets a flag if
1646 a lock is going to be returned.
1648 Severity : enhancement
1650 Description: Checksum enhancements
1651 Details : New checksum enhancements allow for resending RPCs that failed
1654 Severity : enhancement
1656 Description: Tunables on number of dirty pages in cacche
1657 Details : Allow to set limit on number of dirty pages cached.
1662 Description: client crash on unmount - lock still has references
1663 Details : In some error handling cases it was possible to leak a lock
1664 reference on a client while accessing a file. This was not
1665 harmful to the client during operation, but would cause the
1666 client to crash when the filesystem is unmounted.
1669 Frequency : specific case, rare
1671 Description: ETXTBSY on mds though file not in use by client
1672 Details : ETXTBSY is no longer incorrectly returned when attempting to
1673 chmod or chown a directory that the user previously tried to
1674 execute or a currently-executing binary.
1677 Frequency : extremely rare except on liblustre-based clients
1679 Description: Lustre space not freed when files are deleted
1680 Details : Clean up open-unlinked files after client eviction. Previously
1681 the unlink was skipped and the files remained as orphans.
1686 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1687 Details : In some cases it was possible to send a blocking callback to
1688 a client doing a glimpse, even though that client didn't get
1689 a lock granted. When the glimpse lock is cancelled on the OST
1690 the freed lock is left on the waiting list and corrupted the list.
1693 Frequency : all core dumps
1695 Description: Broke core dumps to lustre
1696 Details : Negative dentry may be unhashed if parent does not have UPDATE
1697 lock, but some callers, e.g. do_coredump, expect dentry to be
1698 hashed after successful create, hash it in ll_create_it.
1700 ------------------------------------------------------------------------------
1702 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1704 * Support for kernels:
1705 2.6.9-42.0.2.EL (RHEL 4)
1706 2.6.5-7.276 (SLES 9)
1707 2.4.21-40.EL (RHEL 3)
1708 2.6.12.6 vanilla (kernel.org)
1712 Frequency : always on RHEL 3
1714 Description: Number of open files grows over time
1715 Details : The number of open files grows over time, whether or not
1716 Lustre is started. This was due to a filp leak introduced
1717 by one of our kernel patches.
1719 ------------------------------------------------------------------------------
1721 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1723 * Support for kernels:
1724 2.6.9-42.EL (RHEL 4)
1725 2.6.5-7.276 (SLES 9)
1726 2.4.21-40.EL (RHEL 3)
1727 2.6.12.6 vanilla (kernel.org)
1732 Bugzilla : 5719, 9635, 9792, 9684
1733 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1734 Details : If a server is under heavy load and cannot reply to new
1735 connection requests before the client resends the (re)connect,
1736 the connection handling code can behave badly if two service
1737 threads are concurrently handing separate (re)connections from
1738 the same client. Add better locking to the connection handling
1739 code, and ensure that only a single connection will be processed
1740 for a given client UUID, even if the lock is dropped.
1742 Severity : enhancement
1744 Description: add TCP zero-copy support to kernel
1745 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1746 sends if the hardware supports scatter-gather and checksumming.
1747 This allows socklnd to do client-write and server-read more
1748 efficiently and reduce CPU utilization from skbuf copying.
1751 Frequency : only if NFS exporting from client
1753 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1754 Details : If setting attributes on a file created under NFS that had
1755 never been opened it would be possible to oops the client
1756 if the file had no objects.
1759 Frequency : always for liblustre
1761 Description: liblustre client does MDS+OSTs setattr RPC for each write
1762 Details : When doing a write from a liblustre client, the client
1763 incorrectly issued an RPC to the MDS and each OST the file was
1764 striped over in order to update the timestamps. When writing
1765 with small chunks and many clients this could overwhelm the MDS
1766 with RPCs. In all cases it would slow down the write because
1767 these RPCs are unnecessary.
1769 Severity : enhancement
1771 Description: allow number of MDS service threads to be changed at module load
1772 Details : It is now possible to change the number of MDS service threads
1773 running. Adding "options mds mds_num_threads={N}" to the MDS's
1774 /etc/modprobe.conf will set the number of threads for the next
1775 time Lustre is restarted (assuming the "mds" module is also
1776 reloaded at that time). The default number of threads will
1777 stay the same, 32 for most systems.
1782 Description: OST crash if filesystem is unformatted or corrupt
1783 Details : If an OST is started on a device that has never been formatted
1784 or if the filesystem is corrupt and cannot even mount then the
1785 error handling cleanup routines would dereference a NULL pointer.
1790 Description: NULL pointer deref in llap_from_page.
1791 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1792 incorrect) mapping if the page was truncated/reclaimed while it was
1793 searched for. Check for this condition and skip such pages when
1794 doing readahead. Introduce extra check to llap_from_page() to
1795 verify page->mapping->host is non-NULL (so page is not anonymous).
1798 Frequency : Sometimes when using sys_sendfile
1800 Description: "page not covered by a lock" warnings from ll_readpage
1801 Details : sendfile called ll_readpage without right page locks present.
1802 Now we introduced ll_file_sendfile that does necessary locking
1803 around call to generic_file_sendfile() much like we do in
1807 Frequency : with certain MDS communication failures at client mount time
1809 Description: NULL pointer deref after failed client mount
1810 Details : a client connection request may delayed by the network layer
1811 and not be sent until after the PTLRPC layer has timed out the
1812 request. If the client fails the mount immediately it will try
1813 to clean up before the network times out the request. Add a
1814 reference from the request import to the obd device and delay
1815 the cleanup until the network drops the request.
1818 Frequency : occasionally during client (re)connect
1820 Description: assertion failure during client (re)connect
1821 Details : processing a client connection request may be delayed by the
1822 client or server longer than the client connect timeout. This
1823 causes the client to resend the connection request. If the
1824 original connection request is replied in this interval, the
1825 client may trip an assertion failure in ptlrpc_connect_interpret()
1826 which thought it would be the only running connect process.
1829 Frequency : only with obd_echo servers and clients that are rebooted
1831 Description: kernel BUG accessing uninitialized data structure
1832 Details : When running an obd_echo server it did not start the ping_evictor
1833 thread, and when a client was evicted an uninitialized data
1834 structure was accessed. Start the ping_evictor in the RPC
1835 service startup instead of the OBD startup.
1837 Severity : enhancement
1838 Bugzilla : 10193 (patchless)
1839 Description: Remove dependency on various unexported kernel interfaces.
1840 Details : No longer need reparent_to_init, exit_mm, exit_files,
1841 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1844 Frequency : rare (only users of deprecated and unsupported LDAP config)
1846 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1847 Details : LDAP apparently contains 'lustreName' attributes instead of
1848 'name'. A simple remapping of the name is sufficient.
1851 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1853 Description: waiting_locks_callback trips kernel BUG if client is evicted
1854 Details : Running with the dump_on_timeout debug flag turned on makes
1855 it possible that the waiting_locks_callback() can try to dump
1856 the Lustre kernel debug logs from an interrupt handler. Defer
1857 this log dumping to the expired_lock_main() thread.
1859 Severity : enhancement
1861 Description: Support NFS exporting on 2.6 kernels.
1862 Details : Implement non-rawops metadata methods for NFS server to use without
1863 changing NFS server code.
1866 Frequency : very rare (synthetic metadata workload only)
1868 Description: two racing renames might cause an MDS thread to deadlock
1869 Details : Running the "racer" program may cause one MDS thread to rename
1870 a file from being the source of a rename to being the target of
1871 a rename at exactly the same time that another thread is doing
1872 so, and the second thread has already enqueued these locks after
1873 doing a lookup of the target and is trying to relock them in
1874 order. Ensure that we don't try to re-lock the same resource.
1877 Frequency : only very large systems with liblustre clients
1879 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1880 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1881 all OSTs in parallel. This allows the eviction of stale liblustre
1882 clients to proceed much faster than if they were done in series,
1883 and also offers similar improvements for other set_info RPCs.
1888 Description: excessive CPU usage during initial read phase on client
1889 Details : During the initial read phase on a client, it would agressively
1890 retry readahead on the file, consuming too much CPU and impacting
1891 performance (since 1.4.5.8). Improve the readahead algorithm
1892 to avoid this, and also improve some other common cases (read
1893 of small files in particular, where "small" is files smaller than
1894 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1899 Description: MDS crash when receiving packet with unknown intent.
1900 Details : Do not LBUG in unknown intent case, just return -EFAULT
1902 Severity : enhancement
1903 Bugzilla : 9293, 9385
1904 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1905 Details : Do not serialize getattr (non-intent version) and statfs.
1908 Frequency : occasional, when OST network is overloaded/intermittent
1910 Description: client evicted by OST after bulk IO timeout
1911 Details : If a client sends a bulk IO request (read or write) the OST
1912 may evict the client if it is unresposive to its data GET/PUT
1913 request. This is incorrect if the network is overloaded (takes
1914 too long to transfer the RPC data) or dropped the OST GET/PUT
1915 request. There is no need to evict the client at all, since
1916 the pinger and/or lock callbacks will handle this, and the
1917 client can restart the bulk request.
1920 Frequency : Always when mmapping file with no objects
1922 Description: client crashes when mmapping file with no objects
1923 Details : Check that we actually have objects in a file before doing any
1924 operations on objects in ll_vm_open, ll_vm_close and
1930 Description: Request leak when working with deleted CWD
1931 Details : Introduce advanced request refcount tracking for requests
1932 referenced from lustre intent.
1934 Severity : Enhancement
1936 Description: Cache open file handles on client.
1937 Details : MDS now will return special lock along with openhandle, if
1938 requested and client is allowed to hold openhandle, even if unused,
1939 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1940 closing files for every read/write openration.
1942 Severity : Enhancement
1944 Description: Cache open negative dentries on client when possible.
1945 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1946 negative dentries on lock revocation.
1951 Description: Remounting a client read-only wasn't possible with a zconf mount
1952 Details : It wasn't possible to remount a client read-only with llmount.
1954 Severity : enhancement
1955 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1956 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1957 MPICH is included to add Lustre-specific ADIO interfaces.
1958 This is based closely on the UFS ADIO layer and only differs
1959 in file creation, in order to allow the OST striping to be set.
1960 This is user-contributed code and not supported by CFS.
1965 Description: extended inode attributes (immutable, append-only) work improperly
1966 when 2.4 and 2.6 kernels are used on client/server or vice versa
1967 Details : Introduce kernel-independent values for these flags.
1969 Severity : enhancement
1972 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1973 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1974 granularity. Now, Fractional MB granularity is be supported,
1975 this is very useful for low memory system.
1977 Severity : enhancement
1979 Description: Getattr by fid
1980 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1981 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1982 have locks and avoids getattr rpc requests for it on every stat.
1985 Frequency : Always, for filesystems larger than 2TB
1987 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1988 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1989 filesystems larger than 2TB to be mounted with mballoc enabled.
1992 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1994 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1995 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1996 filesystems larger than 2TB is mounted then the kernel will
1997 silently corrupt the start of the filesystem. CONFIG_LBD is
1998 enabled for all CFS-supported kernels, but the possibility of
1999 this happening with a modified kernel config exists.
2001 Severity : enhancement
2003 Description: add client O_DIRECT support for 2.6 kernels
2004 Details : It is now possible to do O_DIRECT reads and writes to files
2005 in the Lustre client mountpoint on 2.6 kernel clients.
2007 Severity : enhancement
2009 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2010 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2011 parallel, not waiting for response from every OST before sending
2012 a rpc to the next OST.
2017 Description: setattr vs write race when updating file timestamps
2018 Details : Client processes that update a file timestamp into the past
2019 right after writing to the file (e.g. tar) it is possible that
2020 the updated file modification time can be reset to the current
2021 time due to a race between processing the setattr and write RPC.
2023 Severity : enhancement
2025 Description: Bring 'lfs find' closer in line with regular Linux find.
2026 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2027 -print0 options and obtains all the needed info through the lustre
2030 Severity : enhancement
2032 Description: support up to 1024 configured devices on one node
2033 Details : change obd_dev array from statically allocated to dynamically
2034 allocated structs as they are first used to reduce memory usage
2039 Description: Flush dirty partially truncated pages during truncate
2040 Details : Immediatelly flush partially truncated pages in filter_setattr,
2041 this way we completely avoid having any pages in page cache on OST
2042 and can retire ugly workarounds during writes to flush such pages.
2047 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2048 Details : protect inode from truncation within vfs_unlink() context
2049 just take a reference before calling vfs_unlink() and release it
2050 when parent's i_sem is free.
2055 Description: last_id value checked outside lock on OST caused LASSERT failure
2056 Details : If there were multiple MDS->OST object precreate requests in
2057 flight, it was possible that the OST's last object id was checked
2058 outside a lock and incorrectly tripped an assertion. Move checks
2059 inside locks, and discard old precreate requests.
2062 Frequency : always, if extents are used on OSTs
2064 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2065 Details : OSTs using the extents format would not zero the high 16 bits of
2066 the index physical block number. This is not a problem for any
2067 OST filesystems smaller than 16TB, and no kernels support ext3
2068 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2069 new/modified files) and can be fixed for existing filesystems
2070 with e2fsprogs-1.39-cfs1.
2075 Description: import connection selection may be incorrect if timer wraps
2076 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2077 import connections to be ignored if the 32-bit jiffies counter
2078 wraps. Use a 64-bit jiffies counter.
2081 Frequency : very large clusters immediately after boot
2083 Description: LNET request buffers exhausted under heavy short-term load
2084 Details : If a large number of client requests are generated on a service
2085 that has previously never seen so many requests it is possible
2086 that the request buffer growth cannot keep up with the spike in
2087 demand. Instead of dropping incoming requests, they are held in
2088 the LND until the RPC service can accept more requests.
2091 Frequency : Sometimes during replay
2093 Description: Assertion failure in ll_local_open after replay.
2094 Details : If replay happened on an open request reply before we were able
2095 to set replay handler, reply will become not swabbed tripping the
2096 assertion in ll_local_open. Now we set the handler right after
2097 recognising of open request
2100 Frequency : very rare
2102 Description: kernel reports "badness in vsnprintf"
2103 Details : Reading from the "recovery_status" /proc file in small chunks
2104 may cause a negative length in lprocfs_obd_rd_recovery_status()
2105 call to vsnprintf() (which is otherwise harmless). Exit early
2106 if there is no more space in the output buffer.
2108 Severity : enhancement
2110 Description: clear OBD RPC statistics by writing to them
2111 Details : It is now possible to clear the OBD RPC statistics by writing
2112 to the "stats" file.
2117 Description: Client mtime is not the same on different clients after utimes
2118 Details : In some cases, the client was using the utimes() syscall on
2119 a file cached on another node. The clients now validate the
2120 ctime from the MDS + OSTs to determine which one is right.
2125 Description: Inability to activate failout mode
2126 Details : lconf script incorrectly assumed that in pythong string's numeric
2127 value is used in comparisons.
2130 Frequency : always with multiple stripes per file
2132 Description: Inefficient object allocation for mutli-stripe files
2133 Details : When selecting which OSTs to stripe files over, for files with
2134 a stripe count that divides evenly into the number of OSTs,
2135 the MDS is always picking the same starting OST for each file.
2136 Return the OST selection heuristic to the original design.
2141 Description: mount failures may take full timeout to return an error
2142 Details : Under some heavy load conditions it is possible that a
2143 failed mount can wait for the full obd_timeout interval,
2144 possibly several minutes, before reporting an error.
2145 Instead return an error as soon as the status is known.
2147 Frequency : quota enabled and large files being deleted
2149 Description: releasing more than 4GB of quota at once hangs OST
2150 Details : If a user deletes more than 4GB of files on a single OST it
2151 will cause the OST to spin in an infinite loop. Release
2152 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2157 Description: statfs data retrieved from /proc may be stale or zero
2158 Details : When reading per-device statfs data from /proc, in the
2159 {kbytes,files}_{total,free,avail} files, it may appear
2160 as zero or be out of date.
2163 Frequency : systems with MD RAID1 external journal devices
2165 Description: lconf's call to blkid is confused by RAID1 journal devices
2166 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2167 of returning all block devices that match the journal UUID.
2170 Frequency : always, for aggregate stripe size over 4GB
2172 Description: assertion fails when trying to use 4GB stripe size
2173 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2174 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2177 Frequency : always on ppc64
2179 Description: the first write on an ext3 filesystem with mballoc got stuck
2180 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2181 perform endianness conversion.
2183 ------------------------------------------------------------------------------
2185 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2187 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2188 INTEROPERATE with older versions automatically. Please read the
2189 user documentation before upgrading any part of a live system.
2190 * WARNING: Lustre networking configuration changes are required with
2191 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2194 * Support for kernels:
2195 2.6.9-22.0.2.EL (RHEL 4)
2196 2.6.5-7.244 (SLES 9)
2197 2.6.12.6 vanilla (kernel.org)
2200 Severity : enhancement
2201 Bugzilla : 7981/8208
2202 Description: Introduced Lustre Networking (LNET)
2203 Details : LNET is new networking infrastructure for Lustre, it includes
2204 a reorganized network configuration mode (see the user
2205 documentation for full details) as well as support for routing
2206 between different network fabrics. Lustre Networking Devices
2207 (LNDs) for the supported network fabrics have also been
2208 created for this new infrastructure.
2210 Severity : enhancement
2211 Description: Introduced Access control lists
2212 Details : clients can set ACLs on files and directories in order to have
2213 more fine-grained permissions than the standard Unix UGO+RWX.
2214 The MDS must be started with the "-o acl" mount option.
2216 Severity : enhancement
2217 Description: Introduced filesystem quotas
2218 Details : Administrators may now establish per-user quotas on the
2221 Severity : enhancement
2223 Description: Configuration change for the XT3
2224 The PTLLND is now used to run Lustre over Portals on the XT3
2225 The configure option(s) --with-cray-portals are no longer used.
2226 Rather --with-portals=<path-to-portals-includes> is used to
2227 enable building on the XT3. In addition to enable XT3 specific
2228 features the option --enable-cray-xt3 must be used.
2233 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2234 Details : A race between allocating a new llcd and re-getting the llcd_lock
2235 allowed another thread to grab newly-allocated llcd.
2237 Severity : enhancement
2239 Description: 2.6 OST async journal commit and locking fix to improve performance
2240 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2241 2.6 kernels are now async as they already were in 2.4 kernels so
2242 that they can commit concurrently with the network bulk transfer.
2243 For block-allocated files the filter allocation semaphore is held
2244 to avoid filesystem fragmentation during allocation. BKL lock
2245 removed for 2.6 xattr operations where it is no longer needed.
2250 Description: lconf incorrectly determined whether two IP networks could talk
2251 Details : In some more complicated routing and multiple-network
2252 configurations, lconf will avoid trying to make a network
2253 connection to a disjoint part of the IP space. It was doing the
2254 math incorrectly for one set of cases.
2259 Description: Fix for potential infinite loop processing records in an llog.
2260 Details : If an llog record is corrupted/zeroed, it is possible to loop
2261 forever in llog_process(). Validate the llog record length
2262 and skip the remainder of the block on error.
2265 Frequency : occasional (liblustre only)
2267 Description: liblustre could not open files whose last component is a symlink
2268 Details : sysio_path_walk() would incorrectly pass the open intent to
2269 intermediate path components.
2272 Frequency : rare (liblustre only with non-standard tuning)
2273 Bugzilla : 7201 (7350)
2274 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2275 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2276 after it has been released, only do this for referenced locks
2278 Severity : enhancement
2280 Description: specifying an (invalid) directory default stripe_size of -1
2281 would reset the directory default striping
2282 Details : stripe_size -1 was used internally to signal directory stripe
2283 removal, now use "all default" to signal dir stripe removal
2284 as a directory striping of "all default" is not useful
2287 Frequency : common for large clusters running liblustre clients
2289 Description: doing an ls when liblustre clients are running is slow
2290 Details : sending a glimpse AST to a liblustre client waits for every AST
2291 to time out, as liblustre clients will not respond. Since they
2292 cannot cache data we refresh the OST lock LVB from disk instead.
2294 Severity : enhancement
2296 Description: doing an ls at the same time as file IO can be slow
2297 Details : enqueue and other "small" requests can be blocked behind many
2298 large IO requests. Create a new OST IO portal for non-IO
2299 requests so they can be processed faster.
2302 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2304 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2305 created file with mode 000
2306 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2307 allow an ftruncate() as a non-root user to a file with mode 000.
2308 Lustre now respects this flag to disable mode checks when
2309 truncating a file owned by the user
2312 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2315 Description: Revoking locks from clients that went dead or catatonic might take
2317 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2318 cancellation of such locks instant on servers without waiting for
2319 any reply from clients. Clients drops these locks when cancel
2320 notification from server is received without replying.
2323 Frequency : liblustre-only, when liblustre client dies or becomes busy
2325 Description: Doing ls on Linux clients can take a long time with active
2327 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2328 granting such locks to it in the first place if possible. Locks
2329 are taken by proxy on the OST during the read or write and
2330 dropped immediately afterward. Add connect flags handling, do
2331 not grant locks to liblustre clients for glimpse ASTs.
2333 Severity : enhancement
2335 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2336 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2337 sequential reads and behaves badly with applications doing random
2338 reads. Improve it by reading ahead at least read region, and
2339 avoiding excessive large RPC for small reads.
2341 Severity : enhancement
2343 Description: Creating more than 1000 files for a single job may cause a load
2344 imbalance on the OSTs if there are also a large number of OSTs.
2345 Details : qos_prep_create() uses an OST index reseed value that is an
2346 even multiple of the number of available OSTs so that if the
2347 reseed happens in the middle of the object allocation it will
2348 still utilize the OSTs as uniformly as possible.
2353 Description: OST or MDS may oops in ping_evictor_main()
2354 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2355 but doesn't restart at beginning of obd_exports_timed list
2358 Severity : enhancement
2360 Description: improve by-nid export eviction on the MDS and OST
2361 Details : allow multiple exports with the same NID to be evicted at one
2362 time without re-searching the exports list.
2365 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2367 Description: MDS may oops in groups_free()
2368 Details : in rare race conditions a newly allocated group_info struct is
2369 freed again, and this can be NULL. The 2.4 compatibility code
2370 for groups_free() checked for a NULL pointer, but 2.6 did not.
2373 Frequency : common for liblustre clients doing little filesystem IO
2374 Bugzilla : 9352, 7313
2375 Description: server may evict liblustre clients accessing contended locks
2376 Details : if a client is granted a lock or receives a completion AST
2377 with a blocking AST already set it would not reply to the AST
2378 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2381 Frequency : lfs setstripe, only systems with more than 160 OSTs
2383 Description: unable to set striping with a starting offset beyond OST 160
2384 Details : llapi_create_file() incorrectly limited the starting stripe
2385 index to the maximum single-file stripe count.
2388 Frequency : LDAP users only
2390 Description: lconf did not handle in-kernel recovery with LDAP properly
2391 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2393 Severity : enhancement
2395 Description: bind OST threads to NUMA nodes to improve performance
2396 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2397 node and do their allocations there to localize memory access
2399 Severity : enhancement
2401 Description: llmount can determine client NID directly from Myrinet (GM)
2402 Details : the client NID code from gmnalnid was moved directly into
2403 llmount, removing the need to use this or specifying the
2404 client NID explicitly when mounting GM clients with zeroconf
2407 Frequency : if client is started with down MDS
2409 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2410 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2411 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2412 ptlrpc_import_delayed_req() because we didn't check if the
2413 request was interrupted, and we also didn't break out of the
2414 event loop if there was no timeout
2419 Description: data loss during non-page-aligned writes to a single file from
2420 both multiple nodes and multiple threads on one node at same time
2421 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2422 inconsistency led to false short-reads, that were cached and later
2423 used by ->prepare_write() to fill in partially written page,
2424 leading to data loss.
2427 Frequency : always, if lconf --abort_recovery used
2429 Description: lconf --abort_recovery fails with 'Operation not supported'
2430 Details : lconf was attempting to abort recovery on the MDT device and not
2433 Severity : enhancement
2435 Description: remove cleanup logs
2436 Details : replace lconf-generated cleanup logs with lustre internal
2437 cleanup routines. Eliminates the need for client-cleanup and
2440 Severity : enhancement
2442 Description: add support for EAs (user and system) on lustre filesystems
2443 Details : it is now possible to store extended attributes in the Lustre
2444 client filesystem, and with the user_xattr mount option it
2445 is possible to allow users to store EAs on their files also
2447 Severity : enhancement
2449 Description: Add possibility (config option) to show minimal available OST free
2451 Details : When compiled with --enable-mindf configure option, statfs(2)
2452 (and so, df) will return least minimal free space available from
2453 all OSTs as amount of free space on FS, instead of summary of
2454 free spaces of all OSTs.
2456 Severity : enhancement
2458 Description: do not expand extent locks acquired on OST-side
2459 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2460 by server: they are not cached anyway.
2463 Frequency : when mmap is used/binaries executed from Lustre
2465 Description: Unmmap pages before throwing them away from read cache.
2466 Details : llap_shrink cache now attempts to unmap pages before discarding
2467 them (if unmapping failed - do not discard). SLES9 kernel has
2468 extra checks that trigger if this unmapping is not done first.
2473 Description: lconf didn't resolve symlinks before checking to see whether a
2474 given mountpoint was already in use
2477 Frequency : when migrating failover services
2478 Bugzilla : 6395, 9514
2479 Description: When migrating a subset of services from a node (e.g. failback
2480 from a failover service node) the remaining services would
2481 time out and evict clients.
2482 Details : lconf --force (implied by --failover) sets the global obd_timeout
2483 to 5 seconds in order to quickly disconnect, but this caused
2484 other RPCs to time out too quickly. Do not change the global
2485 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2487 Severity : enhancement
2488 Frequency : if MDS is started with down OST
2489 Bugzilla : 9439,5706
2490 Description: Allow startup/shutdown of an MDS without depending on the
2491 availability of the OSTs.
2492 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2493 Add appropriate locking and lov-osc refcounts for safe
2494 cleaning. Add osc abort_inflight calls in case the
2495 synchronize never started.
2498 Frequency : occasional (Cray XT3 only)
2500 Description: root not authorized to access files in CRAY_PORTALS environment
2501 Details : The client process capabilities were not honoured on the MDS in
2502 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2503 been accessed by an authorized user then root was able to access
2504 the file on the local client also. The root user capabilities
2505 are now allowed on the MDS, as this environment has secure UID.
2508 Frequency : occasional
2510 Description: ldiskfs "too long searching" message happens too often
2511 Details : A debugging message (otherwise harmless) prints too often on
2512 the OST console. This has been reduced to only happen when
2513 there are fragmentation problems on the filesystem.
2518 Description: Division by zero in statfs when all OSCs are inactive
2519 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2520 lov_statfs divides by value returned by lov_get_stripecnt().
2524 Bugzilla : 9489, 3273
2525 Description: First write from each client to each OST was only 4kB in size,
2526 to initialize client writeback cache, which caused sub-optimal
2527 RPCs and poor layout on disk for the first writen file.
2528 Details : Clients now request an initial cache grant at (re)connect time
2529 and so that they can start streaming writes to the cache right
2530 away and always do full-sized RPCs if there is enough data.
2531 If the OST is rebooted the client also re-establishes its grant
2532 so that client cached writes will be honoured under the grant.
2537 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2538 Details : Now I/O RPCs go to different portal number and (presumably) fast
2539 lock requests (and glimses) and other RPCs get their own service
2540 threads pool that should be able to service those RPCs
2543 Severity : enhancement
2545 Description: Ability to exchange lustre version between client and servers and
2546 issue warnings at client side if client is too old. Also for
2547 liblustre clients there is ability to refuse connection of too old
2549 Details : New 'version' field is added to connect data structure that is
2550 filled with version info. That info is later checked by server and
2554 Frequency : rare, liblustre only.
2555 Bugzilla : 9296, 9581
2556 Description: Two simultaneous writes from liblustre at offset within same page
2557 might proceed at the same time overwriting eachother with stale
2559 Details : I/O lock withing llu_file_prwv was released too early, before data
2560 actually was hitting the wire. Extended lock-holding time until
2561 server acknowledges receiving data.
2564 Frequency : extremely rare. Never observed in practice.
2566 Description: avoid generating lustre_handle cookie of 0.
2567 Details : class_handle_hash() generates handle cookies by incrementing
2568 global counter, and can hit 0 occasionaly (this is unlikely, but
2569 not impossible, because initial value of cookie counter is
2570 selected randonly). Value of 0 is used as a sentinel meaning
2571 "unassigned handle" --- avoid it. Also coalesce two critical
2572 sections in this function into one.
2574 Severity : enhancement
2576 Description: allow liblustre clients to delegate truncate locking to OST
2577 Details : To avoid overhead of locking, liblustre client instructs OST to
2578 take extent lock in ost_punch() on client's behalf. New connection
2579 flag is added to handle backward compatibility.
2581 Severity : enhancement
2582 Bugzilla : 4928, 7341, 9758
2583 Description: allow number of OST service threads to be specified
2584 Details : a module parameter allows the number of OST service threads
2585 to be specified via "options ost ost_num_threads={N}" in the
2586 OSS's /etc/modules.conf or /etc/modprobe.conf.
2590 Bugzilla : 6146, 9635, 9895
2591 Description: servers crash with bad pointer in target_handle_connect()
2592 Details : In rare cases when a client is reconnecting it was possible that
2593 the connection request was the last reference for that export.
2594 We would temporarily drop the export reference and get a new
2595 one, but this may have been the last reference and the export
2596 was just destroyed. Get new reference before dropping old one.
2598 Severity : enhancement
2599 Frequency : if client is started with failover MDS
2601 Description: Allow multiple MDS hostnames in the mount command
2602 Details : Try to read the configuration from all specified MDS
2603 hostnames during a client mount in case the "primary"
2606 Severity : enhancement
2608 Description: Stop sending data to evicted clients as soon as possible.
2609 Details : Check if the client we are about to send or are sending data to
2610 was evicted already. (Check is done every second of waiting,
2611 for which l_wait_event interface was extended to allow checking
2612 of exit condition at specified intervals).
2615 Frequency : rare, normally only when NFS exporting is done from client
2617 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2619 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2622 Frequency : always, if srand() is called before liblustre initialization
2624 Description: Liblustre uses system PRNG disturbing its usage by user application
2625 Details : Introduce internal to lustre fast and high-quality PRNG for
2626 lustre usage and make liblustre and some other places in generic
2627 lustre code to use it.
2629 Severity : enhancement
2630 Bugzilla : 9477, 9557, 9870
2631 Description: Verify that the MDS configuration logs are updated when xml is
2632 Details : Check if the .xml configuration logs are newer than the config
2633 logs stored on the MDS and report an error if this is the case.
2634 Request --write-conf, or allow starting with --old_conf.
2636 Severity : enhancement
2638 Description: Handle symlinks in the path when checking if Lustre is mounted.
2639 Details : Resolve intermediate symlinks when checking if a client has
2640 mounted a filesystem to avoid duplicate client mounts.
2645 Description: lconf can hit an error exception but still return success.
2646 Details : The lconf command catches the Command error exception at the top
2647 level script context and will exit with the associated exit
2648 status, but doesn't ensure that this exit status is non-zero.
2653 Description: failure of ptlrpc thread startup can cause oops
2654 Details : Starting a ptlrpc service thread can fail if there are a large
2655 number of threads or the server memory is very fragmented.
2656 Handle this without oopsing.
2659 Frequency : always, only if liblustre and non-default acceptor port was used
2661 Description: liblustre cannot connect to servers with non-default acceptor port
2662 Details : tcpnal_set_default_params() was not called and was therefore
2663 ignoring the environment varaible TCPNAL_PORT, as well as other
2664 TCPNAL_ environment variables
2669 Description: two objects could be created on the same OST for a single file
2670 Details : If an OST is down, in some cases it was possible to create two
2671 objects on a single OST for a single file. No problems other
2672 than potential performance impact and spurious error messages.
2676 Bugzilla : 5681, 9562
2677 Description: Client may oops in ll_unhash_aliases
2678 Details : Client dcache may become inconsistent in race condition.
2679 In some cases "getcwd" can fail if the current directory is
2685 Description: Inode refcounting problems in NFS export code
2686 Details : link_raw functions used to call d_instantiate without obtaining
2687 extra inode reference first.
2691 Bugzilla : 9942, 9903
2692 Description: Referencing freed requests leading to crash, memleaks with NFS.
2693 Details : We used to require that call to ll_revalidate_it was always
2694 followed by ll_lookup_it. Also with revalidate_special() it is
2695 possible to call ll_revalidate_it() twice for the same dentry
2696 even if first occurence returned success. This fix changes semantic
2697 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2698 reference on a request referred from the intent.
2699 ll_intent_release() then releases such a request.
2702 Frequency : rare, normally benchmark loads only
2704 Description: unlinked inodes were kept in memory on the client
2705 Details : If a client is repeatedly creating and unlinking files it
2706 can accumulate a lot of stale inodes in the inode slab cache.
2707 If there is no other client load running this can cause the
2708 client node to run out of memory. Instead flush old inodes
2709 from client cache that have the same inode number as a new inode.
2712 Frequency : SLES9 2.6.5 kernel and long filenames only
2713 Bugzilla : 9969, 10379
2714 Description: utime reports stale NFS file handle
2715 Details : SLES9 uses out-of-dentry names in some cases, which confused
2716 the lustre dentry revalidation. Change it to always use the
2720 Frequency : rare, unless heavy write-truncate concurrency is continuous
2721 Bugzilla : 4180, 6984, 7171, 9963, 9331
2722 Description: OST becomes very slow and/or deadlocked during object unlink
2723 Details : filter_destroy() was holding onto the parent directory lock
2724 while truncating+unlinking objects. For very large objects this
2725 may block other threads for a long time and slow overall OST
2726 responsiveness. It may also be possible to get a lock ordering
2727 deadlock in this case, or run out of journal credits because of
2728 the combined truncate+unlink. Solution is to do object truncate
2729 first in one transaction without parent lock, and then do the
2730 final unlink in a new transaction with the parent lock. This
2731 reduces the lock hold time dramatically.
2734 Frequency : rare, 2.4 kernels only
2736 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2737 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2738 before unmounting their filesystem in order to avoid deadlock.
2739 The kernel_locked() function in 2.4 kernels only checks whether
2740 the kernel lock is held, not whether it is this process that is
2741 holding it as 2.6 kernels do.
2746 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2747 Details : The client ptlrpc code may be trying to reconnect to a down
2748 server before a previous connection attempt has timed out.
2749 Increase the reconnect interval to be longer than the connection
2750 timeout interval to avoid sending duplicate connections to
2754 Frequency : echo_client brw_test command
2756 Description: fix echo_client to work with OST preallocated code
2757 Details : OST preallocation code (5137) didn't take echo_client IO path
2758 into account: echo_client calls filter methods outside of any
2759 OST thread and, hence, there is no per-thread preallocated
2760 pages and buffers to use. Solution: hijack pga pages for IO. As
2761 a byproduct, this avoids unnecessary data copying.
2765 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2766 Description: Client can oops in mdc_commit_close() after open replay
2767 Details : It was possible for the MDS to return an open request with no
2768 transaction number in mds_finish_transno() if the client was
2769 evicted, but without actually returning an error. Clients
2770 would later try to replay that open and may trip an assertion
2771 Simplify the client close codepath, and always return an error
2772 from the MDS in case the open is not successful.
2775 Frequency : rare, 2.6 OSTs only
2777 Description: OST may deadlock under high load on fragmented files
2778 Details : If there was a heavy load and highly-fragmented OST filesystems
2779 it was possible to have all the OST threads deadlock waiting on
2780 allocation of biovecs, because the biovecs were not released
2781 until the entire RPC IO was completed. Instead, release biovecs
2782 as soon as they are complete to ensure forward IO progress.
2784 Severity : enhancement
2786 Description: Support for specifying external journal device at mount
2787 Details : If an OST or MDS device is formatted with an external journal
2788 device, this device major/minor is stored in the ext3 superblock
2789 and may not be valid for failover. Allow detecting and
2790 specifying the external journal at mount time.
2795 Description: Mounting an MDS with pending unlinked files may cause oops
2796 Details : target_finish_recovery() calls mds_postrecov() which returned
2797 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2798 considers this an error and immediately begins cleaning up the
2799 lov, just after starting the mds_lov process
2801 Severity : enhancement
2803 Description: Implement 'lfs df' to report actual free space on per-OST basis
2804 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2805 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2806 usage of objects; '-h' to report in human readable format.
2808 ------------------------------------------------------------------------------
2810 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2817 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2818 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2819 at startup during orphan unlinking. Instead, skip buddy update
2820 before setup, as it will be regenerated after recovery is complete.
2825 Description: If an OST is inactive, its locks might reference stale inodes.
2826 Details : lov_change_cbdata() must iterate over all namespaces, even if
2827 they are inactive to clear inode references from the lock.
2829 Severity : enhancement
2830 Frequency : occasional, if non-standard max_dirty_mb used
2832 Description: Client will block write RPCs if not enough grant
2833 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2834 then the client will block writes while waiting for another RPC
2835 to complete instead of consuming its dirty limit. With change
2836 we get improved performance when max_dirty_mb is small.
2838 Severity : enhancement
2839 Bugzilla : 3389, 6253
2840 Description: Add support for supplementary groups on the MDS.
2841 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2842 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2843 lookups for user supplementary groups into a cache.
2847 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2848 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2849 for liblustre O_EXCL is enforced by MDS.
2853 Description: readdir never returns NULL in liblustre.
2854 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2855 dentry in struct dirent.
2859 Frequency : liblustre only, depends on application IO pattern
2860 Description: liblustre clients evicted if not contacting servers
2861 Details : Don't put liblustre clients into the ping_evictor list, so
2862 they will not be evicted by the pinger ever.
2864 Severity : enhancement
2866 Description: Add ability to evict clients by NID from MDS.
2867 Details : By echoing "nid:$NID" string into
2868 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2869 $NID would be instantly evicted from this MDS and from all active
2870 OSTs connected to it.
2874 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2875 Details : lookup_it_finish() used to query file size from OSTs that was not
2880 Description: service threads change working directory to that of init
2881 Details : Starting lustre service threads may pin the working directory
2882 of the parent thread, making that filesystem busy. Threads
2883 now change to the working directory of init to avoid this.
2887 Frequency : during shutdown only
2888 Description: shutdown with a failed MDS or OST can cause unmount to hang
2889 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2894 Frequency : 2.6 only
2895 Description: chmod/chown may include an extra supplementary group
2896 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2897 supplementary group and if none is specified this is used.
2900 Bugzilla : 5479 (6816)
2902 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2903 Details : If lookup is in progress on a file that is unlinked we might try
2904 to revalidate the inode and fail in revalidate after lookup is
2905 complete and ll_file_open() enqueues the open again but
2906 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2909 Frequency : always, if lconf --abort_recovery used
2911 Description: lconf --abort_recovery fails with 'Operation not supported'
2912 Details : lconf was attempting to abort recovery on the MDT device and not
2915 ------------------------------------------------------------------------------
2917 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2922 Frequency : rare (only unsupported configurations with a node running as an
2924 Bugzilla : 6514, 5137
2925 Description: Mounting a Lustre file system on a node running as an OST could
2927 Details : OSTs now preallocates memory needed to write out data at
2928 startup, instead of when needed, to avoid having to
2929 allocate memory in possibly low memory situations.
2930 Specifically, if the file system is mounted on on OST,
2931 memory pressure could force it to try to write out data,
2932 which it needed to allocate memory to do. Due to the low
2933 memory, it would be unable to do so and the node would
2934 become unresponsive.
2936 Severity : enhancement
2938 Description: Addition of lconf --service command line option
2939 Details : lconf now accepts a '--service <arg>' option, which is
2940 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2942 Severity : enhancement
2944 Description: Failover mode is now the default for OSTs.
2945 Details : By default, OSTs will now run in failover mode. To return to
2946 the old behaviour, add '--failout' to the lmc line for OSTs.
2948 Severity : enhancement
2950 Description: Health checks are now provided for MDS and OSTs
2951 Details : Additional detailed health check information on MSD and OSTs
2952 is now provided through the procfs health_check value.
2955 Frequency : occasional, depends on IO load
2957 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2958 after numerous create/delete cycles
2959 Details : The ext3 inode allocation policy would not allocate new inodes
2960 very well on the OSTs because there are no new directories
2961 being created. Instead we look for groups with free space if
2962 the parent directories are nearly full.
2967 Description: Network or server problems during mount may cause partially
2968 mounted clients instead of returning an error.
2969 Details : The config llog parsing code may overwrite the error return
2970 code during mount error handling, returning success instead
2976 Description: MDS can fail to allocate large reply buffers
2977 Details : After long uptimes the MDS can fail to allocate large reply
2978 buffers (e.g. zconf client mount config records) due to memory
2979 fragmentation or consumption by the buffer cache. Preallocate
2980 some large reply buffers so that these replies can be sent even
2981 under memory pressure.
2985 Frequency : rare (liblustre)
2986 Description: fsx running with liblustre complained that using truncate() to
2987 extend the file doesn't work. This patch corrects that issue.
2988 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2989 ATTR_SIZE and lsm use in llu_setattr_raw.
2993 Frequency : rare, only 2.6 kernels
2994 Description: Unusual file access patterns on the MDS may result in inode
2995 data being lost in very rare circumstances.
2996 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2997 code in the 2.6 kernel caused the inode and/or EA data not to
2998 be read from disk, causing single-file corruption.
3002 Frequency : rare, only 2.6 filesystems using extents
3003 Description: Heavy concurrent write and delete load may cause data corruption.
3004 Details : It was possible under high-load situations to have an extent
3005 metadata block in the block device cache from a just-unlinked
3006 file overwrite a newly-allocated data block. We now unmap any
3007 metadata buffers that alias just-allocated data blocks.
3011 Frequency : filesystems with default stripe_count larger than 77
3012 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3013 Details : lconf specifies an inode size of 4096 bytes when the default
3014 stripe_count is larger than 77. This conflicts with the default
3015 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3016 this case to avoid pinning too much memory for large EAs.
3018 ------------------------------------------------------------------------------
3020 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3025 Frequency : rare (extremely heavy IO load with hundreds of clients)
3027 Description: Client is evicted, gets IO error writing to file
3028 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3029 caused clients to be evicted for AST timeouts. The fixes in
3030 bug 5192 mean we no longer need to have such short AST timeouts
3031 so ldlm_timeout has been increased.
3034 Frequency : occasional during --force or --failover shutdown under load
3035 Bugzilla : 5949, 4834
3036 Description: Server oops/LBUG if stopped with --force or --failover under load
3037 Details : a collection of import/export refcount and cleanup ordering
3038 issues fixed for safer force cleanup
3041 Frequency : only filesystems larger than 120 OSTs
3042 Bugzilla : 5990, 6223
3043 Description: lfs getstripe would oops on a very large filesystem
3044 Details : lov_getconfig used kfree on vmalloc'd memory
3047 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3048 Bugzilla : 6242, 6243
3049 Description: reading from files that had been truncated to a non-zero size
3050 but never opened returned no data
3051 Details : ll_file_read() reads zeros from no-object files to EOF
3056 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3057 Details : There's little harm in aborting MDS/OSS recovery and letting it
3058 try again, so I removed the LASSERT and return an error instead.
3060 Severity : enhancement
3062 Description: New debugging infrastructure for tracking down data corruption
3063 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3064 (b) cover more of the client-side code path, and (c) try to narrow
3065 down where problems occurred
3069 Bugzilla : 3819, 4364, 4397, 6313
3070 Description: Racing close and eviction MDS could cause assertion in mds_close
3071 Details : It was possible to get multiple mfd references during close and
3072 client eviction, leading to one thread referencing a freed mfd.
3074 Severity: : enhancement
3075 Bugzilla : 3262, 6359
3076 Description: Attempts to reconnect to servers are now more aggressive.
3077 Details : This builds on the enhanced upcall-less recovery that was added
3078 in 1.4.2. When trying to reconnect to servers, clients will
3079 now try each server in the failover group every 10 seconds. By
3080 default, clients would previously try one server every 25 seconds.
3085 Description: After recovery, certain operations trigger a failed
3086 assertion on a client.
3087 Details : Failing over an mds, using lconf -d --failover, while a
3088 client was doing a readdir() call would cause the client to
3089 LBUG after recovery completed and the readdir() was resent.
3091 Severity : enhancement
3093 Description: Default groups are now added by lconf
3094 Details : You can now run lconf --group <servicename> without having to
3095 manually add groups with lmc.
3098 Frequency : occasional
3100 Description: Nodes with an elan id of 0 trigger a failed assertion
3103 Frequency : always when accessing e.g. tty/console device nodes
3105 Description: tty and some other devices nodes cannot be used on lustre
3106 Details : file's private_data field is used by device data and lustre
3107 values in there got lost. New field was added to struct file to
3108 store fs-specific private data.
3111 Frequency : when exporting Lustre via NFS
3113 Description: NFSD failed occasionally when looking up a path component
3114 Details : NFSD is looking up ".." which was broken in ext3 directories
3115 that had grown large enough to become hashed.
3118 Frequency : Clusters with multiple interfaces not on the same subnet
3120 Description: Nodes will repeatedly try to reconnect to an interface which it
3121 cannot reach and report an error to the log.
3122 Details : Extra peer list entries will be created by lconf with some peers
3123 unreachable. lconf now validates the peer before adding it.
3126 Frequency : Only if a default stripe is set on the filesystem root.
3128 Description: Setting a default stripe on the filesystem root prevented the
3129 filesystem from being remounted.
3130 Details : The client was sending extra request flags in the root getattr
3131 request and did not allocate a reply buffer for the dir EA.
3134 Frequency : occasional, higher if lots of files are accessed by one client
3135 Bugzilla : 6159, 6097
3136 Description: Client trips assertion regarding lsm mismatch/magic
3137 Details : While revalidating inodes the VFS looks up inodes with ifind()
3138 and in rare cases can find an inode that is being freed.
3139 The ll_test_inode() code will free the lsm during ifind()
3140 when it finds an existing inode and then the VFS later attaches
3141 this free lsm to a new inode.
3145 Bugzilla : 6422, 7030
3146 Description: MDS deadlock between mkdir and client eviction
3147 Details : Creating a new file via mkdir or mknod (starting a transaction
3148 and getting the ns lock) can deadlock with client eviction
3149 (gets ns lock and trying to finish a synchronous transaction).
3152 Frequency : occasional
3153 Description: While starting a server, the fsfilt_ext3 module could not be
3155 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3156 kernels. Previously, lconf would still use the ext3 name
3157 when trying to load modules. Now, it will correctly use
3158 ext3 on 2.4 and ldiskfs on 2.6.
3160 Severity : enhancement
3161 Description: The default stripe count has been changed to 1
3162 Details : The interpretation of the default stripe count (0, to lfs
3163 or lmc) has been changed to mean striping across a single
3164 OST, rather than all available. For general usage we have
3165 found a stripe count of 1 or 2 works best.
3167 Severity : enhancement
3168 Description: Add support for compiling against Cray portals.
3169 Details : Conditional compiling for some areas that are different
3173 Frequency : occasional
3174 Bugzilla : 6409, 6834
3175 Description: Creating files with an explicit stripe count may lead to
3176 a failed assertion on the MDS
3177 Details : If some OSTs are full or unavailable, creating files may
3178 trigger a failed assertion on the MDS. Now, Lustre will
3179 try to use other servers or return an error to the
3183 Frequency : occasional
3185 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3186 caused lock timeout during readahead (since 1.4.2).
3187 Details : Processes doing readahead might match a lock that hasn't been
3188 granted yet if there are overlapping and conflicting lock
3189 requests. The readahead process waits on ungranted lock
3190 (original lock is CBPENDING), while OST waits for that process
3191 to cancel CBPENDING read lock and eventually evicts client.
3193 Severity : enhancement
3195 Description: Initial enabling of flock support for clients
3196 Details : Implements fcntl advisory locking and file status functions.
3197 This feature is provided as an optional mount flag (default
3198 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3199 locking are implemented yet, and those that are are not guaranteed
3200 to be completely correct in production environments.
3201 mount -t lustre -o [flock|noflock] ...
3204 Frequency : occasional
3206 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3207 assertion in the ext3 JBD (journaling) layer.
3208 Details : The b_committed_data struct is protected by the big kernel lock
3209 in 2.4 kernels, serializing journal_commit_transaction() and
3210 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3211 access to this struct. In 2.6 kernels there is finer grained
3212 locking to improve SMP performance of the JBD layer.
3216 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3221 Description: merge and cleanup kernel patches.
3224 -----------------------------------------------------------------------------
3226 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3228 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3229 versions of Lustre. Please update all servers and clients to
3230 version 1.4.2 or later at the same time. You must also run
3231 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3232 to update the configuration logs.
3234 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3235 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3236 - swab llog records of type '0' so we get proper header size/idx (5861)
3237 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3238 - fix rename of one directory over another leaking an inode (5953)
3239 - avoid SetPageDirty on 2.6 (5981)
3240 - don't re-add just-being-destroyed locks to the waiting list (5653)
3241 - when creating new directories, inherit the parent's custom
3242 striping settings if present parent (3048)
3243 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3244 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3245 - don't hold client locks on temporary worklist from l_lru (5666)
3246 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3247 - automatically evict dead clients (5921)
3248 - Update file size properly in create+truncate+fstat case (6196)
3249 - Do not unhash mountpoint dentries, do not allow removal of
3251 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3252 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3253 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3254 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3255 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3256 - Correctly handle reads of files with no objects (6243)
3257 - lctl recover will also mark a device active if deactivate used (5933)
3259 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3260 - allow --write-conf on an MDS with different nettype than client (5619)
3261 - don't write config llogs to MDS for mounts not from that MDS (5617)
3262 - lconf should create multiple TCP connections from a client (5201)
3263 - init scripts are now turned off by default; run chkconfig --on
3264 lustre and chkconfig --on lustrefs to use them
3265 - upcalls are no longer needed for clients to recover to failover
3267 - add --abort-recovery option to lconf to abort recovery on device
3269 - add support for an arbitrary number of OSTs (3026)
3270 - Quota support protocol changes.
3271 - forward compatibility changes to wire structs (6007)
3272 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3273 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3274 - improved "lustre status" script
3275 - initialize blocksize for non-regular files (6062)
3276 - added --disable-server and --disable-client configure options (5782)
3277 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3278 - Vanilla 2.4.29 support
3279 - increase maximum number of obd devices to 520 (6242)
3280 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3281 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3282 SLES 9 2.6.5 kernels
3283 - limit stripes per file to 160 (the maximum EA size) (6093)
3285 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3288 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3289 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3290 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3291 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3292 - small ext3 extents cleanups and fixes (5733)
3293 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3294 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3295 - avoid needless client->OST connect, fix handle mismatch (5317)
3296 - fix DLM error path that led to out-of-sync client, long delays (5779)
3297 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3298 - fix several locking issues related to i_size (5492,5624,5654,5672)
3299 - don't move pending lock onto export if it is already evicted (5683)
3300 - fix kernel oops when creating .foo in unlinked directory (5548)
3301 - fix deadlock in obdfilter statistics vs. object create (5811)
3302 - use time_{before,after} to avoid timer jiffies wrap (5882)
3303 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3304 - Do not leak request if server was not able to process it (5154)
3305 - If mds_open unable to find parent dir, make that negative lookup(5154)
3306 - don't create new directories with extent-mapping (5909, 5936)
3308 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3309 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3310 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3311 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3312 - support for RHEL4 kernel on i686 (5773)
3313 - provide error messages when incompatible logs are encountered (5898)
3315 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3316 * version 1.4.0.10 (1.4.1 release candidate 1)
3318 - don't keep a lock reference when lock is not granted (4238)
3319 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3320 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3321 - drop import inflight refcount on signal_completed_replay error (5255)
3322 - unlock page after async write error during send (3677)
3323 - handle missing objects in filter_preprw_read properly (5265)
3324 - no transno return for symlink open, don't save no-trasno open (3440)
3325 - don't try to complete elan receive that already failed (4012)
3326 - free RPC server reply state on error (5406)
3327 - clean up thread from ptlrpc_start_thread() on error (5160)
3328 - readahead could read extra page into cache that wasn't ejected (5388)
3329 - prevent races in class_attach/setup/cleanup/detach (5260)
3330 - don't dereference de->d_inode after l_dput of de (5458)
3331 - use "int" for stripe value returned from lock_to_stripe (5544)
3332 - mballoc allocation and error-checking fixes in 2.6 (5504)
3333 - block device patches to fix I/O request sizes in 2.6 (5482)
3334 - look up hostnames for IB nals (5602)
3335 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3336 - don't start multiple acceptors for the same port (5277)
3337 - fix incorrect LASSERT in mds_getattr_name (5635)
3338 - export a proc file for general "ping" checking (5628)
3339 - fix "lfs check" to not block when the MDS is down (5628)
3341 - service request history (4965)
3342 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3343 - create an "evict_client" /proc entry on OSTs, like the MDS has
3344 - fix mount usage message, return errors per mount(8) (5168)
3345 - change grep [] to grep "[]" in tests so they work in more UMLs
3346 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3347 - remove ancient LOV_MAGIC_V0 EA support (5047)
3348 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3349 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3350 - allow readahead allocations to fail when low on memory (5383)
3351 - mmap locking landed again, after considerable improvement (2828)
3352 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3354 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3357 - send OST transaction number in read/write reply to free req (4966)
3358 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3359 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3360 - the watchdog thread now runs as interruptible (5246)
3361 - flock/lockf fixes (but it's still disabled, pending 5135)
3362 - don't use EXT3 constants in llite code (5094)
3363 - memory shortage at startup could cause assertion (5176)
3365 - reorganization of lov code
3366 - single portals codebase
3368 - add extents/mballoc support (5025)
3369 - direct I/O reads in the obdfilter (4048)
3370 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3372 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3375 - send OST transaction number in read/write reply to free req (4966)
3376 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3377 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3378 - the watchdog thread now runs as interruptible (5246)
3379 - handle missing objects in filter_preprw_read properly (5265)
3380 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3381 - drop import inflight refcount on signal_completed_replay error (5255)
3382 - unlock page after async write error during send (3677)
3383 - return original error code on reconstructed replies (3761)
3384 - no transno return for symlink open, don't save no-trasno open (3440)
3386 - add pid to ldlm debugging output (4922)
3387 - bump the watchdog timeouts -- we can't handle 30sec yet
3388 - extra debugging for orphan dentry/inode bug (5259)
3390 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3393 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3394 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3395 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3396 - fix NULL dereference in /proc/sys/portals/routes (4827)
3397 - allow failed mdc_close() operations to be interrupted (4561)
3398 - stop precreate on OST before MDS would time out on it (4778)
3399 - don't send partial-page writes before EOF from client (4410)
3400 - discard client grant for sub-page writes on large-page clients (4520)
3401 - don't free dentries not owned by NFS code, check generation (4806)
3402 - fix lsm leak if mds_create_objects() fails (4801)
3403 - limit debug_daemon file size, always print CERROR messages (4789)
3404 - use transno after validating reply (3892)
3405 - process timed out requests if import state changes (3754)
3406 - update mtime on OST during writes, return in glimpse (4829)
3407 - add mkfsoptions to LDAP (4679)
3408 - use ->max_readahead method instead of zapping global ra (5039)
3409 - don't interrupt __l_wait_event() during strace
3411 - add software watchdogs to catch hung threads quickly (4941)
3412 - make lustrefs init script start after nfs is mounted
3413 - fix CWARN/ERROR duplication (4930)
3414 - return async write errors to application if possible (2248)
3415 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3416 - print NAL number in %x format (4645)
3417 - update barely-supported suse-2.4.21-171 series (4842)
3418 - support for sles 9 %post scripts
3419 - support for building 2.6 kernel-source packages
3420 - support for sles km_* packages
3422 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3425 - ignore -ENOENT errors in osc_destroy (3639)
3426 - notify osc create thread that OSC is being cleaned up (4600)
3427 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3428 - reconstruct ost_handle() like mds_handle() (4657)
3429 - create a new thread to do import eviction to avoid deadlock (3969)
3430 - let lconf resolve symlinked-to devices (4629)
3431 - don't unlink "objects" from directory with default EA (4554)
3432 - hold socknal file ref over connect in case target is down (4394)
3433 - allow more than 32000 subdirectories in a single directory (3244)
3434 - fix blocks count for O_DIRECT writes (3751)
3435 - OST returns ENOSPC from object create when no space left (4539)
3436 - don't send truncate RPC if file size isn't changing (4410)
3437 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3438 - bind to privileged port in socknal and tcpnal (3689)
3440 - rate limit CERROR/CWARN console message to avoid overload (4519)
3441 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3442 - basic mmap support (3918)
3443 - kernel patch series update from b1_4 (4711)
3445 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3448 - avoid crash during MDS cleanup with OST shut down (2775)
3449 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3450 - don't use bad inodes on the MDS (3744)
3451 - dynamic object preallocation to improve recovery speed (4236)
3452 - don't hold spinlock over lock dumping or change debug flags (4401)
3453 - don't zero obd_dev when it is force cleaned (3651)
3454 - print grants to console if they go negative (4431)
3455 - "lctl deactivate" will stop automatic recovery attempts (3406)
3456 - look for existing locks in ldlm_handle_enqueue() (3764)
3457 - don't resolve lock handle twice in recovery avoiding race (4401)
3458 - revalidate should check working dir is a directory (4134)
3460 - don't always mark "slow" obdfilter messages as errors (4418)
3462 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3465 - don't close LustreDB during write_conf until it is done (3860)
3466 - fix typo in lconf for_each_profile (3821)
3467 - allow dumping logs from multiple threads at one time (3820)
3468 - don't allow multiple threads in OSC recovery (3812)
3469 - fix debug_size parameters (3864)
3470 - fix mds_postrecov to initialize import for llog ctxt (3121)
3471 - replace config semaphore with spinlock (3306)
3472 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3473 - don't allow enqueue to complete on a destroyed export (3822)
3474 - down write_lock before checking llog header bitmap (3825)
3475 - recover from lock replay timeout (3764)
3476 - up llog sem before sending rpc (3652)
3477 - reduce ns lock hold times when setting kms (3267)
3478 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3479 - fix NULL deref and obd_dev leak on setup error (3312)
3480 - replace some LBUG about llog ops with error handling (3841)
3481 - don't match INVALID dentries from d_lookup and spin (3784)
3482 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3483 - fix invalid assertion in ptlrpc_set_wait (3880)
3485 - add libwrap support for the TCP acceptor (3996)
3486 - add /proc/sys/portals/routes for non-root route listing (3994)
3487 - allow setting MDS UUID in .xml (2580)
3488 - print the stack of a process that LBUGs (4228)
3490 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3493 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3494 - make sure to unset replay flag from failed open requests (3440)
3495 - if default stripe count is 0, use OST count for inode size (3636)
3496 - update parent mtime/ctime on client for create/unlink (2611)
3497 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3498 - free recovery state on server during a forced cleanup (3571)
3499 - unregister_reply for resent reqs (3063)
3500 - loop back devices mounting and status check on 2.6 (3563)
3501 - fix resource-creation race that can provoke i_size == 0 (3513)
3502 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3503 - more debugging for page-accounting assertion (3746)
3504 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3505 - avoid deadlock after precreation failure (3758)
3506 - fix race and lock order deadlock in orphan handling (3450, 3750)
3507 - add validity checks when grabbing inodes from l_ast_data (3599)
3509 - add /proc/.../recovery_status to obdfilter (3428)
3510 - lightweight CDEBUG infrastructure, debug daemon (3668)
3511 - change default OSC RPC parameters to be better on small clusters
3512 - turn off OST read cache for files smaller than 32MB
3513 - install man pages and include them in rpms (3100)
3514 - add new init script for (un)mounting lustre filesystems (2593)
3515 - run chkconfig in %post for init scripts (3701)
3516 - drop scimac NAL (unmaintained)
3518 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3521 - clean kiobufs before and after use (3485)
3522 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3523 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3524 - change init script to fail more clearly if not run as root (1528)
3525 - allow clients to reconnect during replay (1742)
3526 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3527 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3528 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3529 - fix cond_resched (3554)
3530 - only evict unfinished clients after recovery (3515)
3531 - allow bulk resend, prevent data loss (3570)
3532 - dynamic ptlrpc request buffer allocation (2102)
3533 - don't allow unlinking open directory if it isn't empty (2904)
3534 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3535 - remove extraneous obd dereference causing LASSERT failure (3334)
3536 - don't use get_cycles() when creating temp. files on the mds (3156)
3537 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3538 - handle EEXIST for set-stripe, set proper directory name (3336)
3540 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3541 - fix ksocknal_fmb_callback() error messages (2918)
3543 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3546 - don't copy lvb into (possibly NULL) reply on error (2983)
3547 - don't deref dentry after dput, don't free lvb on error (2922)
3548 - use the kms to determine writeback rpc length (2947)
3549 - increment oti_logcookies when osc is inactive (2948)
3550 - update client's i_blocks count via lvb messages (2543)
3551 - handle intent open/close of special files properly (1557)
3552 - mount MDS with errors=remount-ro, like obdfilter (2009)
3553 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3554 - don't use cancelling-locks' kms values (2947)
3555 - use highest lock extent for kms, not last one (2925)
3556 - don't dereference ERR_PTR() dentry in error handling path (3107)
3557 - fix thread race in portals_debug_dumplog() (3122)
3558 - create lprocfs device entries at setup instead of at attach (1519)
3559 - common AST error handler, don't evict client on completion race (3145)
3560 - zero nameidata in detach_mnt in 2.6 (3118)
3561 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3562 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3563 - initialize RPC timeout timer earlier for 2.6 (3219)
3564 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3565 - print nal/nid for unknown nid (3258)
3566 - additional checks for oscc recovery before doing precreate (3284)
3567 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3568 - don't crash in mdc_close for bad permissions on open (3285)
3569 - zero i_rdev for non-device files (3147)
3570 - clear page->private before handing to FS, better assertion (3119)
3571 - tune the read pipeline (3236)
3572 - fix incorrect decref of invalidated dentry (2350)
3573 - provide read-ahead stats and refine rpc in flight stats (3328)
3574 - don't hold journal transaction open across create RPC (3313)
3575 - update atime on MDS at close time (3265)
3576 - close LDAP connection when recovering to avoid server load (3315)
3577 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3578 - don't leak open file on MDS after open resend (3325)
3579 - serialize filter_precreate and filter_destroy_precreated (3329)
3580 - loop device shouldn't call sync_dev() for nul device (3092)
3581 - clear page cache after eviction (2766)
3582 - resynchronize MDS->OST in background (2824)
3583 - refuse to mount the same filesystem twice on same mountpoint (3394)
3584 - allow llmount to create routes for mounting behind routers (3320)
3585 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3586 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3587 - fix rare problem with rename on htree directories (3417)
3589 - allow default OST striping configuration per directory (1414)
3590 - fix compilation for qswnal for 2.6 kernels (3125)
3591 - increase maximum number of MDS request buffers for large systems
3592 - change liblustreapi to be useful for external progs like lfsck (3098)
3593 - increase local configuration timeout for slow disks (3353)
3594 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3596 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3599 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3600 - don't overwrite extent policy data in reply if lock was blocked (2901)
3601 - drop filter export grants atomically with removal from device (2663)
3602 - del obd_self_export from work_list in class_disconnect_exports (2908)
3603 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3604 - swab reply message in mdc_close, other PPC fixes (2464)
3605 - fix destroying of named logs (2325)
3606 - overwrite old logs when running lconf --write_conf (2264)
3607 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3608 - fix race in target_handle_connect (2898)
3609 - mds_reint_create() should take same inode create lock (2926)
3610 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3611 - don't close files for self_export to avoid uninitialized obd (2936)
3612 - allow MDS with the same name as client node (2939)
3613 - hold dentry reference for closed log files for unlink (2325)
3614 - reserve space for all logs during transactions (2059)
3615 - don't evict page beyond end of stripe extent (2925)
3616 - don't oops on a deleted current working directory (2399)
3617 - handle hard links to targets without a parent properly (2517)
3618 - don't dereference NULL lock when racing during eviction (2867)
3619 - don't grow lock extents when lots of conflicting locks (2919)
3621 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3624 - account for cache space usage on clients to avoid data loss (974)
3625 - lfsck support in lustre kernel code (2349)
3626 - reduce journal credits needed for BRW writes (2370)
3627 - orphan handling to avoid losing space on client/server crashes
3628 - ptlrpcd can be blocked, stopping ALL progress (2477)
3629 - use lock value blocks to assist in proper KMS, faster stat (1021)
3630 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3631 - recovery for initial connections (2355)
3632 - fixes for mds_cleanup_orphans (1934)
3633 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3634 - block all file creations until orphan recovery completes (1901)
3635 - client remove rq_connection from request struct (2423)
3636 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3637 - recovery timer race (2670)
3638 - mdc_close recovey bug (2532)
3639 - ptlrpc cleanup bug (2710)
3640 - mds timeout on local locks (2588)
3641 - namespace lock held during RPCs (2431)
3642 - handle interrupted sync write properly (2503)
3643 - don't try to handle a message that hasn't been replied to (2699)
3644 - client assert failure during cleanup after abort recovery (2701)
3645 - leak mdc device after failed mount (2712)
3646 - ptlrpc_check_set allows timedout requests to complete (2714)
3647 - wait for inflight reqs when ptlrpcd finishes (2710)
3648 - make sure unregistered services are removed from the srv_list
3649 - reset bulk XID's when resending them (caught by 1138 test)
3650 - unregister_bulk after timeout
3651 - fix lconf error (2694)
3652 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3653 - readahead locks pages, leaves pending causing memory pressure (2673)
3654 - increase OST request buffers to 4096 on large machines (2729)
3655 - fix up permission of existing directories in simple_mkdir (2661)
3656 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3657 - don't assign transno to errored transactions (2742)
3658 - don't delete objects on OST if given a bogus objid from MDS (2751)
3659 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3660 - if rq_no_resend, then timeout request after recovery (2432)
3661 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3662 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3663 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3664 - MDS recovery shouldn't delete objects that it has given out (2730)
3665 - if enqueue arrives after completion, don't clobber LVB (2819)
3666 - don't unlock pages twice when trigger_group_io returns error (2814)
3667 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3668 - don't write pages to disk if there was an error (1450)
3669 - don't ping imports that have recovery disabled (2676)
3670 - take buffered bytes into account when balancing socknal conn (2817)
3671 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3672 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3673 - remove little-endian swabbing of llog records (1987)
3674 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3675 - retry reposting request buffers if they fail (1191)
3676 - grow extent at grant time to avoid granting a revoked lock (2809)
3677 - lock revoke doesn't evict page if covered by a second lock (2765)
3678 - disable VM readahead to avoid reading outside lock extents (2805)
3680 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3681 - updated kernel patches for hp-2.4.20 kernel (2681)
3683 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3686 - fix truncated write corruption (2366)
3687 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3689 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3690 - ensure len(uuid) < 37 in lmc (1171)
3691 - fix ia64 OOPS in llog_test (2255)
3692 - zero end of page at obdfilter for partial page writes (2648)
3693 - don't leave stale dentries around after renames (bug 2428)
3694 - fix timeouts when evicting a client with a single lock held (2642)
3695 - set deadline for the initial HELLO message to drain (2634)
3696 - print out dotted-quad IP addresses in the socknal (2302)
3698 - additional debugging for MDS client eviction problem (2443)
3699 - fix mkfsoptions support for osts (2603, 2604)
3701 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3704 - add series for the vanilla 2.6.0 kernel
3705 - add series for the vanilla 2.4.24 kernel
3706 - add series for a cray x86/64 UL kernel drop
3707 - fix xattr patches for the vanilla 2.4.19 series
3709 - generate true UUIDs in lmc (1171)
3710 - have portals stack dumping break in UML (2466)
3711 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3712 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3713 - wake callback waiting threads on client eviction (2460)
3714 - Add --ptldebug and --subsystem to lmc (1719)
3715 - update assertion to allow safe interrupt allocation
3716 - set rq_no_resend for cancel requests (2432)
3717 - recalculate ptlrpcd timeout after resend (2494)
3718 - call vfs_rmdir when removing pending directories (2368)
3719 - fix renaming a file to itself (2429)
3720 - lmc creates a default one-stripe lov (2454)
3721 - expand procfs space to handle large clusters (2326)
3722 - increase UML stack to avoid overflow
3723 - update lconf's list of debug and subsystem masks
3724 - fix lfs find --obd (2510)
3725 - /proc tunable for disabling filter read caching (2591)
3726 - stop rpm packages from altering slapd.conf (2301)
3727 - disable nagle in the socknal under 0conf (2578)
3728 - choose mds inode size based on stripe count (2572)
3729 - fix kernel-source rpm problems (2516)
3731 - add --disable-doc to avoid pdf generation (2421)
3732 - update documentation, tests, type-os, comments
3733 - avoid format warnings on ia64
3734 - remove the TOE NAL
3735 - tiny code cleanups by removing unused fields
3737 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3740 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3741 - remove the most blatant lies from BUILDING file (2371)
3742 - change default debug level to reasonable production setting
3743 - reduce client side cache size to reduce cache flush time
3744 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3745 - make TCP zerocopy and pinger support enabled by default (2476)
3746 - sync writes completed after process exits caused crashes (2319)
3747 - maintain correct mount count on the MDS (2356)
3748 - backout 1557, because 2316 wasn't really fixed
3749 - better file I/O statistics gathering in /proc
3750 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3751 - another kernel patch to fix zero-copy TCP function export
3752 - don't take duplicate lock when processing re-sent getattr (2420)
3753 - lctl uses obd_self_export instead of creating new conn (2353)
3754 - MDS/OST recovery case which requires object creation asserted (2425)
3755 - move lfs from /usr/sbin to /usr/bin in packages
3756 - fix race between mds_client_add and mds_client_free (2417)
3757 - use kmalloc instead of slabs in portals (2430)
3758 - don't create duplicate records when a failover MDS is present (2442)
3759 - remove unnecessary mount age check (2332)
3760 - don't remove directory inodes from locks prematurely (2451)
3761 - don't break if MDS service name is the same as hostname (2103)
3762 - fix races in client write RPC generation when cache full (2482)
3764 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3767 - remove now-unused request->rq_obd (278)
3768 - if an allocation fails, print out how much memory we've used (1933)
3769 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3770 - add the "configurable stack size" patch to most series files (1256)
3771 - ability to write large log records, for 100+ OST configs (2306)
3772 - fix NULL deref when filter_prep fails (2314)
3773 - fix operator precedence error in filter_sync
3774 - dynamic allocation of socknal TX descriptors (2315)
3775 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3776 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3777 - ensure that CWARN messages in Portals always get to the syslog
3778 - __init/__exit are not for prototype decls (ldlm_init/exit)
3779 - x86-64 compile warning fixes
3780 - fix gateway LMC keyword conflict (2318)
3781 - fix MDS lock inversions in getattr/reint paths (1844)
3782 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3783 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3784 - DLM race condition prevented some lock evictions (2328)
3785 - ENOMEM detection and retry on socknal sends (2230)
3786 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3787 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3788 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3789 - better error messages when a client is rejected during recovery (1505)
3790 - avoid cancelling locks which were never granted, after failure (2330)
3791 - fix i_sem/journal inversion in mds_client_add (2333)
3792 - fix truncate/getattr lock cycle deadlock (2334)
3793 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3794 - fix two rare exit paths which could leak an l_lock() ref (2321)
3795 - fencepost error in MDS/OST orphan recovery (2226)
3796 - make log record alignment 8 bytes (1988)
3797 - lstripe now fails when requested offset > ost_count (2237)
3798 - ensure that all kernel series have a complete list.h (1607)
3799 - fix crashes in special-file operations (2316)
3800 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3802 - allow configurable automake binary, for testing new versions
3803 - small update to the lfs documentation
3805 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3807 * fix negative export reference count in fsfilt_sync (2312)
3809 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3810 * release candidate 0.9.1
3812 - orphans are moved into the PENDING directory for possible recovery
3813 - replayed opens now open by fid for orphan/rename safety (1042)
3814 - last close of an orphan inode generates a transno (683)
3815 - chdir() and mount() now pin the directory entry (1020)
3816 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3817 - discard very old requests without processing them (1502)
3818 - remove some common, well-understood CERRORs (1505)
3819 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3820 - clear "grant" flags in OST replies until OST grant code lands (1644)
3821 - fix read performance by not clobbering i_blksize on client (1598)
3822 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3823 - make LRU size a /proc tunable, clears locks when reduced (707)
3824 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3825 - server locks take references on exports now (1558)
3826 - build fixes for 2.4.20-rh trees (1663)
3827 - return an error from lov_create if all OSCs are inactive (1751)
3828 - fix import levels when a reconnect happens without a timeout (1597)
3829 - exit early from mds_open if we get a lookup error (1749)
3830 - partial page read at EOF wouldn't wait for disk before sending (1642)
3831 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3832 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3833 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3834 - write/truncate lock inversion (1639)
3835 - Don't auto-load obdclass, portals modules during cleanup (1495)
3836 - fix timestamps from jumping to "now" (1763)
3837 - extra journal assertions (1648)
3838 - add an extra multiunlink test (1771)
3839 - fix read_record/write_record API (1776)
3840 - fix leak of offset_extent, possible incorrect i_size later (1772)
3841 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3842 - Debugging for the kqswnal_get_idle_tx problems (1820)
3843 - Allow recovery to be attempted multiple times (1536)
3844 - Write out MDS last_rcvd file after it is first created (1600)
3845 - Fix tx_descriptor leak in failed transmit situations (1827)
3846 - ext3 journaling fixes for assertion failure after IO error (1871)
3847 - class_export_put() on freed export after completion AST error (1896)
3848 - Fix revalidate looping in VFS (1322)
3849 - Don't access a freed export during MDS_REINT timeout (1521)
3850 - Add open-unlink recovery support on the MDS (1673,1764)
3851 - Return an error if no MDS data was read from last_rcvd (1946)
3852 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3853 - Don't setup a disk device that doesn't match exported UUID (317)
3854 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3855 - avoid committing NULL handle in force close
3856 - local.sh is now a one-stripe LOV configuration
3857 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3858 - fix ext3 htree duplicate directory entry corruption (1516)
3859 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3860 - update to new LOV EA format (2097)
3861 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3863 2003-06-15 Phil Schwan <phil@clusterfs.com>
3866 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3867 - per-import/export recovery handling (958, 931, 959)
3868 - multiple last-rcvd slots, for serving multiple FSes (949)
3869 - connections are again shared between multiple imp/exports (963, 964)
3870 - "umount -f" would hang if any requests needed to be sent (393, 978)
3871 - avoid pinning large req buffer by copying for queued messages (989)
3872 - add "uuid" to "lctl device" command to help upcalls (991)
3873 - "open" RPCs with transnos would confuse recovery counters (1037)
3874 - do proper endian conversion of all wire messages (288, 340, 891)
3875 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3876 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3877 - fix ptlrpc_connection leak in target_handle_connect (1174)
3878 - fix import refcounting bug in OST and MDS cleanup (1134)
3879 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3880 - fix very unlikely obd_types race condition (501)
3881 - remove osc_open hack for echo_client (1187)
3882 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3883 - a failure in read_inode2 leads to deadlock (1139)
3884 - cancel ack-locks as soon as transaction is committed (1072)
3885 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3886 - make sure to commitrw after any preprw to avoid deadlock (1162)
3887 - failing to execute a file in a lustre FS would lock inode (1203)
3888 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3889 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3890 - obd_uuid2tgt would walk off the end of the list (1255)
3891 - on IA64 the osc would give portals incorrect bulk size (1258)
3892 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3893 - fix lock inversion caused by new llite matching code (1282)
3894 - limit the number of dirty pages on a client to 10MB (1286)
3895 - timed out locks were not being corrected cancelled (1289)
3896 - fix O_DIRECT above 4GB on IA-32 (1292)
3897 * major user-visible changes
3898 - fail out/fail over policy now controlled by the upcall (993)
3900 - add OBD_PING to check server availability and failure (954)
3901 - lustre messages are now sent in sending host order (288, 340, 891)
3902 - add eadatalen to MDS getattr reply (340)
3903 - OST read replies may contain second buffer, with per-page status (593)
3905 2003-03-11 Phil Schwan <phil@clusterfs.com>
3908 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3909 - failed open()s could cause deadlock; fixed (867, 869)
3910 - stop cancelling OST locks when files are closed (481)
3911 - overlapping XID spaces caused network corruption (851, 853)
3912 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3913 - setattr_raw added, to do single-RPC, server-side setattrs
3914 - lmc/lconf syntax change for OST UUIDs
3915 - fix crashy race condition between ptlrpc_free_req and osc_close
3916 - don't use request in mdc_enqueue if we hit a timeout (889)
3917 - don't set the inode i_size for regular files from the MDS (896)
3918 - handle out of order completion AST (842)
3919 - don't LBUG if a lock request times out after receiving AST (913)
3920 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3921 - if a bad lock AST arrives, send an error instead of dropping entirely
3922 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3923 - fix leak in bulk IO when only partially completed (899, 900, 926)
3924 - fix O_DIRECT for ia64 (55)
3925 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3926 - C-z after timeout could hang a process forever; fixed (977)
3928 - client-side I/O cache (678, 924, 929, 941, 970)
3930 - READPAGE and SETATTRs which don't take server-side locks get
3933 2003-02-11 Phil Schwan <phil@clusterfs.com>
3936 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3937 - Fix fsfilt_extN_readpage() to read a full page of directory
3938 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3939 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3940 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3941 - Most intent operations were removed, in favour of a new RPC mode
3942 that does a single RPC to the server and bypasses most of the VFS
3943 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3944 - Aggressively cancel local locks on DLM servers
3945 - mds_reint_unlink sends EA to the client if it's the last nlink.
3946 client uses that EA to unlink OST objects.
3947 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3948 - recursive symlinks were fixed (439)
3949 - fixed NULL deref in DEBUG_REQ
3950 - filter_update_lastobjid no longer calls sync, which annoyed extN
3951 - fixed multi-client small-writes to a single file problem (445)
3952 - fixed mtime updates during file writes (607)
3953 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3954 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3955 - fixed Linux OST size reporting problem (444, 656)
3956 - OST now updates object mtime with writes or setattr (607, 619)
3957 - client verifies file size before zeroing page past EOF (445)
3958 - OST now writes last allocated objid to disk with allocation (108)
3959 - LOV on echo now works (409)
3961 - mds_reint_unlink sends a new buffer, with the EA included. this
3962 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3963 set if a regular file was being unlinked, and it was the last link
3964 - use PtlGet from the target for bulk writes (315)
3965 - OST now updates object mtime with writes or setattr (607, 619)
3966 - LDLM now has a grant-time callback to revalidate locked items, if
3968 - Many MDS operations were reorganized to combat race conditions
3970 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3971 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3973 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3976 - Fully reactivate OST imports after reconnection (512, others)
3977 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3978 - More graceful error handling for truncating on dead OST (515)
3979 - Don't error out unless we're actually accessing dead stripes (474)
3980 - Fix garbage sizes when stripes are missing (410)
3981 - LRU counters were broken, causing constant lock purge (433, 432)
3982 - garbage on read from stripes with failed OSTs (441)
3983 - mark OSCs as active before reconnecting during recovery (438)
3984 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3985 - lfind did not preserve OST order in output (443)
3986 - symlinks cause hung clients, incorrect data (439)
3987 - stop dereferencing request after dropping refcount (457)
3988 - don't LASSERT(spin_is_locked) on non-SMP (455)
3989 - fixes for many rename() bugs
3990 - fstat didn't correctly synchronize attributes (399)
3991 - server must handle lock cancellation during blocking AST prep (487)
3992 - bulk descriptors were free()d too soon (511)
3993 - fix paths in lconf, which would load incorrect modules (451, 507)
3994 - fix confusing lconf 'host not found' error message (386)
3995 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3996 - fix race condition in mdc_blocking_ast() for inode access (526)
3997 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3998 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3999 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4000 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4001 - fix lstripe in several ways (kernel side) (527)
4002 - fix request leak in ldlm_cli_enqueue (262)
4003 - incorrect OSC was marked inactive after OST failure
4004 - call mds_fs_cleanup before unmounting filesystem (524)
4005 - fix races between taking ns_lock and ldlm_lock_change_resource
4006 - fix races updating LOV export open file list
4007 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4008 - fix recovery NULL deref in ldlm_cli_cancel_unused
4009 - fix some DLM races by using new hash table for lock handles (419)
4010 - permit the client to specify desired inodes, at replay
4011 - duplicate requests when we queue them for replay reintegration
4012 - fix last_rcvd offset calculation
4013 - sync after each recovered transaction, so we always make progress
4014 - never, not always, ERESTART requests without transnos
4015 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4017 - skip replay if the MDS says that the client is already connected
4018 - don't check for a recovery-enabled export to match lctl's UUID
4019 - don't INC_USE_COUNT for phantom exports
4020 - don't crash when cleaning up phantom exports (567)
4021 - don't double-finish or set replay data for errored mdc_open requests
4022 - abort requests when they time out, so we don't get old replies
4023 - send/receive replies for AST messages again
4024 - if the client says that it doesn't have the lock, cancel it on the
4026 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4027 mark it as destroyed, it will all work out when decref is called
4028 - fix module use counts (22, 581)
4030 - ASTs now expect a reply (server cancels lock on error reply)
4032 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4035 - fix many simultaneous client startup (392)
4036 - fix dentry->d_it clobbering
4037 - credentials weren't being shipped for readdir/getattr operations
4038 - remove invalid assertions triggered during some concurrent MD
4040 - proper Lustre versions added (336, 389)
4041 - fix memory leak for create error case (398)
4042 - fix LOV locking bug that would get cli/srv out of sync
4043 - fix echo client over LOV (409)
4044 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4045 - fix double-O_EXCL intent crash (424)
4046 - avoid sending multiple lock CANCELs (352)
4048 - MDS can do multi-client recovery (modulo bugs in new code)
4050 - many updates, edits, cleanups
4052 2002-11-18 Phil Schwan <phil@clusterfs.com>
4055 - fix null d_it dereference (346)
4056 - fix full OST/dbench hang (333)
4057 - fix permission problem with file removal (286)
4058 - fix removal of OSCs from LOV when they fail
4059 - fix NULL deref during bulk timeout (214)
4060 - fix problems related to multiple filesystems on one MDS (241)
4061 - fixed serious subtle metadata locking bugs
4062 - free locks on clients when inodes are removed due to memory
4064 - fix inode pointer in lock data (285)
4065 - partial support for multiple MDS on a single host (241)
4066 - data locks weren't cancelled at clear_inode time (290, 311)
4067 - intent locks could lead to unbounded lock growth (205)
4068 - added a maximum lock count, an LRU list, and a flusher
4069 - fix multiple rename (365)
4070 - properly abstracted the echo client
4071 - OSC locked 1 byte too many; fixed
4072 - rewrote brw callback code:
4073 - fixed recovery bugs related to LOVs (306)
4074 - fixed too-many-pages-in-one-write crash (191)
4075 - fixed (again) crash in sync_io_timeout (214)
4076 - probably fixed callback-related race (385)
4078 - Add capability to MDS protocol
4079 - LDLM cancellations and callbacks on different portals
4081 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4084 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4085 - defer open object destruction to close time (601981, 138)
4086 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4087 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4088 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4089 - don't call obd_setattr in ll_file_release for destroyed objects
4090 * protocol change to lustre_msg: move |version| and add |flags|
4091 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4092 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4093 * added replay of create, unlink, link and rename operations during
4094 MDS failover; recovery should be much more robust now
4095 * remove failed OSCs from LOVs (only lov_create uses this so far)
4096 * the lustre-HOWTO was brought (more) up to date (582544)
4098 2002-10-23 Phil Schwan <phil@clusterfs.com>
4101 - in-use dentries weren't being reused properly (617851)
4102 - prevent multiple LDLM setup (599178)
4103 - fix LOV size calculations for truncate (617853)
4104 - fix client handling of MDS intent errors (POSIX)
4105 - fix permission bug in lovstripe.c test (624321)
4106 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4107 - truncate past end of file could corrupt data
4108 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4109 - a race in recovery could return ETIMEDOUT to apps (623947)
4110 - building outside the source directory was fixed
4111 * the lustre-HOWTO was brought (more) up to date (582544)
4112 * major progress was made on recovery functionality
4114 2002-10-10 Phil Schwan <phil@clusterfs.com>
4117 - recovery deadlock fix
4118 - rm -rf causes LBUG fix (617817)
4119 - file open by multiple tasks fix (618962)
4120 - directory permissions bugs (602707 and 620007)
4121 - journal_stop fixed with locking (611313)
4122 - O_APPEND failures resolved (618273, perhaps 614459)
4123 - lconf PATH fix (619770)
4124 - IA64 build fix (621450)
4125 - RPC buffer sizes scale with amount of memory
4127 2002-10-01 Phil Schwan <phil@clusterfs.com>
4130 - locks would be cancelled without throwing away data pages,
4131 resulting in inconsistent data (605627)
4132 - inode attributes were not always being refreshed (605627, 612449)
4133 - lconf now continues to cleanup after lctl reports an error
4134 - MDS now enforces user permissions (602707)
4135 - lprocfs cleanup fixed, but not yet enabled (614157)
4136 - fixed infinite server hang, should a client not respond to an AST
4137 - avoid going into recovery if user calls readlink() with a buffer
4138 that's too small (613941)
4139 - AST RPCs no longer require replies (614867) -- this may be changed
4140 - don't crash server if client sends an IOV that's too big (611336)
4141 - fixed lock conversion deadlock (611892)
4142 - fixed the following of symlinks (614622)
4143 * recovery: the server can remove locks from a client that dies, other
4144 clients can make progress
4145 * more extN patch fixes
4146 * compile-time configurable ptlrpc buffer allocations
4148 - collaborative read cache document
4149 - Lustre Lite Performance CDR document-in-progress
4151 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4154 - fix typo in patch-2.4.18
4156 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4159 - clear ptlrpc request each time in handle_incoming_request()
4160 - unlink of files now destroys the object on the OST
4162 2002-09-19 Peter Braam <braam@clusterfs.com>
4164 * add hard link support
4165 * change obdfile creation method
4166 * kernel patch changed
4168 2002-09-19 Peter Braam <braam@clusterfs.com>
4171 - stack overflow bug in extN fixed
4173 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4175 * documentation updates
4176 - add man pages for config tools
4177 - update tests/README to describe testing with new config tools
4178 - finish metadata API descriptions
4179 * bug fixes and cleanups
4180 - statfs workaround for 16TB limit
4181 - LOV stripe allocation improved, can stripe on subset of OSTs
4182 - LOV file size/IO offset was wrong for files > 4GB in size
4183 - object EA data was being dropped, caused files to be unreadable
4184 - memory overflow with non-LOV OST caused memory corruption
4185 - fixed regression tests to work with new config tools, obdfilter
4186 - fixed bug when directory size became larger than 1 block
4187 - fixed bug (for single client case) when PWD was deleted
4188 - invalidate local directory pages when doing intent-based ops
4189 - avoid LDLM oops when lock callback contained bad data
4191 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4193 * documentation updates
4194 * bug fixes and cleanups
4195 - configuration tools
4198 - 64-bit compile warnings
4199 - 64-bit internal statfs data
4201 * test_brw on persistent OST devices
4203 * lprocfs (disabled)
4205 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4207 * documentation updates
4208 * bug fixes and cleanups
4209 * configuration tools
4211 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4214 * many small fixes to 0_5_4
4215 * io/network handling
4216 * thinkos in MDS operations
4218 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4221 * crucial basic fixes to 0.5.3
4222 * IOR, Iozone work over Elan
4225 2002-08-07 Phil Schwan <phil@clusterfs.com>
4226 * version 0_5_3, our first alpha
4227 * we use the new Portals iovs
4228 * documentation updates
4229 * bug fixes and cleanups
4230 * small changes in the DLM wire protocol
4232 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4233 * version 0_5_1 with some initial stability,
4234 * locking on MD and file I/O.
4235 * documentation updates
4236 * several bug fixes since 0.5.0
4237 * small changes in wire protocol
4239 2002-07-18 Phil Schwan <phil@clusterfs.com>
4241 * delivered as Lustre Light Alpha
4242 * fixed a crash after handling invalid MDS requests
4243 * fixed directory pages for architectures with non-4k pages sizes
4245 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4246 * release version v0_4_4
4247 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4249 * Fixes a number of interruption problems with OST operations.
4250 * Update documentation for portals header changes
4251 * Move all wire protocol structs/defines to lustre_idl.h
4252 * Fixes symlink length bug.
4253 * Add tcpdump to repository.
4255 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4256 * release version v0_4_3
4257 * Fixes statfs for inodes on extN.
4258 * Fixes bug in runtests which would delete /etc/hosts.
4259 * Use 64-bit object IDs wherever possible (not into VFS though)
4260 Remove ost_get_info, which is unused by lustre, and out of date.
4262 2002-07-03 Peter Braam <braam@clusterfs.com>
4263 * release version v0_4_2 Fixes a lookup error (type not passed)
4264 * move forward to head of Portals
4265 * move forward to latest Lustre kernel
4267 2002-06-25 Peter Braam <braam@clusterfs.com>
4268 * release version v0_4_1. Hopefully stable on single node use.